Prepare v0.15.0 release documentation

.gitignore

@@ -22,6 +22,9 @@
 go.work
 go.work.sum
 
+# Local ReleaseKit executable
+.releasekit/
+
 # env file
 .env
 
@@ -67,4 +70,3 @@ Icon
 .AppleDesktop
 Network Trash Folder
 Temporary Items
-.apdisk

AGENTS.md

@@ -1,2 +1,6 @@
-Please review `docs/development.md` for initial orientation in this repository
-and follow its task-specific reading guide.
+Start with `docs/development.md` for orientation and use its policy reading guide
+to select the policies applicable to the task. Do not read the entire suite by
+default.
+
+Read and follow `docs/policy/agents.md` before recommending, planning, or
+implementing repository changes.

COPYRIGHT

@@ -0,0 +1,23 @@
+PromptKit
+
+Copyright (C) 2026 Eric Rakestraw
+
+SPDX-License-Identifier: LGPL-3.0-or-later
+
+This library is free software: you can redistribute it and/or modify it
+under the terms of the GNU Lesser General Public License as published by
+the Free Software Foundation, either version 3 of the License, or
+(at your option) any later version.
+
+The GNU Lesser General Public License version 3 is provided in LICENSE.
+
+This library is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU Lesser General Public License for more details.
+
+Policy documents copied or adapted from Standard Repository Policies remain licensed
+under CC-BY-SA-4.0. See docs/oss/attribution.md for the affected material,
+upstream credit, and license links.
+
+Third-party code and assets are subject to their respective license terms.

README.md

@@ -8,7 +8,8 @@ gitea.maximumdirect.net/eric/promptkit
 ```
 
 The root `promptkit` package provides the supported public engine. Consumers
-can configure filesystem or in-memory prompt, profile, and schema sources,
+can configure file, embedded, or programmatic backend, profile, and prompt
+definitions alongside filesystem or in-memory schemas,
 prepare requests without generation, run requests with the built-in
 OpenAI-compatible client, or inject their own model client and artifact reader.
 
@@ -20,9 +21,12 @@ Run the maintained offline preparation example from the repository root:
 go run ./examples/go-library/prepare
 ```
 
-It loads a repository-local prompt, supplies an in-memory profile and inline
-input, and prints deterministic preparation metadata without contacting a
-provider or requiring credentials. Read the
+It loads a repository-local prompt, supplies a programmatic profile and inline
+input, inspects the captured configuration, and prints deterministic
+preparation metadata without contacting a provider or requiring credentials.
+The [consumer guide](docs/consumers/pkg-promptkit.md) also shows a local backend,
+ordered file/embedded sources, programmatic mappings, and how to replace an
+engine when source files change. Read the
 [example source](examples/go-library/prepare/main.go), the
 [Go package consumer guide](docs/consumers/pkg-promptkit.md), and the
 [framework format reference](docs/formats.md) to build a consumer workflow.
@@ -33,6 +37,47 @@ boundary and constraints that framework work must preserve.
 
 ## Release Guidance
 
+The [v0.15.0 release notes](docs/releases/v0.15.0.md) cover error-only
+validation of concrete requests with `Engine.ValidateRequest`.
+
+Consumers upgrading to `v0.12.0` should read the
+[v0.12.0 release notes](docs/releases/v0.12.0.md) for the intentional ConfigKit
+configuration and public construction API migration.
+
+The [v0.11.2 release notes](docs/releases/v0.11.2.md) cover policy alignment
+and copied-policy attribution updates.
+
+The [v0.11.1 release notes](docs/releases/v0.11.1.md) cover the MIT-licensed
+catalog updates and licensing documentation improvements.
+
+Consumers upgrading to `v0.11.0` should read the
+[v0.11.0 changelog and migration guide](docs/releases/v0.11.0.md), particularly
+the default retry behavior and license change.
+
+Consumers upgrading to the `v0.10.0` profile-policy transition
+should read the [v0.10.0 transition guide](docs/releases/v0.10.0.md).
+
+Consumers upgrading from `v0.8.0` to `v0.9.0` should read the
+[v0.9.0 changelog and migration guide](docs/releases/v0.9.0.md).
+
+Consumers upgrading from `v0.7.0` to `v0.8.0` can consult the
+[v0.8.0 changelog and migration guide](docs/releases/v0.8.0.md).
+
+Earlier adopters can consult the
+[v0.7.0 changelog and migration guide](docs/releases/v0.7.0.md).
+
+Consumers upgrading from `v0.5.0` to `v0.6.0` can consult the
+[v0.6.0 changelog and migration guide](docs/releases/v0.6.0.md).
+
+Consumers upgrading from `v0.4.0` to `v0.5.0` can consult the
+[v0.5.0 changelog and migration guide](docs/releases/v0.5.0.md).
+
+Consumers upgrading from `v0.3.0` to `v0.4.0` should read the
+[v0.4.0 changelog and adoption guide](docs/releases/v0.4.0.md).
+
+Consumers upgrading from `v0.2.0` to `v0.3.0` should read the
+[v0.3.0 changelog](docs/releases/v0.3.0.md).
+
 Consumers moving from `v0.1.0` to `v0.2.0` should read the
 [v0.2.0 changelog and migration guide](docs/releases/v0.2.0.md).
 
@@ -41,4 +86,8 @@ Consumers moving from `v0.1.0` to `v0.2.0` should read the
 [Scriptorium](https://gitea.maximumdirect.net/eric/scriptorium) is the CLI and
 HTTP application built on Promptkit.
 
-Promptkit is licensed under the [GNU General Public License version 3](LICENSE).
+## License
+
+Promptkit is licensed under the [LGPL v3.0 or later](COPYRIGHT).
+
+[Open Source attribution notices](docs/oss/attribution.md) are available.

appended_messages_contract_test.go

@@ -0,0 +1,129 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"sync/atomic"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestAppendedMessageRoleConstantsAreStrings(t *testing.T) {
+	var (
+		developer string = promptkit.RoleDeveloper
+		system    string = promptkit.RoleSystem
+		user      string = promptkit.RoleUser
+		assistant string = promptkit.RoleAssistant
+	)
+	if developer != "developer" || system != "system" || user != "user" || assistant != "assistant" {
+		t.Fatalf("unexpected role constants: %q %q %q %q", developer, system, user, assistant)
+	}
+}
+
+func TestAppendedMessagesComposeFrozenEffectivePrompt(t *testing.T) {
+	cacheControl := &promptkit.CacheControl{Type: promptkit.CacheControlEphemeral, TTL: "1h"}
+	appended := []promptkit.RenderedMessage{
+		{Role: " \tAsSiStAnT\n", Content: "previous response"},
+		{Role: promptkit.RoleUser, Content: "corrective request", CacheControl: cacheControl},
+	}
+	client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "output"}}
+	engine := newSimpleEngine(t, "configured prompt", promptkit.WithLLMClient(client))
+	request := promptkit.RunRequest{PromptID: "prompt", AppendedMessages: appended}
+
+	plain, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+	if err != nil {
+		t.Fatalf("prepare plain request: %v", err)
+	}
+	prepared, err := engine.Prepare(context.Background(), request)
+	if err != nil {
+		t.Fatalf("prepare appended request: %v", err)
+	}
+	expected := []promptkit.RenderedMessage{
+		{Role: promptkit.RoleUser, Content: "configured prompt"},
+		{Role: promptkit.RoleAssistant, Content: "previous response"},
+		{Role: promptkit.RoleUser, Content: "corrective request", CacheControl: &promptkit.CacheControl{Type: promptkit.CacheControlEphemeral, TTL: "1h"}},
+	}
+	if !reflect.DeepEqual(prepared.Messages, expected) || prepared.PromptHash != plain.PromptHash || prepared.RenderedPromptHash == plain.RenderedPromptHash {
+		t.Fatalf("prepared effective prompt = %#v, plain = %#v", prepared, plain)
+	}
+
+	equivalent, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt", AppendedMessages: []promptkit.RenderedMessage{
+		{Role: promptkit.RoleAssistant, Content: "previous response"},
+		{Role: promptkit.RoleUser, Content: "corrective request", CacheControl: &promptkit.CacheControl{Type: promptkit.CacheControlEphemeral, TTL: "1h"}},
+	}})
+	if err != nil || equivalent.RenderedPromptHash != prepared.RenderedPromptHash {
+		t.Fatalf("normalized-equivalent request = (%#v, %v), want matching rendered hash %q", equivalent, err, prepared.RenderedPromptHash)
+	}
+
+	for _, variant := range []promptkit.RunRequest{
+		{PromptID: "prompt", AppendedMessages: []promptkit.RenderedMessage{{Role: promptkit.RoleAssistant, Content: "changed"}, expected[2]}},
+		{PromptID: "prompt", AppendedMessages: []promptkit.RenderedMessage{expected[2], expected[1]}},
+		{PromptID: "prompt", AppendedMessages: []promptkit.RenderedMessage{{Role: promptkit.RoleAssistant, Content: "previous response"}, {Role: promptkit.RoleUser, Content: "corrective request"}}},
+	} {
+		variantPrepared, prepareErr := engine.Prepare(context.Background(), variant)
+		if prepareErr != nil || variantPrepared.RenderedPromptHash == prepared.RenderedPromptHash || variantPrepared.PromptHash != prepared.PromptHash {
+			t.Fatalf("variant preparation = (%#v, %v)", variantPrepared, prepareErr)
+		}
+	}
+
+	result, err := engine.Run(context.Background(), request)
+	if err != nil || result.RenderedPromptHash != prepared.RenderedPromptHash || !reflect.DeepEqual(client.requests[0].Prompt.Messages, expected) {
+		t.Fatalf("run result = (%#v, %v), request = %#v", result, err, client.requests)
+	}
+
+	execution, err := engine.PrepareExecution(context.Background(), request)
+	if err != nil {
+		t.Fatalf("prepare execution: %v", err)
+	}
+	appended[0].Content = "changed caller content"
+	cacheControl.TTL = ""
+	firstDetails := execution.Details()
+	firstDetails.Messages[1].Content = "changed details content"
+	firstDetails.Messages[2].CacheControl.Type = "changed details cache"
+	secondDetails := execution.Details()
+	if !reflect.DeepEqual(secondDetails.Messages, expected) || secondDetails.RenderedPromptHash != prepared.RenderedPromptHash {
+		t.Fatalf("prepared execution details = %#v, want frozen %#v", secondDetails, expected)
+	}
+
+	result, err = engine.RunPrepared(context.Background(), execution)
+	if err != nil || result.RenderedPromptHash != prepared.RenderedPromptHash || !reflect.DeepEqual(client.requests[1].Prompt.Messages, expected) {
+		t.Fatalf("prepared result = (%#v, %v), requests = %#v", result, err, client.requests)
+	}
+}
+
+func TestInvalidAppendedMessagesFailBeforeModelWork(t *testing.T) {
+	var modelCalls atomic.Int64
+	engine, err := promptkit.NewEngine(promptkit.Config{}, promptkit.WithLLMClient(countingLLMClient{calls: &modelCalls}))
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	request := promptkit.RunRequest{
+		PromptID: "unreached",
+		AppendedMessages: []promptkit.RenderedMessage{{
+			Role:    "unsupported-role",
+			Content: "sensitive appended content",
+		}},
+	}
+
+	operations := []struct {
+		name string
+		run  func() error
+	}{
+		{name: "Prepare", run: func() error { _, err := engine.Prepare(context.Background(), request); return err }},
+		{name: "PrepareExecution", run: func() error { _, err := engine.PrepareExecution(context.Background(), request); return err }},
+		{name: "Run", run: func() error { _, err := engine.Run(context.Background(), request); return err }},
+	}
+	for _, operation := range operations {
+		t.Run(operation.name, func(t *testing.T) {
+			err := operation.run()
+			if !errors.Is(err, promptkit.ErrInvalidRequest) {
+				t.Fatalf("error = %v, want ErrInvalidRequest", err)
+			}
+		})
+	}
+	if modelCalls.Load() != 0 {
+		t.Fatalf("invalid appended message invoked the model %d times", modelCalls.Load())
+	}
+}

assembled_workflow_benchmark_test.go

@@ -0,0 +1,237 @@
+package promptkit_test
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+var (
+	assembledPreparedSink *promptkit.PreparedRun
+	assembledResultSink   *promptkit.RunResult
+	assembledDetailsSink  promptkit.PreparedRun
+	assembledHandleSink   *promptkit.PreparedExecution
+)
+
+type assembledWorkflowClient struct{ content string }
+
+func (client assembledWorkflowClient) Generate(
+	context.Context,
+	promptkit.GenerateRequest,
+) (*promptkit.GenerateResponse, error) {
+	return &promptkit.GenerateResponse{Content: client.content}, nil
+}
+
+type assembledWorkflowFixture struct {
+	engine          *promptkit.Engine
+	request         promptkit.RunRequest
+	expectedMessage string
+	response        string
+}
+
+func assembledWorkflowSchema(b *testing.B) []byte {
+	b.Helper()
+	properties := make(map[string]any, 256)
+	for index := range 256 {
+		properties[fmt.Sprintf("property-%03d", index)] = map[string]any{"type": "string"}
+	}
+	data, err := json.Marshal(map[string]any{
+		"$schema":              "https://json-schema.org/draft/2020-12/schema",
+		"type":                 "object",
+		"properties":           properties,
+		"additionalProperties": false,
+	})
+	if err != nil {
+		b.Fatal(err)
+	}
+	return data
+}
+
+func assembledWorkflowExtras(entries int) map[string]any {
+	providers := make(map[string]any, entries)
+	for index := range entries {
+		providers[fmt.Sprintf("provider-%03d", index)] = map[string]any{
+			"enabled": true,
+			"labels":  []any{fmt.Sprintf("tier-%d", index%4), fmt.Sprintf("region-%d", index%3)},
+			"limits":  map[string]any{"burst": index + 1, "period": "minute"},
+			"routes":  []any{map[string]any{"weight": index + 1, "fallback": true}},
+		}
+	}
+	return map[string]any{"provider_options": providers}
+}
+
+func newAssembledWorkflowFixture(b *testing.B, name string) assembledWorkflowFixture {
+	b.Helper()
+	message := "Hello {{.name}}"
+	response := "ok"
+	output := map[string]any{"format": "text", "validation_mode": "none"}
+	profile := map[string]any{"endpoint": "http://example.test/v1", "model": "benchmark-model"}
+	options := []promptkit.Option{promptkit.WithLLMClient(assembledWorkflowClient{content: response})}
+	var expectedMessage string
+
+	switch name {
+	case "small":
+		expectedMessage = "Hello Ada"
+	case "large_message":
+		prefix := strings.Repeat("x", 256*1024)
+		message = prefix + "{{.name}}"
+		expectedMessage = prefix + "Ada"
+	case "mutable_payload":
+		message = `{"subject":"{{.name}}"}`
+		expectedMessage = `{"subject":"Ada"}`
+		response = "{}"
+		profile["extra_params"] = assembledWorkflowExtras(64)
+		output = map[string]any{
+			"format": "json", "validation_mode": "json_schema", "schema_path": "workflow-schema.json",
+		}
+		options[0] = promptkit.WithLLMClient(assembledWorkflowClient{content: response})
+		options = append(options, promptkit.WithSchemaFS(fstest.MapFS{
+			"workflow-schema.json": &fstest.MapFile{Data: assembledWorkflowSchema(b)},
+		}, "."))
+	default:
+		b.Fatalf("unknown assembled workflow fixture %q", name)
+	}
+
+	config := map[string]any{
+		"profiles": map[string]any{"profile": profile},
+		"prompts": map[string]any{"workflow": map[string]any{"v1": map[string]any{
+			"default_profile": "profile",
+			"messages":        []any{map[string]any{"role": "user", "content": message}},
+			"output":          output,
+		}}},
+	}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
+		Source: configkit.Mapping("assembled workflow benchmark", config),
+	}}}, options...)
+	if err != nil {
+		b.Fatal(err)
+	}
+	request := promptkit.RunRequest{
+		PromptID: "workflow", PromptVersion: "v1", Vars: map[string]string{"name": "Ada"},
+	}
+	return assembledWorkflowFixture{
+		engine: engine, request: request, expectedMessage: expectedMessage, response: response,
+	}
+}
+
+func verifyAssembledWorkflowFixture(b *testing.B, fixture assembledWorkflowFixture) {
+	b.Helper()
+	ctx := context.Background()
+	prepared, err := fixture.engine.Prepare(ctx, fixture.request)
+	if err != nil {
+		b.Fatal(err)
+	}
+	if prepared.PromptID != "workflow" || prepared.PromptVersion != "v1" || len(prepared.Messages) != 1 ||
+		prepared.Messages[0].Content != fixture.expectedMessage {
+		b.Fatalf("prepared fixture output is invalid: %+v", prepared)
+	}
+	handle, err := fixture.engine.PrepareExecution(ctx, fixture.request)
+	if err != nil {
+		b.Fatal(err)
+	}
+	details := handle.Details()
+	if details.PromptID != "workflow" || len(details.Messages) != 1 || details.Messages[0].Content != fixture.expectedMessage {
+		handle.Discard()
+		b.Fatalf("prepared execution details are invalid: %+v", details)
+	}
+	handle.Discard()
+	result, err := fixture.engine.Run(ctx, fixture.request)
+	if err != nil {
+		b.Fatal(err)
+	}
+	if result.RawOutput != fixture.response {
+		b.Fatalf("run output=%q, want %q", result.RawOutput, fixture.response)
+	}
+	if fixture.response == "{}" && result.Validation.Status != promptkit.ValidationPassed {
+		b.Fatalf("structured response validation=%+v, want passed", result.Validation)
+	}
+}
+
+func BenchmarkEnginePreparation(b *testing.B) {
+	for _, name := range []string{"small", "large_message", "mutable_payload"} {
+		b.Run(name, func(b *testing.B) {
+			fixture := newAssembledWorkflowFixture(b, name)
+			verifyAssembledWorkflowFixture(b, fixture)
+			ctx := context.Background()
+			b.Run("prepare", func(b *testing.B) {
+				b.ReportAllocs()
+				b.ResetTimer()
+				for b.Loop() {
+					prepared, err := fixture.engine.Prepare(ctx, fixture.request)
+					if err != nil {
+						b.Fatal(err)
+					}
+					assembledPreparedSink = prepared
+				}
+				b.StopTimer()
+				assembledPreparedSink = nil
+			})
+			b.Run("prepared_execution_creation_and_discard", func(b *testing.B) {
+				b.ReportAllocs()
+				b.ResetTimer()
+				for b.Loop() {
+					handle, err := fixture.engine.PrepareExecution(ctx, fixture.request)
+					if err != nil {
+						b.Fatal(err)
+					}
+					handle.Discard()
+					assembledHandleSink = handle
+				}
+				b.StopTimer()
+				assembledHandleSink = nil
+			})
+		})
+	}
+}
+
+func BenchmarkEngineRun(b *testing.B) {
+	for _, name := range []string{"small", "large_message", "mutable_payload"} {
+		b.Run(name, func(b *testing.B) {
+			fixture := newAssembledWorkflowFixture(b, name)
+			verifyAssembledWorkflowFixture(b, fixture)
+			ctx := context.Background()
+			b.ReportAllocs()
+			b.ResetTimer()
+			for b.Loop() {
+				result, err := fixture.engine.Run(ctx, fixture.request)
+				if err != nil {
+					b.Fatal(err)
+				}
+				assembledResultSink = result
+			}
+			b.StopTimer()
+			assembledResultSink = nil
+		})
+	}
+}
+
+func BenchmarkPreparedExecutionDetails(b *testing.B) {
+	for _, name := range []string{"small", "large_message", "mutable_payload"} {
+		b.Run(name, func(b *testing.B) {
+			fixture := newAssembledWorkflowFixture(b, name)
+			verifyAssembledWorkflowFixture(b, fixture)
+			handle, err := fixture.engine.PrepareExecution(context.Background(), fixture.request)
+			if err != nil {
+				b.Fatal(err)
+			}
+			defer handle.Discard()
+			initial := handle.Details()
+			if len(initial.Messages) != 1 || initial.Messages[0].Content != fixture.expectedMessage {
+				b.Fatalf("prepared details fixture is invalid: %+v", initial)
+			}
+			b.ReportAllocs()
+			b.ResetTimer()
+			for b.Loop() {
+				assembledDetailsSink = handle.Details()
+			}
+			b.StopTimer()
+			assembledDetailsSink = promptkit.PreparedRun{}
+		})
+	}
+}

backends.go

@@ -1,74 +1,69 @@
 package promptkit
 
 import (
+	"gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/promptkit/internal/backend"
-	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 )
 
-// BackendOpenRouter is the reserved ID of Promptkit's built-in OpenRouter
-// backend.
+// BackendOpenRouter identifies the maintained OpenRouter backend.
 const BackendOpenRouter = backend.OpenRouterID
 
-// Backend configures one engine-scoped OpenAI-compatible backend.
-//
-// Backend has no stable JSON representation. Use keyed literals so additions
-// to this configuration value do not break source compatibility.
-type Backend struct {
-	// ID is the stable, case-sensitive registry key. NewEngine trims it and
-	// requires a non-blank value. BackendOpenRouter is reserved.
-	ID string
-	// Endpoint is the OpenAI-compatible base endpoint. NewEngine trims it and
-	// requires an absolute HTTP or HTTPS URL with a host and without user
-	// information, a query string, or a fragment. Paths are allowed.
-	Endpoint string
-	// APIKeyEnv optionally names the environment variable containing the API
-	// key. NewEngine trims it and requires the portable form
-	// [A-Za-z_][A-Za-z0-9_]*. Store only the name, never a credential value.
-	APIKeyEnv string
-	// ExtraParams contains backend-wide request defaults. Values must be
-	// JSON-compatible, finite, acyclic, and keyed by non-empty strings. Keys
-	// must not be model, session_id, messages, temperature, max_tokens, top_p,
-	// service_tier, reasoning_effort, or response_format. An empty map supplies
-	// no defaults. NewEngine deeply copies the map.
-	ExtraParams map[string]any
-	// ConcurrencyLimit is the maximum number of simultaneous model-generation
-	// calls allowed for this backend within one Engine. Zero leaves the backend
-	// unlimited. A negative value makes NewEngine fail with ErrInvalidConfig.
-	ConcurrencyLimit int
-	// QueueCapacity controls how many additional Run calls may be admitted
-	// beyond ConcurrencyLimit. Nil uses 1024 when ConcurrencyLimit is positive;
-	// a pointer uses its exact value, including zero. The pointed-to value must
-	// be non-negative, and QueueCapacity must be nil when ConcurrencyLimit is
-	// zero. Their sum must fit in an int. WithBackend copies the value and does
-	// not retain the pointer.
+// BackendRakestrawHome identifies the maintained Rakestrawhome backend.
+const BackendRakestrawHome = backend.RakestrawHomeID
+
+// BackendLocal is a conventional ID for a caller-configured local backend.
+const BackendLocal = "local"
+
+// BackendConfig is a presence-aware convenience value for a keyed backend
+// definition. Nil pointer fields are omitted, while non-nil pointers preserve
+// explicit zero or blank values. A nil ExtraParams map is omitted; a non-nil
+// empty map supplies an empty mapping. Use raw ConfigKit Mapping inputs for
+// explicit null. BackendConfig has no stable JSON representation.
+type BackendConfig struct {
+	// Endpoint is the required absolute HTTP(S) base URL after all layers merge.
+	Endpoint *string
+	// APIKeyEnv optionally names an environment variable, not a credential.
+	APIKeyEnv *string
+	// ConcurrencyLimit is zero for unlimited generation; negative is invalid.
+	ConcurrencyLimit *int
+	// QueueCapacity defaults to 1024 for positive concurrency when omitted.
+	// An explicit value, including zero, requires positive concurrency.
 	QueueCapacity *int
+	// ExtraParams supplies JSON-compatible provider request defaults.
+	ExtraParams map[string]any
 }
 
-// WithBackend adds one Backend registration to the constructed Engine.
-//
-// Registrations accumulate in option order. Every normalized ID must be unique
-// across consumer registrations and built-ins; a duplicate or invalid
-// definition makes NewEngine fail with ErrInvalidConfig. In particular,
-// BackendOpenRouter cannot be replaced. The immutable registration is scoped
-// to the resulting Engine and cannot be enumerated, replaced, removed, or
-// mutated after construction. WithBackend does not install package-global
-// state.
-func WithBackend(backend Backend) Option {
-	queueCapacity := 0
-	queueCapacitySet := backend.QueueCapacity != nil
-	if queueCapacitySet {
-		queueCapacity = *backend.QueueCapacity
+// BackendSource creates a labeled ConfigKit mapping from keyed backend values.
+// It encodes fields without validating them; NewEngine validates the final
+// effective configuration after all selected source layers are combined.
+// Mapping captures a copy of supported input values at creation; unsupported
+// values fail when the source is loaded.
+func BackendSource(label string, backends map[string]BackendConfig) configkit.Source {
+	definitions := make(map[string]any, len(backends))
+	for id, config := range backends {
+		fields := make(map[string]any)
+		if config.Endpoint != nil {
+			fields["endpoint"] = *config.Endpoint
+		}
+		if config.APIKeyEnv != nil {
+			fields["api_key_env"] = *config.APIKeyEnv
+		}
+		if config.ConcurrencyLimit != nil {
+			fields["concurrency_limit"] = *config.ConcurrencyLimit
+		}
+		if config.QueueCapacity != nil {
+			fields["queue_capacity"] = *config.QueueCapacity
+		}
+		if config.ExtraParams != nil {
+			fields["extra_params"] = providerMapForMapping(config.ExtraParams)
+		}
+		definitions[id] = fields
 	}
-	return optionFunc(func(options *engineOptions) error {
-		options.backends = append(options.backends, domain.Backend{
-			ID:               backend.ID,
-			Endpoint:         backend.Endpoint,
-			APIKeyEnv:        backend.APIKeyEnv,
-			ExtraParams:      backend.ExtraParams,
-			ConcurrencyLimit: backend.ConcurrencyLimit,
-			QueueCapacity:    queueCapacity,
-			QueueCapacitySet: queueCapacitySet,
-		})
-		return nil
-	})
+	return configkit.Mapping(label, map[string]any{"backends": definitions})
+}
+
+// LocalBackend supplies endpoint and concurrency pointers for a caller-chosen
+// backend ID; it does not select an ID or validate either value.
+func LocalBackend(endpoint string, concurrencyLimit int) BackendConfig {
+	return BackendConfig{Endpoint: &endpoint, ConcurrencyLimit: &concurrencyLimit}
 }

capacity_contract_test.go

@@ -7,6 +7,7 @@ import (
 	"testing"
 	"time"
 
+	"gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/promptkit"
 )
 
@@ -60,7 +61,18 @@ func TestEngineRejectsRunBeforeCompletionWhenAdmissionIsFull(t *testing.T) {
 
 	awaitCapacitySignal(t, reader.entered, "first artifact read")
 
-	result, err := engine.Run(context.Background(), capacityInputRequest("http://second.example/v1"))
+	canceledContext, cancel := context.WithCancel(context.Background())
+	cancel()
+	result, err := engine.Run(canceledContext, capacityInputRequest("http://canceled.example/v1"))
+	if result != nil || !errors.Is(err, context.Canceled) {
+		t.Fatalf("canceled capacity admission=(%+v, %v), want context cancellation", result, err)
+	}
+	var canceledCapacityErr *promptkit.CapacityError
+	if errors.Is(err, promptkit.ErrCapacityExceeded) || errors.As(err, &canceledCapacityErr) {
+		t.Fatalf("canceled admission exposed capacity rejection: %v", err)
+	}
+
+	result, err = engine.Run(context.Background(), capacityInputRequest("http://second.example/v1"))
 	if result != nil {
 		t.Fatalf("capacity rejection returned partial result: %+v", result)
 	}
@@ -70,6 +82,21 @@ func TestEngineRejectsRunBeforeCompletionWhenAdmissionIsFull(t *testing.T) {
 	if errors.Is(err, promptkit.ErrInvalidRequest) || errors.Is(err, promptkit.ErrLLMGenerate) {
 		t.Fatalf("capacity rejection had an unrelated category: %v", err)
 	}
+	var capacityErr *promptkit.CapacityError
+	if !errors.As(err, &capacityErr) || capacityErr == nil {
+		t.Fatalf("capacity rejection=%v, want CapacityError", err)
+	}
+	if capacityErr.BackendID != "limited" {
+		t.Fatalf("capacity backend ID=%q, want limited", capacityErr.BackendID)
+	}
+	capacityErr.BackendID = "changed"
+
+	result, err = engine.Run(context.Background(), capacityInputRequest("http://third.example/v1"))
+	var subsequentCapacityErr *promptkit.CapacityError
+	if result != nil || !errors.As(err, &subsequentCapacityErr) ||
+		subsequentCapacityErr == nil || subsequentCapacityErr.BackendID != "limited" {
+		t.Fatalf("subsequent capacity rejection=(%+v, %v), want independent limited CapacityError", result, err)
+	}
 	if calls := reader.callCount(); calls != 1 {
 		t.Fatalf("artifact calls=%d, want only the admitted run", calls)
 	}
@@ -124,17 +151,7 @@ func TestUnlimitedBackendsRetainInjectedClientConcurrency(t *testing.T) {
 		{
 			name: "endpoint-only profile",
 			configure: func(t *testing.T, client promptkit.LLMClient) *promptkit.Engine {
-				engine, err := promptkit.NewEngine(promptkit.Config{},
-					promptkit.WithPromptFS(contractPromptFS("prompt", "profile", "message"), "."),
-					promptkit.WithProfiles(promptkit.Profile{
-						ID: "profile", Endpoint: "http://endpoint.example/v1", Model: "model",
-					}),
-					promptkit.WithLLMClient(client),
-				)
-				if err != nil {
-					t.Fatalf("construct endpoint-only engine: %v", err)
-				}
-				return engine
+				return newSimpleEngine(t, "message", promptkit.WithLLMClient(client))
 			},
 		},
 	}
@@ -174,6 +191,19 @@ func TestCapacityExceededSentinelContract(t *testing.T) {
 	if promptkit.ErrCapacityExceeded == nil {
 		t.Fatal("ErrCapacityExceeded is nil")
 	}
+	var nilCapacityErr *promptkit.CapacityError
+	zeroCapacityErr := &promptkit.CapacityError{}
+	populatedCapacityErr := &promptkit.CapacityError{BackendID: "limited"}
+	for _, capacityErr := range []error{nilCapacityErr, zeroCapacityErr} {
+		if !errors.Is(capacityErr, promptkit.ErrCapacityExceeded) {
+			t.Fatalf("capacity error=%v, want ErrCapacityExceeded", capacityErr)
+		}
+	}
+	var discoveredCapacityErr *promptkit.CapacityError
+	if !errors.As(populatedCapacityErr, &discoveredCapacityErr) || discoveredCapacityErr != populatedCapacityErr {
+		t.Fatalf("populated capacity error is not discoverable: %v", populatedCapacityErr)
+	}
+
 	for _, unrelated := range []error{
 		promptkit.ErrInvalidConfig,
 		promptkit.ErrInvalidRequest,
@@ -181,7 +211,8 @@ func TestCapacityExceededSentinelContract(t *testing.T) {
 		promptkit.ErrValidation,
 	} {
 		if errors.Is(promptkit.ErrCapacityExceeded, unrelated) ||
-			errors.Is(unrelated, promptkit.ErrCapacityExceeded) {
+			errors.Is(unrelated, promptkit.ErrCapacityExceeded) ||
+			errors.Is(populatedCapacityErr, unrelated) {
 			t.Fatalf("ErrCapacityExceeded aliases unrelated sentinel %v", unrelated)
 		}
 	}
@@ -210,27 +241,35 @@ func newBackendCapacityEngine(
 	reader promptkit.ArtifactReader,
 ) *promptkit.Engine {
 	t.Helper()
-	promptFS := contractPromptFS("prompt", "profile", "message")
+	promptID, content := "prompt", "message"
+	var inputs []any
 	if reader != nil {
-		promptFS = contractInputPromptFS()
+		promptID, content = "input-prompt", `{{input "input"}}`
+		inputs = []any{map[string]any{"name": "input", "required": true}}
 	}
-	options := []promptkit.Option{
-		promptkit.WithPromptFS(promptFS, "."),
-		promptkit.WithBackend(promptkit.Backend{
-			ID:               "limited",
-			Endpoint:         "http://backend.example/v1",
-			ConcurrencyLimit: limit,
-			QueueCapacity:    queueCapacity,
-		}),
-		promptkit.WithProfiles(promptkit.Profile{
-			ID: "profile", BackendID: "limited", Model: "model",
-		}),
-		promptkit.WithLLMClient(client),
+	backend := map[string]any{"endpoint": "http://backend.example/v1", "concurrency_limit": limit}
+	if queueCapacity != nil {
+		backend["queue_capacity"] = *queueCapacity
 	}
+	definition := map[string]any{
+		"default_profile": "profile",
+		"messages":        []any{map[string]any{"role": "user", "content": content}},
+		"output":          map[string]any{"format": "text", "validation_mode": "none"},
+	}
+	if inputs != nil {
+		definition["inputs"] = inputs
+	}
+	options := []promptkit.Option{promptkit.WithLLMClient(client)}
 	if reader != nil {
 		options = append(options, promptkit.WithArtifactReader(reader))
 	}
-	engine, err := promptkit.NewEngine(promptkit.Config{}, options...)
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
+		Source: configkit.Mapping("capacity fixture", map[string]any{
+			"backends": map[string]any{"limited": backend},
+			"profiles": map[string]any{"profile": map[string]any{"backend": "limited", "model": "model"}},
+			"prompts":  map[string]any{promptID: map[string]any{"1": definition}},
+		}),
+	}}}, options...)
 	if err != nil {
 		t.Fatalf("construct capacity engine: %v", err)
 	}

capacity_error.go

@@ -0,0 +1,40 @@
+package promptkit
+
+import (
+	"fmt"
+	"strings"
+)
+
+// CapacityError reports bounded admission rejected for a selected backend.
+//
+// Engine-produced values identify only rejection at Promptkit's bounded
+// [Engine.Run] or [Engine.RunPrepared] admission boundary. BackendID is the
+// normalized registered backend ID used for routing and capacity; endpoint
+// overrides do not change it. Every engine-produced value is nonnil and has a
+// nonblank BackendID. Provider errors, active-generation waiting, and caller
+// cancellation are not represented by this type.
+//
+// Callers own returned values and may mutate BackendID without affecting engine
+// state or another error. CapacityError and its default Go encoding have no
+// stable JSON contract. Consumer-constructed values do not establish that an
+// engine rejected work.
+type CapacityError struct {
+	// BackendID is the normalized registered backend ID whose admission was
+	// rejected.
+	BackendID string
+}
+
+// Error returns diagnostic wording that is not a parsing contract. It is safe
+// to call on a nil receiver or a value with a blank BackendID.
+func (e *CapacityError) Error() string {
+	if e == nil || strings.TrimSpace(e.BackendID) == "" {
+		return ErrCapacityExceeded.Error()
+	}
+	return fmt.Sprintf("backend %q admission: %v", e.BackendID, ErrCapacityExceeded)
+}
+
+// Unwrap returns ErrCapacityExceeded so errors.Is and errors.As can be used
+// together. It is safe to call on a nil receiver or a zero value.
+func (e *CapacityError) Unwrap() error {
+	return ErrCapacityExceeded
+}

configuration_diff_contract_test.go

@@ -0,0 +1,243 @@
+package promptkit_test
+
+import (
+	"errors"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestConfigurationInspectionDiffsCapturedValues(t *testing.T) {
+	left := comparisonInspection(t, "earlier-model")
+	right := comparisonInspection(t, "later-model")
+
+	differences, err := left.Diff(right)
+	if err != nil {
+		t.Fatalf("compare captured configurations: %v", err)
+	}
+	if len(differences) != 1 {
+		t.Fatalf("difference count = %d, want 1: %+v", len(differences), differences)
+	}
+	difference := differences[0]
+	if got, want := difference.Path.String(), "/profiles/offline/model"; got != want {
+		t.Fatalf("difference path = %q, want %q", got, want)
+	}
+	if !strings.Contains(difference.Left.YAML, "earlier-model") || !strings.Contains(difference.Right.YAML, "later-model") {
+		t.Fatalf("difference snapshots = (%q, %q)", difference.Left.YAML, difference.Right.YAML)
+	}
+
+	reversed, err := right.Diff(left)
+	if err != nil {
+		t.Fatalf("compare reversed configurations: %v", err)
+	}
+	if len(reversed) != 1 || reversed[0].Path.String() != difference.Path.String() ||
+		!strings.Contains(reversed[0].Left.YAML, "later-model") ||
+		!strings.Contains(reversed[0].Right.YAML, "earlier-model") {
+		t.Fatalf("reversed difference = %+v", reversed)
+	}
+
+	equal := comparisonInspection(t, "earlier-model")
+	if equal, err := left.Diff(equal); err != nil || len(equal) != 0 {
+		t.Fatalf("compare equal configurations = (%+v, %v), want no differences", equal, err)
+	}
+	if equal, err := left.Diff(left); err != nil || len(equal) != 0 {
+		t.Fatalf("compare view with itself = (%+v, %v), want no differences", equal, err)
+	}
+}
+
+func TestConfigurationInspectionDiffRejectsZeroViews(t *testing.T) {
+	valid := comparisonInspection(t, "current-model")
+	var zero promptkit.ConfigurationInspection
+
+	for _, comparison := range []struct {
+		name        string
+		left, right promptkit.ConfigurationInspection
+	}{
+		{name: "zero left", left: zero, right: valid},
+		{name: "zero right", left: valid, right: zero},
+		{name: "both zero", left: zero, right: zero},
+	} {
+		t.Run(comparison.name, func(t *testing.T) {
+			differences, err := comparison.left.Diff(comparison.right)
+			if !errors.Is(err, promptkit.ErrInvalidConfig) || len(differences) != 0 {
+				t.Fatalf("Diff() = (%+v, %v), want ErrInvalidConfig and no records", differences, err)
+			}
+		})
+	}
+}
+
+func TestConfigurationInspectionDiffIgnoresSourceHistory(t *testing.T) {
+	left := comparisonInspectionFrom(t, "earlier source", "same-model", nil)
+	right := comparisonInspectionFrom(t, "later source", "same-model", nil)
+	if !hasSourceLabel(left, "earlier source") || !hasSourceLabel(right, "later source") {
+		t.Fatal("comparison fixtures did not retain their distinct source labels")
+	}
+
+	differences, err := left.Diff(right)
+	if err != nil || len(differences) != 0 {
+		t.Fatalf("compare equal effective values with different sources = (%+v, %v), want no differences", differences, err)
+	}
+}
+
+func TestConfigurationInspectionDiffReportsPresenceAndSequences(t *testing.T) {
+	left := comparisonInspectionFrom(t, "before", "same-model", map[string]any{
+		"sequence": []any{"earlier", "unchanged"},
+	})
+	right := comparisonInspectionFrom(t, "after", "same-model", map[string]any{
+		"nullable": nil,
+		"sequence": []any{"later", "unchanged"},
+	})
+
+	differences, err := left.Diff(right)
+	if err != nil {
+		t.Fatalf("compare presence and sequence values: %v", err)
+	}
+	if len(differences) != 2 {
+		t.Fatalf("difference count = %d, want 2: %+v", len(differences), differences)
+	}
+	byPath := differencesByPath(differences)
+	nullable, ok := byPath["/backends/local/extra_params/nullable"]
+	if !ok || nullable.Left.Present || !nullable.Right.Present || strings.TrimSpace(nullable.Right.YAML) != "null" {
+		t.Fatalf("absent-to-null difference = (%+v, %t)", nullable, ok)
+	}
+	sequence, ok := byPath["/backends/local/extra_params/sequence"]
+	if !ok || !sequence.Left.Present || !sequence.Right.Present ||
+		!strings.Contains(sequence.Left.YAML, "earlier") || !strings.Contains(sequence.Right.YAML, "later") {
+		t.Fatalf("atomic sequence difference = (%+v, %t)", sequence, ok)
+	}
+}
+
+func TestConfigurationInspectionDiffProtectsSensitiveValuesAndReturnedRecords(t *testing.T) {
+	left := comparisonInspectionFrom(t, "before", "same-model", map[string]any{
+		"token":   "left-secret-sentinel",
+		"visible": "before-value",
+	})
+	right := comparisonInspectionFrom(t, "after", "same-model", map[string]any{
+		"token":   "right-secret-sentinel",
+		"visible": "after-value",
+	})
+
+	differences, err := left.Diff(right)
+	if err != nil {
+		t.Fatalf("compare sensitive values: %v", err)
+	}
+	if len(differences) != 2 {
+		t.Fatalf("difference count = %d, want 2: %+v", len(differences), differences)
+	}
+	byPath := differencesByPath(differences)
+	secret, ok := byPath["/backends/local/extra_params/token"]
+	if !ok || !secret.Left.Redacted || !secret.Right.Redacted || !secret.HiddenChanged {
+		t.Fatalf("sensitive difference = (%+v, %t)", secret, ok)
+	}
+	if strings.Contains(secret.Left.YAML, "left-secret-sentinel") || strings.Contains(secret.Left.YAML, "right-secret-sentinel") ||
+		strings.Contains(secret.Right.YAML, "left-secret-sentinel") || strings.Contains(secret.Right.YAML, "right-secret-sentinel") {
+		t.Fatalf("sensitive snapshots disclosed a value: %+v", secret)
+	}
+	visible, ok := byPath["/backends/local/extra_params/visible"]
+	if !ok || visible.Left.Redacted || visible.Right.Redacted || visible.HiddenChanged ||
+		!strings.Contains(visible.Left.YAML, "before-value") || !strings.Contains(visible.Right.YAML, "after-value") {
+		t.Fatalf("public sibling difference = (%+v, %t)", visible, ok)
+	}
+
+	differences[0].Path = configkit.Keys("caller", "mutation")
+	differences[0].Left.YAML = "caller mutation"
+	differences[0].Right.YAML = "caller mutation"
+	again, err := left.Diff(right)
+	if err != nil {
+		t.Fatalf("repeat comparison after mutating returned records: %v", err)
+	}
+	if len(again) != 2 {
+		t.Fatalf("repeat difference count = %d, want 2: %+v", len(again), again)
+	}
+	againByPath := differencesByPath(again)
+	if _, exists := againByPath["/caller/mutation"]; exists {
+		t.Fatal("mutating returned path reached retained inspection")
+	}
+	if _, exists := againByPath["/backends/local/extra_params/token"]; !exists {
+		t.Fatalf("original sensitive path was lost after mutation: %+v", again)
+	}
+	if _, exists := againByPath["/backends/local/extra_params/visible"]; !exists {
+		t.Fatalf("original public path was lost after mutation: %+v", again)
+	}
+	for _, difference := range again {
+		if difference.Left.YAML == "caller mutation" || difference.Right.YAML == "caller mutation" {
+			t.Fatalf("mutating returned snapshots reached retained inspection: %+v", difference)
+		}
+	}
+}
+
+func TestConfigurationInspectionDiffMasksSensitiveShapeReplacement(t *testing.T) {
+	left := comparisonInspectionFrom(t, "before", "same-model", map[string]any{
+		"details": map[string]any{
+			"token": "nested-secret-sentinel",
+			"label": "public detail",
+		},
+	})
+	right := comparisonInspectionFrom(t, "after", "same-model", map[string]any{
+		"details": "replacement",
+	})
+
+	differences, err := left.Diff(right)
+	if err != nil {
+		t.Fatalf("compare sensitive shape replacement: %v", err)
+	}
+	if len(differences) != 1 || differences[0].Path.String() != "/backends/local/extra_params/details" {
+		t.Fatalf("shape replacement differences = %+v", differences)
+	}
+	difference := differences[0]
+	if !difference.Left.Redacted || !difference.Right.Redacted || !difference.HiddenChanged {
+		t.Fatalf("shape replacement masking = %+v", difference)
+	}
+	if strings.Contains(difference.Left.YAML, "nested-secret-sentinel") || strings.Contains(difference.Right.YAML, "nested-secret-sentinel") {
+		t.Fatalf("shape replacement disclosed a sensitive value: %+v", difference)
+	}
+}
+
+func comparisonInspection(t *testing.T, model string) promptkit.ConfigurationInspection {
+	t.Helper()
+	return comparisonInspectionFrom(t, "comparison fixture", model, nil)
+}
+
+func comparisonInspectionFrom(t *testing.T, source, model string, extraParams map[string]any) promptkit.ConfigurationInspection {
+	t.Helper()
+	backend := map[string]any{"endpoint": "https://example.invalid/v1"}
+	if extraParams != nil {
+		backend["extra_params"] = extraParams
+	}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: configkit.Mapping(source, map[string]any{
+			"backends": map[string]any{"local": backend},
+			"profiles": map[string]any{"offline": map[string]any{
+				"backend": "local",
+				"model":   model,
+			}},
+		})}}}},
+	}})
+	if err != nil {
+		t.Fatalf("construct comparison engine: %v", err)
+	}
+	inspection, err := engine.InspectConfiguration()
+	if err != nil {
+		t.Fatalf("inspect comparison engine: %v", err)
+	}
+	return inspection
+}
+
+func differencesByPath(differences []configkit.Difference) map[string]configkit.Difference {
+	byPath := make(map[string]configkit.Difference, len(differences))
+	for _, difference := range differences {
+		byPath[difference.Path.String()] = difference
+	}
+	return byPath
+}
+
+func hasSourceLabel(view promptkit.ConfigurationInspection, label string) bool {
+	for _, source := range view.Sources() {
+		if source.Label == label {
+			return true
+		}
+	}
+	return false
+}

configuration_inspection.go

@@ -0,0 +1,99 @@
+package promptkit
+
+import (
+	"fmt"
+	"io"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit/internal/configuration"
+)
+
+// ConfigurationInspection is a read-only view of one Engine's captured
+// effective configuration and source history. It does not expose typed
+// configuration values, filesystem contexts, or runtime credentials.
+// Report content may include application prompt and provider data, so callers
+// choose where and to whom to disclose it.
+type ConfigurationInspection struct {
+	source configuration.Inspection
+}
+
+// InspectConfiguration returns the ConfigKit inspection view captured by this
+// Engine. It does not reload sources, resolve credentials, or contact a model.
+// A nil Engine returns ErrInvalidConfig.
+func (e *Engine) InspectConfiguration() (ConfigurationInspection, error) {
+	if e == nil || e.internal == nil {
+		return ConfigurationInspection{}, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+	return ConfigurationInspection{source: e.internal.InspectConfiguration()}, nil
+}
+
+// Diff compares this view's captured effective configuration with other. This
+// view is the earlier (left) value and other is the later (right) value. The
+// comparison delegates to ConfigKit's frozen effective-value semantics, so
+// source or resolution history alone does not create differences. It does not
+// reread external sources or compare referenced file contents, resolved
+// credential contents, later execution preparation, or request-specific
+// overrides. Either zero view returns ErrInvalidConfig and no records; other
+// ConfigKit errors are returned unchanged. An empty diff is not evidence of
+// equivalent engine behavior or execution. Records can contain ordinary
+// application data, so callers choose where and to whom to disclose them.
+func (v ConfigurationInspection) Diff(other ConfigurationInspection) ([]configkit.Difference, error) {
+	if v.source == nil || other.source == nil {
+		return nil, ErrInvalidConfig
+	}
+	return configuration.Diff(v.source, other.source)
+}
+
+// Metadata returns detached effective presence and history records.
+func (v ConfigurationInspection) Metadata() configkit.Metadata {
+	if v.source == nil {
+		return configkit.Metadata{}
+	}
+	return v.source.Metadata()
+}
+
+// Sources returns a detached list of loaded source descriptions.
+func (v ConfigurationInspection) Sources() []configkit.SourceInfo {
+	if v.source == nil {
+		return nil
+	}
+	return v.source.Sources()
+}
+
+// Diagnostics returns detached nonfatal load diagnostics.
+func (v ConfigurationInspection) Diagnostics() []configkit.Diagnostic {
+	if v.source == nil {
+		return nil
+	}
+	return v.source.Diagnostics()
+}
+
+// Explain returns detached effective and historical values for a known path.
+// Dynamic credential-shaped provider keys are redacted by ConfigKit; ordinary
+// prompt and provider content is not automatically safe to disclose.
+func (v ConfigurationInspection) Explain(path configkit.Path) (configkit.Explanation, bool) {
+	if v.source == nil {
+		return configkit.Explanation{}, false
+	}
+	return v.source.Explain(path)
+}
+
+// WriteEffective writes ConfigKit's redacted effective YAML. Writer errors
+// remain available through errors.Is and errors.As. The zero view returns
+// ErrInvalidConfig.
+func (v ConfigurationInspection) WriteEffective(w io.Writer) error {
+	if v.source == nil {
+		return ErrInvalidConfig
+	}
+	return v.source.WriteEffective(w)
+}
+
+// WriteExplanation writes ConfigKit's redacted history for path. Unknown paths
+// produce no output; writer errors remain inspectable. The zero view returns
+// ErrInvalidConfig.
+func (v ConfigurationInspection) WriteExplanation(w io.Writer, path configkit.Path) error {
+	if v.source == nil {
+		return ErrInvalidConfig
+	}
+	return v.source.WriteExplanation(w, path)
+}

configuration_inspection_contract_test.go

@@ -0,0 +1,247 @@
+package promptkit_test
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{
+			{Inputs: []configkit.Input{{Source: configkit.Mapping("defaults", map[string]any{
+				"backends": map[string]any{"local": map[string]any{
+					"endpoint": "https://example.invalid/v1", "api_key_env": "OPTIONAL_KEY",
+					"extra_params": map[string]any{"token": "private-sentinel", "visible": "public-value"},
+				}},
+				"profiles": map[string]any{
+					"base.v1":    map[string]any{"backend": "local", "model": "default-model"},
+					"literal/id": map[string]any{"extends": "base.v1"},
+				},
+			})}}},
+			{Inputs: []configkit.Input{{Source: configkit.Mapping("operator", map[string]any{
+				"profiles": map[string]any{"base.v1": map[string]any{"model": "operator-model"}},
+			})}}},
+		},
+	}})
+	if err != nil {
+		t.Fatalf("construct layered engine: %v", err)
+	}
+	view, err := engine.InspectConfiguration()
+	if err != nil {
+		t.Fatalf("inspect configuration: %v", err)
+	}
+	modelPath := configkit.Keys("profiles", "base.v1", "model")
+	model, found := view.Explain(modelPath)
+	if !found || !strings.Contains(model.Effective.YAML, "operator-model") || len(model.History) < 2 {
+		t.Fatalf("layered model explanation = (%+v, %t)", model, found)
+	}
+	inheritedPath := configkit.Keys("profiles", "literal/id", "model")
+	inherited, found := view.Explain(inheritedPath)
+	if !found || !strings.Contains(inherited.Effective.YAML, "operator-model") || len(inherited.History) == 0 || !view.Metadata().IsExplicit(inheritedPath) {
+		t.Fatalf("literal-key inheritance explanation = (%+v, %t)", inherited, found)
+	}
+	if inheritedPath.String() != "/profiles/literal~1id/model" {
+		t.Fatalf("literal-key path = %q", inheritedPath)
+	}
+	secretPath := configkit.Keys("backends", "local", "extra_params", "token")
+	secret, found := view.Explain(secretPath)
+	if !found || !secret.Effective.Redacted || strings.Contains(secret.Effective.YAML, "private-sentinel") {
+		t.Fatalf("credential-shaped provider value was not redacted: (%+v, %t)", secret, found)
+	}
+	var effective bytes.Buffer
+	if err := view.WriteEffective(&effective); err != nil {
+		t.Fatalf("write effective configuration: %v", err)
+	}
+	if strings.Contains(effective.String(), "private-sentinel") || !strings.Contains(effective.String(), "OPTIONAL_KEY") || !strings.Contains(effective.String(), "public-value") {
+		t.Fatalf("unexpected effective report: %s", effective.String())
+	}
+	var explanation bytes.Buffer
+	if err := view.WriteExplanation(&explanation, secretPath); err != nil || strings.Contains(explanation.String(), "private-sentinel") {
+		t.Fatalf("secret explanation = (%q, %v)", explanation.String(), err)
+	}
+
+	sources := view.Sources()
+	if len(sources) == 0 {
+		t.Fatal("inspection returned no sources")
+	}
+	originalLabel := sources[0].Label
+	sources[0].Label = "changed"
+	if view.Sources()[0].Label != originalLabel {
+		t.Fatal("source mutation reached retained inspection")
+	}
+	model.History[0].After.YAML = "changed"
+	if again, ok := view.Explain(modelPath); !ok || again.History[0].After.YAML == "changed" {
+		t.Fatal("explanation mutation reached retained inspection")
+	}
+	history := view.Metadata().History(modelPath)
+	if len(history) == 0 {
+		t.Fatal("metadata history is empty")
+	}
+	history[0].Reasons = []string{"changed"}
+	if again := view.Metadata().History(modelPath); len(again) == 0 || (len(again[0].Reasons) > 0 && again[0].Reasons[0] == "changed") {
+		t.Fatal("metadata mutation reached retained inspection")
+	}
+}
+
+type failingInspectionWriter struct{ err error }
+
+func (writer failingInspectionWriter) Write([]byte) (int, error) { return 0, writer.err }
+
+func TestConfigurationInspectionZeroAndWriterErrors(t *testing.T) {
+	var engine *promptkit.Engine
+	if _, err := engine.InspectConfiguration(); !errors.Is(err, promptkit.ErrInvalidConfig) {
+		t.Fatalf("nil engine inspection = %v", err)
+	}
+	var empty promptkit.ConfigurationInspection
+	path := configkit.Keys("profiles", "missing")
+	if _, found := empty.Metadata().Lookup(path); found || len(empty.Sources()) != 0 || len(empty.Diagnostics()) != 0 {
+		t.Fatal("zero inspection reported configuration")
+	}
+	if _, found := empty.Explain(path); found {
+		t.Fatal("zero inspection explained an unknown path")
+	}
+	if err := empty.WriteEffective(io.Discard); !errors.Is(err, promptkit.ErrInvalidConfig) {
+		t.Fatalf("zero effective writer = %v", err)
+	}
+	if err := empty.WriteExplanation(io.Discard, path); !errors.Is(err, promptkit.ErrInvalidConfig) {
+		t.Fatalf("zero explanation writer = %v", err)
+	}
+
+	valid, err := promptkit.NewEngine(promptkit.Config{})
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	view, err := valid.InspectConfiguration()
+	if err != nil {
+		t.Fatalf("inspect configuration: %v", err)
+	}
+	writeErr := errors.New("report writer failed")
+	writer := failingInspectionWriter{err: writeErr}
+	if err := view.WriteEffective(writer); !errors.Is(err, writeErr) {
+		t.Fatalf("effective writer error = %v", err)
+	}
+	if err := view.WriteExplanation(writer, configkit.Keys("profiles", "mistral-small-3", "model")); !errors.Is(err, writeErr) {
+		t.Fatalf("explanation writer error = %v", err)
+	}
+}
+
+func TestConfigurationInspectionUsesCapturedSourcesWithoutProviderCalls(t *testing.T) {
+	source := &unreadableAfterConstructionFS{FS: fstest.MapFS{
+		"config/prompt.yaml": &fstest.MapFile{Data: []byte(capturedPromptDocument)},
+		"config/body.tmpl":   &fstest.MapFile{Data: []byte("hello")},
+	}}
+	inputs, err := promptkit.FilesystemInputs("inspection source", source, "config")
+	if err != nil {
+		t.Fatalf("select inputs: %v", err)
+	}
+	otherSource := &unreadableAfterConstructionFS{FS: fstest.MapFS{
+		"config/prompt.yaml": &fstest.MapFile{Data: []byte(capturedPromptDocument)},
+		"config/body.tmpl":   &fstest.MapFile{Data: []byte("hello")},
+	}}
+	otherInputs, err := promptkit.FilesystemInputs("comparison source", otherSource, "config")
+	if err != nil {
+		t.Fatalf("select comparison inputs: %v", err)
+	}
+	var modelCalls atomic.Int64
+	client := countingLLMClient{calls: &modelCalls}
+	endpoint, model, otherModel := "https://example.invalid/v1", "offline-model", "comparison-model"
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: inputs}, {Inputs: []configkit.Input{{Source: promptkit.ProfileSource("offline profile", map[string]promptkit.ProfileConfig{
+			"offline": {Endpoint: &endpoint, Model: &model},
+		})}}}},
+	}}, promptkit.WithLLMClient(client))
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	otherEngine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: otherInputs}, {Inputs: []configkit.Input{{Source: promptkit.ProfileSource("comparison profile", map[string]promptkit.ProfileConfig{
+			"offline": {Endpoint: &endpoint, Model: &otherModel},
+		})}}}},
+	}}, promptkit.WithLLMClient(client))
+	if err != nil {
+		t.Fatalf("construct comparison engine: %v", err)
+	}
+	source.closed.Store(true)
+	otherSource.closed.Store(true)
+	view, err := engine.InspectConfiguration()
+	if err != nil {
+		t.Fatalf("inspect configuration: %v", err)
+	}
+	otherView, err := otherEngine.InspectConfiguration()
+	if err != nil {
+		t.Fatalf("inspect comparison configuration: %v", err)
+	}
+	var expected configkit.Difference
+	for index := 0; index < 3; index++ {
+		differences, err := view.Diff(otherView)
+		if err != nil || len(differences) != 1 {
+			t.Fatalf("compare captured configurations = (%+v, %v), want one difference", differences, err)
+		}
+		if differences[0].Path.String() != "/profiles/offline/model" ||
+			!strings.Contains(differences[0].Left.YAML, model) ||
+			!strings.Contains(differences[0].Right.YAML, otherModel) {
+			t.Fatalf("captured difference = %+v", differences[0])
+		}
+		if index == 0 {
+			expected = differences[0]
+		} else if differences[0].Path.String() != expected.Path.String() ||
+			differences[0].Left.YAML != expected.Left.YAML || differences[0].Right.YAML != expected.Right.YAML {
+			t.Fatalf("captured difference changed between reads: first=%+v next=%+v", expected, differences[0])
+		}
+	}
+	if _, found := view.Explain(configkit.Keys("prompts", "captured", "1", "messages").Index(0).Key("content_file")); !found {
+		t.Fatal("captured prompt content path was not inspectable")
+	}
+	var report bytes.Buffer
+	if err := view.WriteEffective(&report); err != nil || !strings.Contains(report.String(), "captured") {
+		t.Fatalf("write captured report = (%q, %v)", report.String(), err)
+	}
+	if source.reads.Load() != 0 || otherSource.reads.Load() != 0 || modelCalls.Load() != 0 {
+		t.Fatalf("inspection or comparison touched source or provider: reads=(%d,%d) requests=%d", source.reads.Load(), otherSource.reads.Load(), modelCalls.Load())
+	}
+
+	const calls = 24
+	var wait sync.WaitGroup
+	errs := make(chan error, calls)
+	for index := 0; index < calls; index++ {
+		wait.Add(1)
+		go func(index int) {
+			defer wait.Done()
+			switch index % 3 {
+			case 0:
+				_, err := engine.InspectConfiguration()
+				errs <- err
+			case 1:
+				_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "captured", SessionID: "direct"})
+				errs <- err
+			case 2:
+				differences, err := view.Diff(otherView)
+				if err == nil && (len(differences) != 1 || differences[0].Path.String() != expected.Path.String() ||
+					differences[0].Left.YAML != expected.Left.YAML || differences[0].Right.YAML != expected.Right.YAML) {
+					err = errors.New("concurrent comparison returned an unexpected difference")
+				}
+				errs <- err
+			}
+		}(index)
+	}
+	wait.Wait()
+	close(errs)
+	for err := range errs {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	if source.reads.Load() != 0 || otherSource.reads.Load() != 0 || modelCalls.Load() != calls/3 {
+		t.Fatalf("concurrent inspection/comparison/execution = source reads (%d,%d), model calls %d", source.reads.Load(), otherSource.reads.Load(), modelCalls.Load())
+	}
+}

configuration_sources.go

@@ -0,0 +1,83 @@
+package promptkit
+
+import (
+	"context"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
+)
+
+// DirectoryInputs selects the YAML files beneath an explicitly chosen host
+// directory as one strict-composition group. The returned order is deterministic;
+// it does not assign precedence among those files. Discovery is recursive over
+// .yaml and .yml files. An unavailable root fails; an empty directory yields
+// no inputs. Imports are disabled unless callers enable them on the inputs.
+func DirectoryInputs(directory string) ([]configkit.Input, error) {
+	root, err := filepath.Abs(directory)
+	if err != nil {
+		return nil, fmt.Errorf("resolve configuration directory: %w", err)
+	}
+	info, err := os.Stat(root)
+	if err != nil {
+		return nil, fmt.Errorf("inspect configuration directory: %w", err)
+	}
+	if !info.IsDir() {
+		return nil, fmt.Errorf("configuration root is not a directory: %s", root)
+	}
+	root, err = filepath.EvalSymlinks(root)
+	if err != nil {
+		return nil, fmt.Errorf("resolve configuration directory: %w", err)
+	}
+	files, err := filecatalog.FindYAMLFiles(context.Background(), root)
+	if err != nil {
+		return nil, fmt.Errorf("discover configuration files: %w", err)
+	}
+	inputs := make([]configkit.Input, 0, len(files))
+	for _, file := range files {
+		inputs = append(inputs, configkit.Input{
+			Source:     configkit.File(file),
+			Filesystem: &configkit.FilesystemContext{Root: root},
+		})
+	}
+	return inputs, nil
+}
+
+// FilesystemInputs selects the YAML files beneath root in a caller-owned
+// filesystem as one strict-composition group. Discovery is recursive over
+// .yaml and .yml files, with deterministic ordering and no precedence among
+// files. The caller retains ownership of the filesystem; an unavailable root
+// fails and an empty directory yields no inputs. Imports remain disabled unless
+// callers enable them on inputs.
+func FilesystemInputs(label string, filesystem fs.FS, root string) ([]configkit.Input, error) {
+	if filesystem == nil {
+		return nil, fmt.Errorf("configuration filesystem is nil")
+	}
+	root = filecatalog.CleanFSRoot(root)
+	if !fs.ValidPath(root) {
+		return nil, fmt.Errorf("configuration root is not a valid filesystem path: %s", root)
+	}
+	info, err := fs.Stat(filesystem, root)
+	if err != nil {
+		return nil, fmt.Errorf("inspect configuration root: %w", err)
+	}
+	if !info.IsDir() {
+		return nil, fmt.Errorf("configuration root is not a directory: %s", root)
+	}
+	files, err := filecatalog.FindFSYAMLFiles(context.Background(), filesystem, root)
+	if err != nil {
+		return nil, fmt.Errorf("discover configuration files: %w", err)
+	}
+	namespace := configkit.NewFilesystem(label, filesystem)
+	inputs := make([]configkit.Input, 0, len(files))
+	for _, file := range files {
+		inputs = append(inputs, configkit.Input{
+			Source:     namespace.File(file),
+			Filesystem: &configkit.FilesystemContext{FS: &namespace, Root: root},
+		})
+	}
+	return inputs, nil
+}

configuration_sources_test.go

@@ -0,0 +1,83 @@
+package promptkit_test
+
+import (
+	"io/fs"
+	"os"
+	"path/filepath"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+type sourceDocument struct {
+	First  string `yaml:"first"`
+	Second string `yaml:"second"`
+}
+
+func TestDirectoryInputsSelectsSortedFilesWithRootContext(t *testing.T) {
+	root := t.TempDir()
+	if err := os.Mkdir(filepath.Join(root, "nested"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	for name, data := range map[string]string{
+		"a.yaml": "first: one\n", "nested/b.yml": "second: two\n", "nested/body.md": "message",
+	} {
+		if err := os.WriteFile(filepath.Join(root, name), []byte(data), 0o600); err != nil {
+			t.Fatal(err)
+		}
+	}
+	inputs, err := promptkit.DirectoryInputs(root)
+	if err != nil || len(inputs) != 2 || inputs[0].Imports || inputs[1].Imports {
+		t.Fatalf("directory inputs: %+v, %v", inputs, err)
+	}
+	assertSourceInputs(t, inputs, "message")
+	if _, err := promptkit.DirectoryInputs(filepath.Join(root, "missing")); err == nil {
+		t.Fatal("missing selected directory succeeded")
+	}
+	if empty, err := promptkit.DirectoryInputs(t.TempDir()); err != nil || len(empty) != 0 {
+		t.Fatalf("empty directory: %+v, %v", empty, err)
+	}
+}
+
+func TestFilesystemInputsSelectsEmbeddedFilesWithRootContext(t *testing.T) {
+	files := fstest.MapFS{
+		"settings/a.yaml":         {Data: []byte("first: one\n")},
+		"settings/nested/b.yml":   {Data: []byte("second: two\n")},
+		"settings/nested/body.md": {Data: []byte("message")},
+	}
+	inputs, err := promptkit.FilesystemInputs("embedded", files, "settings")
+	if err != nil || len(inputs) != 2 || inputs[0].Imports || inputs[1].Imports {
+		t.Fatalf("filesystem inputs: %+v, %v", inputs, err)
+	}
+	assertSourceInputs(t, inputs, "message")
+	if _, err := promptkit.FilesystemInputs("embedded", files, "missing"); err == nil {
+		t.Fatal("missing selected root succeeded")
+	}
+}
+
+func assertSourceInputs(t *testing.T, inputs []configkit.Input, expected string) {
+	t.Helper()
+	result, err := configkit.LoadLayered(configkit.Input{Source: configkit.NoSource()}, configkit.LayeredOptions[sourceDocument]{
+		Underlays: []configkit.Underlay{{Inputs: inputs}},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if result.Config.First != "one" || result.Config.Second != "two" {
+		t.Fatalf("configuration files did not compose: %+v", result.Config)
+	}
+	field, found := result.Metadata().Lookup(configkit.Keys("second"))
+	if !found || len(field.Origin.Locations) != 1 {
+		t.Fatalf("missing second-field origin: %+v", field)
+	}
+	context, found := result.SourceContext(field.Origin.Locations[0].SourceID)
+	if !found || context.BaseDir != "nested" {
+		t.Fatalf("unexpected rooted source context: %+v", context)
+	}
+	data, err := fs.ReadFile(context.FS, "nested/body.md")
+	if err != nil || string(data) != expected {
+		t.Fatalf("rooted resource read: %q, %v", data, err)
+	}
+}

construction_contract_test.go

@@ -0,0 +1,231 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestProfileEnvironmentLookupAcceptsNonShellNames(t *testing.T) {
+	const name = "promptkit.test-api-key"
+	for _, test := range []struct {
+		name, value, direct, want string
+		unset, clear              bool
+	}{
+		{name: "set", value: " synthetic-key ", want: "Bearer synthetic-key"},
+		{name: "empty"},
+		{name: "whitespace", value: "  "},
+		{name: "unset", unset: true},
+		{name: "direct precedence", value: "synthetic-key", direct: "direct-key", want: "Bearer direct-key"},
+		{name: "explicit clear", value: "synthetic-key", clear: true},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			t.Setenv(name, test.value)
+			if test.unset {
+				if err := os.Unsetenv(name); err != nil {
+					t.Fatal(err)
+				}
+			}
+			cfg := simpleConfig("hello")
+			cfg.Sources.Underlays = []configkit.Underlay{{Inputs: []configkit.Input{*cfg.Sources.Main}}}
+			lowerName := name
+			cfg.Sources.Underlays = append(cfg.Sources.Underlays, configkit.Underlay{Inputs: []configkit.Input{{
+				Source: promptkit.ProfileSource("default credentials", map[string]promptkit.ProfileConfig{
+					"profile": {APIKeyEnv: &lowerName},
+				}),
+			}}})
+			env := " " + name + " "
+			if test.clear {
+				env = ""
+			}
+			cfg.Sources.Main = &configkit.Input{Source: promptkit.ProfileSource("credentials", map[string]promptkit.ProfileConfig{
+				"profile": {APIKeyEnv: &env},
+			})}
+			var calls int
+			var auth string
+			cfg.HTTPClient = &http.Client{Transport: roundTripFunc(func(request *http.Request) (*http.Response, error) {
+				calls++
+				auth = request.Header.Get("Authorization")
+				return &http.Response{StatusCode: http.StatusOK, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(`{"choices":[{"message":{"content":"ok"}}]}`))}, nil
+			})}
+			engine, err := promptkit.NewEngine(cfg)
+			if err != nil {
+				t.Fatal(err)
+			}
+			_, err = engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", APIKey: test.direct})
+			if err != nil || calls != 1 || auth != test.want {
+				t.Fatalf("run: err=%v calls=%d authorization=%q, want %q", err, calls, auth, test.want)
+			}
+		})
+	}
+}
+
+func TestPublicConstructionWithLocalBackendAndProgrammaticProfile(t *testing.T) {
+	backendID, model := "workstation", "offline-model"
+	prompt := configkit.Input{Source: configkit.Mapping("prompt", map[string]any{
+		"prompts": map[string]any{"hello": map[string]any{"1": map[string]any{
+			"default_profile": "offline",
+			"messages":        []any{map[string]any{"role": "user", "content": "Hello"}},
+			"output":          map[string]any{"format": "text", "validation_mode": "none"},
+		}}},
+	})}
+	client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "reply"}}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{
+			{Source: promptkit.BackendSource("local backend", map[string]promptkit.BackendConfig{
+				backendID: promptkit.LocalBackend("http://localhost:8000/v1", 2),
+			})},
+			{Source: promptkit.ProfileSource("profile", map[string]promptkit.ProfileConfig{
+				"offline": {Backend: &backendID, Model: &model},
+			})},
+		}}},
+		Main: &prompt,
+	}}, promptkit.WithLLMClient(client))
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	result, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "hello"})
+	if err != nil {
+		t.Fatalf("run prompt: %v", err)
+	}
+	if result.RawOutput != "reply" || result.SelectedBackendID != backendID || result.EffectiveModelParams.Model != model {
+		t.Fatalf("unexpected result: %+v", result)
+	}
+	if len(client.requests) != 1 || client.requests[0].Target.Endpoint != "http://localhost:8000/v1" {
+		t.Fatalf("unexpected model request: %+v", client.requests)
+	}
+}
+
+func TestPublicConstructionCombinesEmbeddedOperatorAndProgrammaticInputs(t *testing.T) {
+	embedded, err := promptkit.FilesystemInputs("embedded defaults", fstest.MapFS{
+		"config/default.yaml": &fstest.MapFile{Data: []byte(`
+profiles:
+  application:
+    endpoint: https://example.invalid/v1
+    model: default-model
+prompts:
+  hello:
+    "1":
+      default_profile: application
+      messages:
+        - role: user
+          content: Hello
+      output:
+        format: text
+        validation_mode: none
+`)},
+	}, "config")
+	if err != nil {
+		t.Fatalf("select embedded defaults: %v", err)
+	}
+	operatorDirectory := t.TempDir()
+	if err := os.WriteFile(filepath.Join(operatorDirectory, "override.yaml"), []byte("profiles:\n  application:\n    model: operator-model\n"), 0o644); err != nil {
+		t.Fatalf("write operator input: %v", err)
+	}
+	operator, err := promptkit.DirectoryInputs(operatorDirectory)
+	if err != nil {
+		t.Fatalf("select operator inputs: %v", err)
+	}
+	programmaticModel := "programmatic-model"
+	programmatic := configkit.Input{Source: promptkit.ProfileSource("programmatic settings", map[string]promptkit.ProfileConfig{
+		"application": {Model: &programmaticModel},
+	})}
+	for _, test := range []struct {
+		name  string
+		apply bool
+		want  string
+	}{
+		{name: "embedded", want: "default-model"},
+		{name: "operator", apply: true, want: "operator-model"},
+		{name: "programmatic", apply: true, want: "programmatic-model"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			underlays := []configkit.Underlay{{Inputs: embedded}}
+			if test.apply {
+				underlays = append(underlays, configkit.Underlay{Inputs: operator})
+			}
+			if test.name == "programmatic" {
+				underlays = append(underlays, configkit.Underlay{Inputs: []configkit.Input{programmatic}})
+			}
+			engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: underlays}})
+			if err != nil {
+				t.Fatalf("construct engine: %v", err)
+			}
+			prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "hello"})
+			if err != nil {
+				t.Fatalf("prepare prompt: %v", err)
+			}
+			if prepared.EffectiveModelParams.Model != test.want {
+				t.Fatalf("model = %q, want %q", prepared.EffectiveModelParams.Model, test.want)
+			}
+		})
+	}
+}
+
+func TestPublicConstructionPreservesConfigKitLoadError(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
+		Source: configkit.Mapping("invalid", map[string]any{"profiles": map[string]any{"broken": map[string]any{"unknown": true}}}),
+	}}})
+	var loadErr *configkit.LoadError
+	if engine != nil || !errors.Is(err, promptkit.ErrInvalidConfig) || !errors.As(err, &loadErr) {
+		t.Fatalf("construction = (%#v, %v), want invalid config preserving LoadError", engine, err)
+	}
+}
+
+func TestPublicConstructionRejectsSuppliedInvalidMainInput(t *testing.T) {
+	main := configkit.Input{}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &main}})
+	if engine != nil || !errors.Is(err, promptkit.ErrInvalidConfig) {
+		t.Fatalf("construction = (%#v, %v), want invalid supplied main input", engine, err)
+	}
+}
+
+func TestPublicConstructionCanOverlayMaintainedBackend(t *testing.T) {
+	endpoint := "http://local-openrouter.example/v1"
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.BackendSource("operator backend", map[string]promptkit.BackendConfig{
+			promptkit.BackendOpenRouter: {Endpoint: &endpoint},
+		})}}}},
+	}})
+	if err != nil {
+		t.Fatalf("construct engine with backend overlay: %v", err)
+	}
+	inspection, err := engine.InspectProfile(context.Background(), "mistral-small-3")
+	if err != nil {
+		t.Fatalf("inspect catalog profile: %v", err)
+	}
+	if inspection.EffectiveModelParams.Endpoint != endpoint {
+		t.Fatalf("effective endpoint = %q, want %q", inspection.EffectiveModelParams.Endpoint, endpoint)
+	}
+}
+
+func TestPreparationRejectsAmbiguousPromptVersion(t *testing.T) {
+	definition := func(content string) map[string]any {
+		return map[string]any{
+			"messages": []any{map[string]any{"role": "user", "content": content}},
+			"output":   map[string]any{"format": "text", "validation_mode": "none"},
+		}
+	}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Main: &configkit.Input{Source: configkit.Mapping("versions", map[string]any{
+			"prompts": map[string]any{"hello": map[string]any{
+				"1": definition("first"), "2": definition("second"),
+			}},
+		})},
+	}})
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	if result, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "hello"}); result != nil || !errors.Is(err, promptkit.ErrPromptLoad) {
+		t.Fatalf("ambiguous preparation = (%#v, %v), want ErrPromptLoad", result, err)
+	}
+}

consumer_test_helpers_test.go

@@ -0,0 +1,73 @@
+package promptkit_test
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"sync/atomic"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+type fakeLLMClient struct {
+	response  *promptkit.GenerateResponse
+	responses []*promptkit.GenerateResponse
+	err       error
+	requests  []promptkit.GenerateRequest
+}
+
+func (f *fakeLLMClient) Generate(_ context.Context, request promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
+	f.requests = append(f.requests, request)
+	if f.err != nil {
+		return nil, f.err
+	}
+	if len(f.responses) > 0 {
+		index := len(f.requests) - 1
+		if index >= len(f.responses) {
+			return nil, fmt.Errorf("no response configured for generation %d", index+1)
+		}
+		return f.responses[index], nil
+	}
+	return f.response, nil
+}
+
+type countingLLMClient struct{ calls *atomic.Int64 }
+
+func (client countingLLMClient) Generate(context.Context, promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
+	if client.calls != nil {
+		client.calls.Add(1)
+	}
+	return &promptkit.GenerateResponse{Content: "ok"}, nil
+}
+
+type roundTripFunc func(*http.Request) (*http.Response, error)
+
+func (f roundTripFunc) RoundTrip(request *http.Request) (*http.Response, error) { return f(request) }
+
+// newSimpleEngine exercises the public constructor with a complete keyed
+// document. Tests that need layering or resources select their own inputs.
+func newSimpleEngine(t *testing.T, content string, options ...promptkit.Option) *promptkit.Engine {
+	t.Helper()
+	engine, err := promptkit.NewEngine(simpleConfig(content), options...)
+	if err != nil {
+		t.Fatalf("construct consumer test engine: %v", err)
+	}
+	return engine
+}
+
+func simpleConfig(content string) promptkit.Config {
+	return promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Main: &configkit.Input{Source: configkit.Mapping("consumer test", map[string]any{
+			"profiles": map[string]any{"profile": map[string]any{
+				"endpoint": "http://example.test/v1", "model": "model",
+			}},
+			"prompts": map[string]any{"prompt": map[string]any{"1": map[string]any{
+				"default_profile": "profile",
+				"messages":        []any{map[string]any{"role": "user", "content": content}},
+				"output":          map[string]any{"format": "text", "validation_mode": "none"},
+			}}},
+		})},
+	}}
+}

convert.go

@@ -1,7 +1,9 @@
 package promptkit
 
 import (
+	"fmt"
 	"reflect"
+	"unicode/utf8"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
@@ -12,19 +14,59 @@ func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) {
 	if err != nil {
 		return domain.RunRequest{}, err
 	}
+	appendedMessages, err := toDomainAppendedMessages(req.AppendedMessages)
+	if err != nil {
+		return domain.RunRequest{}, err
+	}
+	var observer domain.ExecutionObserver
+	if req.Observer != nil {
+		callback := req.Observer
+		observer = func(event domain.ExecutionEvent) { callback(fromDomainExecutionEvent(event)) }
+	}
 	return domain.RunRequest{
-		PromptID:      req.PromptID,
-		PromptVersion: req.PromptVersion,
-		ProfileID:     req.ProfileID,
-		SessionID:     req.SessionID,
-		APIKey:        req.APIKey,
-		Inputs:        toDomainArtifactRefMap(req.Inputs),
-		Vars:          copyStringMap(req.Vars),
-		Execution:     execution,
-		Validation:    toDomainOutputContractPtr(req.Validation),
+		PromptID:         req.PromptID,
+		PromptVersion:    req.PromptVersion,
+		ProfileID:        req.ProfileID,
+		SessionID:        req.SessionID,
+		APIKey:           req.APIKey,
+		Inputs:           toDomainArtifactRefMap(req.Inputs),
+		Vars:             copyStringMap(req.Vars),
+		Execution:        execution,
+		Validation:       toDomainOutputContractPtr(req.Validation),
+		AppendedMessages: appendedMessages,
+		TransientRetries: copyIntPtr(req.TransientRetries),
+		Observer:         observer,
 	}, nil
 }
 
+func toDomainAppendedMessages(messages []RenderedMessage) ([]domain.RenderedMessage, error) {
+	if messages == nil {
+		return nil, nil
+	}
+
+	converted := make([]domain.RenderedMessage, len(messages))
+	for index, message := range messages {
+		if !utf8.ValidString(message.Content) {
+			return nil, fmt.Errorf("appended message %d content must be valid UTF-8", index)
+		}
+		role, err := domain.NormalizeMessageRole(message.Role)
+		if err != nil {
+			return nil, fmt.Errorf("appended message %d role: %w", index, err)
+		}
+
+		cacheControl, err := domain.NormalizeCacheControl(toDomainCacheControl(message.CacheControl))
+		if err != nil {
+			return nil, fmt.Errorf("appended message %d cache_control: %w", index, err)
+		}
+		converted[index] = domain.RenderedMessage{
+			Role:         role,
+			Content:      message.Content,
+			CacheControl: cacheControl,
+		}
+	}
+	return converted, nil
+}
+
 func fromDomainPreparedRun(prepared *domain.PreparedRun) *PreparedRun {
 	if prepared == nil {
 		return nil
@@ -90,8 +132,22 @@ func toDomainGenerateResponse(resp *GenerateResponse) *domain.GenerateResponse {
 		return nil
 	}
 	return &domain.GenerateResponse{
-		Content: resp.Content,
-		Usage:   toDomainTokenUsage(resp.Usage),
+		Content:       resp.Content,
+		Usage:         toDomainTokenUsage(resp.Usage),
+		UsagePresence: toDomainTokenUsagePresence(resp.UsagePresence),
+	}
+}
+
+func toDomainTokenUsagePresence(presence *TokenUsagePresence) *domain.TokenUsagePresence {
+	if presence == nil {
+		return nil
+	}
+	return &domain.TokenUsagePresence{
+		PromptTokens:     presence.PromptTokens,
+		CompletionTokens: presence.CompletionTokens,
+		TotalTokens:      presence.TotalTokens,
+		CachedTokens:     presence.CachedTokens,
+		CacheWriteTokens: presence.CacheWriteTokens,
 	}
 }
 
@@ -170,6 +226,40 @@ func fromDomainExecutionTarget(target domain.ExecutionTarget) ExecutionTarget {
 	}
 }
 
+func fromDomainProfileInspection(inspection *domain.ProfileInspection) *ProfileInspection {
+	if inspection == nil {
+		return nil
+	}
+	return &ProfileInspection{
+		ProfileID:            inspection.ProfileID,
+		EffectiveModelParams: fromDomainExecutionTarget(inspection.EffectiveModelParams),
+		APIKeyRequired:       inspection.APIKeyRequired,
+	}
+}
+
+func fromDomainPromptInspection(inspection *domain.PromptInspection) *PromptInspection {
+	if inspection == nil {
+		return nil
+	}
+	inputs := make([]PromptInputDefinition, len(inspection.Inputs))
+	for i, input := range inspection.Inputs {
+		inputs[i] = PromptInputDefinition{
+			Name:        input.Name,
+			Required:    input.Required,
+			ContentType: input.ContentType,
+			Description: input.Description,
+		}
+	}
+	return &PromptInspection{
+		PromptID:         inspection.PromptID,
+		PromptVersion:    inspection.PromptVersion,
+		PromptHash:       inspection.PromptHash,
+		DefaultProfileID: inspection.DefaultProfileID,
+		Inputs:           inputs,
+		OutputContract:   fromDomainOutputContract(inspection.OutputContract),
+	}
+}
+
 func fromDomainExecutionTargetPresence(presence domain.ExecutionTargetPresence) ExecutionTargetPresence {
 	return ExecutionTargetPresence{
 		Temperature:    presence.Temperature,
@@ -251,6 +341,16 @@ func fromDomainRenderedMessages(messages []domain.RenderedMessage) []RenderedMes
 	return out
 }
 
+func toDomainCacheControl(cacheControl *CacheControl) *domain.CacheControl {
+	if cacheControl == nil {
+		return nil
+	}
+	return &domain.CacheControl{
+		Type: domain.CacheControlType(cacheControl.Type),
+		TTL:  cacheControl.TTL,
+	}
+}
+
 func fromDomainCacheControl(cacheControl *domain.CacheControl) *CacheControl {
 	if cacheControl == nil {
 		return nil

convert_appended_messages_test.go

@@ -0,0 +1,108 @@
+package promptkit
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestToDomainAppendedMessages(t *testing.T) {
+	tests := []struct {
+		name        string
+		messages    []RenderedMessage
+		want        []domain.RenderedMessage
+		wantErr     string
+		privateRole string
+		privateText string
+	}{
+		{
+			name: "normalizes supported roles",
+			messages: []RenderedMessage{
+				{Role: " DeVeLoPeR ", Content: "developer"},
+				{Role: "SyStEm", Content: "system"},
+				{Role: "\tUsEr\n", Content: "user"},
+				{Role: "assistant", Content: "assistant"},
+			},
+			want: []domain.RenderedMessage{
+				{Role: domain.RoleDeveloper, Content: "developer"},
+				{Role: domain.RoleSystem, Content: "system"},
+				{Role: domain.RoleUser, Content: "user"},
+				{Role: domain.RoleAssistant, Content: "assistant"},
+			},
+		},
+		{
+			name:     "invalid role UTF-8",
+			messages: []RenderedMessage{{Role: string([]byte{0xff}), Content: "private-content"}},
+			wantErr:  "appended message 0 role",
+		},
+		{
+			name:     "invalid content UTF-8",
+			messages: []RenderedMessage{{Role: "private-role", Content: string([]byte{0xff})}},
+			wantErr:  "appended message 0 content",
+		},
+		{
+			name:        "unsupported role",
+			messages:    []RenderedMessage{{Role: "private-role", Content: "private-content"}},
+			wantErr:     "appended message 0 role",
+			privateRole: "private-role",
+			privateText: "private-content",
+		},
+		{
+			name:        "invalid cache control",
+			messages:    []RenderedMessage{{Role: RoleUser, Content: "private-content", CacheControl: &CacheControl{Type: "private-cache"}}},
+			wantErr:     "appended message 0 cache_control",
+			privateText: "private-content",
+		},
+		{
+			name: "empty and whitespace content",
+			messages: []RenderedMessage{
+				{Role: RoleUser, Content: ""},
+				{Role: RoleAssistant, Content: " \t\n "},
+			},
+			want: []domain.RenderedMessage{
+				{Role: domain.RoleUser, Content: ""},
+				{Role: domain.RoleAssistant, Content: " \t\n "},
+			},
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			got, err := toDomainAppendedMessages(test.messages)
+			if test.wantErr != "" {
+				if err == nil || !strings.Contains(err.Error(), test.wantErr) {
+					t.Fatalf("error = %v, want %q", err, test.wantErr)
+				}
+				for _, privateValue := range []string{test.privateRole, test.privateText} {
+					if privateValue != "" && strings.Contains(err.Error(), privateValue) {
+						t.Fatalf("error exposed appended message data %q: %v", privateValue, err)
+					}
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("toDomainAppendedMessages() error = %v", err)
+			}
+			if !reflect.DeepEqual(got, test.want) {
+				t.Fatalf("toDomainAppendedMessages() = %#v, want %#v", got, test.want)
+			}
+		})
+	}
+}
+
+func TestToDomainAppendedMessagesCopiesCacheControl(t *testing.T) {
+	cacheControl := &CacheControl{Type: CacheControlEphemeral, TTL: "1h"}
+	messages := []RenderedMessage{{Role: RoleUser, Content: "content", CacheControl: cacheControl}}
+	request, err := toDomainRunRequest(RunRequest{AppendedMessages: messages})
+	if err != nil {
+		t.Fatalf("toDomainRunRequest() error = %v", err)
+	}
+
+	messages[0].Content = "changed"
+	cacheControl.TTL = ""
+	if request.AppendedMessages[0].Content != "content" || request.AppendedMessages[0].CacheControl.TTL != "1h" {
+		t.Fatalf("domain request did not retain an independent appended-message copy: %#v", request.AppendedMessages)
+	}
+}

doc.go

@@ -1,25 +1,36 @@
 // Package promptkit provides an embeddable engine for preparing and executing
 // prompt-defined LLM workflows.
 //
-// Applications construct an [Engine] with [NewEngine], select filesystem or
-// in-memory sources and optional engine-scoped [Backend] registrations, and
-// call [Engine.Prepare] or [Engine.Run]. Concrete registries, repositories,
+// Applications construct an [Engine] with [NewEngine], select ConfigKit
+// sources for prompts, profiles, and backends, and
+// call [Engine.InspectConfiguration], [Engine.InspectPrompt],
+// [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.ListProfiles],
+// [Engine.ListPrompts], [Engine.Prepare], [Engine.PrepareExecution],
+// [Engine.Run], or [Engine.RunPrepared]. Concrete registries, repositories,
 // validators, and the built-in OpenAI-compatible client remain internal
 // implementation details.
 //
 // # Concurrency and ownership
 //
-// An Engine supports concurrent Prepare and Run calls. Engine-local backend
-// policies bound admitted Run calls and model generations where configured,
-// while different backend pools and unlimited backends continue independently.
-// An injected [LLMClient] or [ArtifactReader] can therefore still receive
-// concurrent calls and must be safe for that use.
+// An Engine supports concurrent InspectConfiguration, InspectPrompt,
+// InspectRequirements, InspectProfile, ListProfiles, ListPrompts, Prepare,
+// PrepareExecution, Run, and RunPrepared calls. Engine-local backend policies
+// bound admitted Run and RunPrepared calls and model generations where
+// configured, while different backend pools and unlimited backends continue
+// independently. An injected [LLMClient] or [ArtifactReader] can therefore
+// still receive concurrent calls and must be safe for that use.
 //
-// NewEngine copies in-memory profiles and backend definitions. Prepare and Run
-// copy request maps, slices, pointer values, and JSON-compatible extra
-// parameters before using them. Returned values and values passed to extension
-// interfaces are likewise isolated from engine state. Callers own those copies
-// and may mutate them after the call that supplied or returned them.
+// ConfigKit captures the selected configuration when NewEngine constructs an
+// Engine. Prepare, PrepareExecution, and Run copy request maps, slices, pointer
+// values, and JSON-compatible extra parameters before using them. InspectPrompt,
+// InspectRequirements, and InspectProfile return copied inspection values.
+// ListProfiles and ListPrompts return caller-owned identity slices.
+// Returned values and values passed to extension interfaces are likewise
+// isolated from engine state.
+// Callers own those copies and may mutate them after the call that supplied or
+// returned them. [CapacityError] values are caller-owned and may be mutated
+// without affecting engine state or another error. Immutable [GenerationError]
+// values are also caller-owned and do not retain shared engine state.
 //
 // # Security and sensitive data
 //
@@ -45,10 +56,20 @@
 // [GenerateResponse], [ExecutionTargetPresence], and the string value types
 // used by those values.
 //
-// Construction values, including [Config], [Backend], [RunRequest],
-// [ArtifactRef], [ExecutionTargetOverride], [Profile], and
-// [OpenAICompatibleProfileConfig], do not have stable JSON representations.
-// Direct API keys are nevertheless excluded from JSON for every public value.
+// Construction, inspection, handle, and error values, including [Config],
+// [ConfigurationSources], [ConfigurationInspection], [BackendConfig],
+// [ProfileConfig], [RunRequest],
+// [ArtifactRef], [ExecutionTargetOverride], [ProfileInspection],
+// [PromptIdentity], [PromptInputDefinition], [PromptInspection],
+// [RequirementInspectionRequest], [RequirementInspection], [PreparedExecution],
+// [CapacityError], [GenerationError], [RetryExhaustedError], and
+// [ExecutionEvent], do not have stable JSON
+// representations. Direct API keys are nevertheless excluded from JSON for
+// every public value.
+
+// Provider-derived [GenerationError] accessor values are untrusted and can
+// contain sensitive request or schema fragments. Applications must apply their
+// own disclosure policy before logging, displaying, or returning them.
 //
 // JSON timestamps use time.Time's RFC 3339 encoding and are omitted when zero.
 // PreparedRun and RunResult durations are encoded as integer milliseconds in

docs/adr/0001-configuration-boundary.md

@@ -0,0 +1,60 @@
+# 0001: Configuration Ownership At The ConfigKit Handoff
+
+## Status
+
+Accepted.
+
+## Date
+
+2026-09-30
+
+## Context
+
+PromptKit is adopting ConfigKit to standardize configuration semantics and
+simplify its loading machinery. Configuration includes prompts, profiles, and
+backend definitions contributed by maintained catalogs and consuming applications.
+ConfigKit supplies loading, composition, precedence, inheritance, and inspection
+capabilities. PromptKit supplies LLM-specific interpretation and execution.
+
+A clear handoff assigns each library responsibility for the configuration it
+owns and keeps validation aligned with the values an application actually uses.
+
+## Decision
+
+ConfigKit owns configuration processing up to the handoff of the final effective
+configuration. PromptKit takes ownership at that handoff and applies its domain
+interpretation and validation to the final configuration.
+
+The handoff includes effective presence/null information and resource context
+supplied by ConfigKit. PromptKit uses this information to interpret effective
+settings and resolve referenced resources.
+
+PromptKit passes through ConfigKit's inspection capabilities for explicit
+operator use. Configuration history and source explanations serve that
+inspection purpose; domain validation operates on the final effective values.
+
+This boundary applies uniformly to application configuration and maintained
+catalog configuration, including backend definitions.
+
+## Alternatives Considered
+
+- **Independent source or catalog preflight validation:** preserves checks on
+  individual contributions, but duplicates configuration processing and evaluates
+  values that may be replaced before handoff. Raw-source and intermediate-state
+  validation are outside PromptKit's scope, including equivalent test-only checks.
+- **History-based source auditing:** can identify overwritten contributions, but
+  couples domain enforcement to configuration history. Historical inspection is
+  an operator capability rather than a validation mechanism.
+
+## Consequences
+
+PromptKit's configuration logic and tests focus on the final effective values,
+their domain meaning, and resulting execution behavior. ConfigKit owns failures
+arising during its processing, which PromptKit preserves through its error
+boundary. Catalog maintainers own their repositories' publishing checks.
+
+The adoption work routes backend configuration through the same handoff and
+retires PromptKit's source-level and intermediate-configuration validation.
+Effective resource interpretation and explicit operator inspection remain
+supported. This concentrates configuration semantics in ConfigKit and reduces
+duplicated machinery and maintenance in PromptKit.

docs/consumers/pkg-promptkit.md

@@ -19,32 +19,313 @@ The following Go fragments are illustrative and omit surrounding package,
 import, and error-handling code. Use the maintained examples for complete
 programs.
 
+## Go Compatibility
+
+The `go` directive in the selected release's [module metadata](../../go.mod)
+declares its minimum supported consumer Go version. Maintainers build and
+validate releases with the locally selected Go patch release; the minimum
+does not simply track that patch version.
+
+Successful release validation on the selected patch is sufficient to presume
+support across its Go series, subject to the declared minimum and concrete
+evidence of incompatibility. Specific compatibility concerns are investigated
+as they arise. Minor releases may raise the minimum; patch releases preserve
+it. Release notes explain increases and any required consumer action. See the
+[release policy](../policy/release.md#consumer-minimum-and-release-toolchain).
+
 ## Construct An Engine
 
-Create an engine with
-[`NewEngine`](../../engine.go). A directory-backed setup supplies a prompt
-directory and may supply profile and schema directories:
+Create an engine with [`NewEngine`](../../engine.go). Select each directory
+explicitly and place its inputs in an ordered ConfigKit underlay. A group
+strictly composes its files; later groups can overlay earlier ones:
 
 ```go
+promptInputs, err := promptkit.DirectoryInputs("prompts")
+if err != nil { return err }
+profileInputs, err := promptkit.DirectoryInputs("profiles")
+if err != nil { return err }
 engine, err := promptkit.NewEngine(promptkit.Config{
-	PromptDir:  "prompts",
-	ProfileDir: "profiles",
-	SchemaDir:  "schemas",
+	Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{
+		{Inputs: promptInputs},
+		{Inputs: profileInputs},
+	}},
+	SchemaDir: "schemas",
 })
 ```
 
-Options support single-file or `fs.FS` sources, in-memory profiles,
-engine-scoped backends, and injected artifact or model clients. Consult the
-[constructor and option GoDoc](../../engine.go) for composition, precedence,
-validation, and default transport behavior. Source discovery, format
-validation, and profile precedence are defined by the
-[framework format reference](../formats.md).
+For a single highest-priority document, set `Sources.Main` to a
+`*configkit.Input`; omit it for ordinary layered construction. A single file
+can use `configkit.File` directly. `ProfileSource`, `BackendSource`, and raw
+`configkit.Mapping` supply programmatic definitions. Schema and runtime
+collaborator options remain available. Construction validates the final
+effective configuration and captures referenced prompt content; construct a
+new engine to observe configuration edits. See the [constructor GoDoc](../../engine.go)
+for the exact public contract and the [framework format reference](../formats.md)
+for definition fields.
+
+## Supply Embedded Application Defaults
+
+Use `FilesystemInputs` for embedded defaults and `DirectoryInputs` for an
+explicitly chosen operator directory. A later programmatic group can override
+both. No directory name has built-in precedence:
+
+```go
+defaults, err := promptkit.FilesystemInputs("embedded defaults", applicationFS, "config")
+if err != nil { return err }
+operator, err := promptkit.DirectoryInputs(operatorDirectory)
+if err != nil { return err }
+profile := promptkit.ProfileSource("application choices", map[string]promptkit.ProfileConfig{
+	"summary": {Model: &selectedModel},
+})
+engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+	Underlays: []configkit.Underlay{
+		{Inputs: defaults},
+		{Inputs: operator},
+		{Inputs: []configkit.Input{{Source: profile}}},
+	},
+}})
+```
+
+Only add the operator group when the application selects that directory; an
+explicitly selected missing or unreadable root is an error. Each helper finds
+`.yaml` and `.yml` files recursively and disables imports by default. The
+caller may enable imports on individual inputs. A nil `Sources.Main` is an
+empty main source, not another directory lookup.
+
+For a programmatic document with explicit null or a prompt definition, use a
+labeled mapping as an input. A mapping owns a snapshot of supported values:
+
+```go
+document := configkit.Mapping("application configuration", map[string]any{
+	"profiles": map[string]any{
+		"local-summary": map[string]any{
+			"backend": "local", "model": "example-model",
+			"temperature": nil, // Suppress an inherited provider control.
+		},
+	},
+})
+input := configkit.Input{Source: document}
+// Add input to a later underlay, or set Sources.Main = &input.
+```
+
+`ProfileSource` and `BackendSource` are smaller typed conveniences when null
+is not needed. A caller underlay can overlay catalog backend fields just as it
+can overlay profiles; there is no registration-only backend namespace.
+
+When configuration or message files change, build and validate a new engine,
+then switch the application to that engine for subsequent work. Existing
+engines and prepared handles retain their captured definitions; schema and
+request artifact resources still follow their runtime lifecycle. Inspection
+does not watch or reload sources automatically.
+
+## Discover Available Profiles And Prompts
+
+Use [`Engine.ListProfiles`](../../listing.go) and
+[`Engine.ListPrompts`](../../listing.go) to discover exact identities from the
+engine's final captured configuration, then inspect a selected prompt:
+
+```go
+profiles, err := engine.ListProfiles(ctx)
+if err != nil {
+	return err
+}
+if len(profiles) > 0 {
+	_, err := engine.InspectProfile(ctx, profiles[0])
+	if err != nil {
+		return err
+	}
+}
+prompts, err := engine.ListPrompts(ctx)
+if err != nil {
+	return err
+}
+if len(prompts) > 0 {
+	selected := prompts[0]
+	inspection, err := engine.InspectPrompt(ctx, selected.ID, selected.Version)
+	if err != nil {
+		return err
+	}
+	_ = inspection
+}
+```
+
+These lists describe effective membership, not ConfigKit contribution history,
+and do not promise that a profile is ready to execute. Use inspection for
+details and preparation to resolve request resources. The linked method GoDoc
+owns ordering, ownership, concurrency, and error behavior; identifiers may be
+sensitive application data.
+
+## Inspect The Captured Configuration
+
+Use [`Engine.InspectConfiguration`](../../configuration_inspection.go) for an
+operator-selected report of effective configuration, source descriptions,
+diagnostics, and path history. For a literal ID containing a slash, construct
+the semantic path with `configkit.Keys`:
+
+```go
+view, err := engine.InspectConfiguration()
+if err != nil { return err }
+path := configkit.Keys("profiles", "team/summary", "model")
+explanation, found := view.Explain(path)
+if found && explanation.Effective.Redacted {
+	// Apply the application's disclosure policy before sharing the report.
+}
+```
+
+`WriteEffective` and `WriteExplanation` produce ConfigKit-redacted reports;
+the application decides whether those reports may be disclosed. Credential-
+shaped provider keys are redacted, but arbitrary prompt and provider content
+is not necessarily safe to share. Inspection uses the engine's captured result
+and does not reopen sources or resolve credentials. `InspectProfile` instead
+shows backend-completed execution settings for one profile, and `Prepare` or
+`PreparedExecution.Details` shows request-specific rendered messages, inputs,
+and validation resources. None of those views is a substitute for the others.
+
+To review changes between two independently captured configurations, compare
+their inspection views:
+
+```go
+earlier, err := earlierEngine.InspectConfiguration()
+if err != nil {
+	return err
+}
+later, err := candidateEngine.InspectConfiguration()
+if err != nil {
+	return err
+}
+differences, err := earlier.Diff(later)
+if err != nil {
+	return err
+}
+for _, difference := range differences {
+	if difference.HiddenChanged {
+		// Report a protected change without inferring its value.
+	}
+	if difference.Left.Redacted || difference.Right.Redacted {
+		// Apply the application's disclosure policy before presenting snapshots.
+	}
+}
+```
+
+See the [`ConfigurationInspection.Diff` GoDoc](../../configuration_inspection.go)
+and ConfigKit's [v0.7.0 effective-configuration diff guidance](https://gitea.maximumdirect.net/eric/ConfigKit/src/tag/v0.7.0/docs/config.md#effective-configuration-diffs)
+for the exact comparison contract. The operation compares only frozen effective
+configuration values: it does not reread external files or compare their
+contents, resolved credential contents, later execution preparation, or
+request-specific overrides. An empty diff is not proof of equivalent execution.
+Records can contain ordinary application data, so use the application's
+disclosure policy when presenting them.
+
+## Inspect A Prompt Before Preparation
+
+Use [`Engine.InspectPrompt`](../../engine.go) to check one configured prompt's
+declared inputs and output workflow without creating placeholder inputs or
+resolving a profile:
+
+```go
+inspection, err := engine.InspectPrompt(ctx, "meeting.summary", "")
+if err != nil {
+	return err
+}
+
+for _, input := range inspection.Inputs {
+	// Compare the declared input with application configuration.
+}
+```
+
+Use this configuration-time boundary when the application needs only the
+declared prompt interface. Use `InspectProfile` separately when it must also
+check a configured profile. Use `Prepare` when it needs inputs, schemas, or
+rendered messages, and use prepared execution when that work must remain tied
+to later execution. The method's [GoDoc](../../engine.go) owns exact fields,
+hash, ownership, and error semantics.
+
+## Inspect Requirements Before Preparation
+
+Use [`Engine.InspectRequirements`](../../requirements.go) when the application
+needs the prompt's recognizable input and caller-variable dependencies for a
+particular context before it has input artifacts. Unlike `InspectPrompt`, it
+also walks captured template syntax and selects a profile identity; unlike
+`Prepare`, it does not load artifacts or render messages.
+
+```go
+requirements, err := engine.InspectRequirements(ctx, promptkit.RequirementInspectionRequest{
+	PromptID: "meeting.summary",
+	Vars:     map[string]string{"audience": "project leads"},
+})
+if err != nil {
+	return err
+}
+
+for _, input := range requirements.Inputs {
+	if input.Kind == promptkit.RequirementRequired {
+		// Plan to bind this input before preparation.
+	}
+	if input.Kind == promptkit.RequirementOptional && input.Explanation != "" {
+		// A recognizable reference may occur only on a known skipped path.
+	}
+}
+for _, variable := range requirements.Variables {
+	if variable.Kind == promptkit.RequirementRequired && !variable.Supplied {
+		// The key is required even though this inspection did not receive it.
+	}
+}
+if !requirements.Complete {
+	// Preserve the partial inventory and account for its discovery issues.
+}
+```
+
+Optional entries can include recognizable inputs and variables found only in
+inactive branches. A missing required variable is reported separately from
+discovery completeness: missing a value used only for direct text substitution
+does not by itself make the inventory incomplete. Conversely, an unresolved
+dependency-bearing branch can make discovery incomplete while retaining known
+conditional entries. Do not treat an empty partial inventory as proof that no
+input bindings are needed. See the method's [GoDoc](../../requirements.go) for
+the exact classifications, supported syntax, ownership, and error contract.
+
+## Validate A Concrete Request
+
+Use [`Engine.ValidateRequest`](../../engine.go) when the application has a
+complete `RunRequest` and needs to know whether that concrete invocation passes
+preparation:
+
+```go
+request := promptkit.RunRequest{
+	PromptID: "meeting.summary",
+	Vars:     map[string]string{"audience": "project leads"},
+	Inputs: map[string]promptkit.ArtifactRef{
+		"note": promptkit.Inline("Synthetic meeting notes"),
+	},
+}
+if err := engine.ValidateRequest(ctx, request); err != nil {
+	return err
+}
+```
+
+Unlike `InspectRequirements`, this checks supplied values, the selected template
+branches, and request bindings through the existing preparation path. It checks
+only this invocation, does not aggregate errors, and loads every supplied input
+according to ordinary preparation behavior, even if a branch does not use it.
+It makes no model call and returns no prepared snapshot, but artifact and schema
+resource I/O and its time and memory costs remain. Success does not certify
+provider acceptance or later execution success. See the method's GoDoc for its
+exact error and context behavior.
+
+Scriptorium can map the returned error to its own diagnostics and exit status.
+Its `validate` and `run` commands should construct equivalent requests for
+equivalent caller inputs so their preparation behavior agrees. Separate
+subprocess calls prepare independently; an opaque `PrepareExecution` handle
+cannot be transferred from `validate` to `run`. Use `PrepareExecution` when one
+process and engine must execute the exact prepared snapshot.
 
 ## Prepare Without Model Execution
 
 [`Engine.Prepare`](../../engine.go) resolves the selected prompt and profile,
 loads inputs and any structured-output schema, and renders messages without
-calling a model client:
+calling a model client. Choose it when the prepared value is the final
+inspection or persistence result and no later execution must be tied to that
+exact snapshot:
 
 ```go
 prepared, err := engine.Prepare(ctx, promptkit.RunRequest{
@@ -58,15 +339,62 @@ prepared, err := engine.Prepare(ctx, promptkit.RunRequest{
 The maintained
 [offline preparation example](../../examples/go-library/prepare/main.go)
 shows a complete runnable setup with a prompt file, in-memory profile, and
-inline input. Exact request requirements and prepared-result fields belong to
-the [`RunRequest` and `PreparedRun` GoDoc](../../types.go).
+inline input. It demonstrates validation and preparation with the same request
+to make the two operations concrete; this performs preparation twice and is
+illustrative, not a requirement to validate before `Prepare` or `Run`. Exact
+request requirements and prepared-result fields belong to the
+[`RunRequest` and `PreparedRun` GoDoc](../../types.go).
+
+## Prepare Now And Execute The Same Snapshot Later
+
+Use [`Engine.PrepareExecution`](../../engine.go) when an application must
+inspect or persist preflight details before deciding whether to start model
+work, while ensuring that later execution uses those exact rendered messages,
+inputs, target settings, and validation resources:
+
+```go
+preparedExecution, err := engine.PrepareExecution(ctx, promptkit.RunRequest{
+	PromptID: "meeting.summary",
+	Inputs: map[string]promptkit.ArtifactRef{
+		"note": promptkit.Inline("Synthetic meeting notes"),
+	},
+})
+if err != nil {
+	return err
+}
+defer preparedExecution.Discard()
+
+details := preparedExecution.Details()
+// Inspect or persist an application-selected safe subset of details.
+
+result, err := engine.RunPrepared(ctx, preparedExecution)
+```
+
+Preparation does not call the model or reserve backend capacity.
+`RunPrepared` executes from the retained snapshot rather than reloading
+consumer sources. The handle is opaque in-process state, while `Details`
+contains rendered content and remains subject to the application's data
+handling policy. The
+[`PreparedExecution` and method GoDoc](../../prepared_execution.go) and
+[engine operation GoDoc](../../engine.go) own exact lifecycle, engine-binding,
+credential, cancellation, timing, and error semantics.
+
+If the request supplied `Observer`, successful preparation retains that
+function for `RunPrepared`; preparation itself emits nothing. Observation
+starts only when the owning engine successfully claims the handle. Reused,
+discarded, foreign, and invalid handles emit nothing. `Discard` clears the
+retained callback without invoking it. The function's closure is retained by
+identity, so synchronize any shared state it uses. Prepared start events can
+include identities already resolved in the frozen snapshot.
 
 ## Execute And Validate
 
 [`Engine.Run`](../../engine.go) performs the same preparation, invokes the
 configured model client, classifies the generated artifact, and validates the
-content. A completed content check may return `ValidationFailed` in the result;
-an operational inability to validate returns an error.
+content in one call. Choose it when the application does not need a preflight
+boundary tied to the eventual execution. A completed content check may return
+`ValidationFailed` in the result; an operational inability to validate returns
+an error.
 
 The maintained
 [offline execution example](../../examples/go-library/run/main.go) injects a
@@ -81,6 +409,98 @@ semantics. The
 [OpenAI-compatible integration contract](../integrations/openai-compatible-chat.md)
 owns the built-in client's outbound HTTP behavior.
 
+### Observe An Ordinary Run
+
+Set `RunRequest.Observer` to receive synchronous metadata events from `Run`:
+
+```go
+request.Observer = func(event promptkit.ExecutionEvent) {
+	// Translate selected metadata into an application-owned record.
+}
+result, err := engine.Run(ctx, request)
+```
+
+An ordinary run emits `started` before request conversion, followed by any
+attempt and retry-delay completions, then one `finished` event on a normal
+return. `Prepare` and `PrepareExecution` emit no events. A start event has no
+resolved prompt, profile, or model; later events gain identities only as
+selection succeeds. The terminal event has dispatched-attempt, transient-retry,
+repair-call, duration, and presence-aware usage totals. Its usage is an
+aggregate, not an additional billable attempt. A known zero differs from an
+unknown count, and incomplete totals must not be treated as complete cost
+accounting. A failed operation still returns no partial `RunResult`.
+An injected client that leaves `UsagePresence` nil can report known nonzero
+counts, but its zero counts remain unknown to observations; supply explicit
+presence when measured zeros matter. Existing `RunResult.Usage` sums returned
+values independently of observation presence.
+
+Callbacks run synchronously, in order for one operation, and can overlap for
+different operations. Synchronize shared callback state and avoid blocking on
+work that needs the same run's admission slot. Callback panics propagate;
+there is no durable event history or promise of a terminal event after panic.
+Events omit content and error objects, but selected identifiers may still be
+sensitive. See the [observation GoDoc](../../observations.go) for exact fields.
+The [offline proposal and assessment example](../../examples/go-library/observe/main.go)
+shows application-owned event collection and a separate semantic acceptance
+decision using the built-in HTTP path without network access.
+
+### Repair A Structured Result
+
+Set a small additional-call budget when a structurally invalid result can be
+corrected automatically:
+
+```go
+request.Validation = &promptkit.OutputContract{
+	Format:         promptkit.FormatJSON,
+	ValidationMode: promptkit.ValidationJSONSchema,
+	SchemaPath:     "events.schema.json",
+	RepairAttempts: 1,
+}
+```
+
+Each repair attempt is another model generation, so it can increase latency and
+usage; `RunResult.Usage` is cumulative and `Validation.RepairAttempts` reports
+repair generations rather than transient HTTP attempts. Exhaustion still
+returns the final failed validation result. `basic` validation can also repair
+an empty candidate, but structural
+validity is not evidence of factual or domain correctness. See the
+[output-contract format reference](../formats.md#output-contract) and
+[`OutputContract` GoDoc](../../types.go) for the exact budget and eligibility
+rules.
+
+The built-in client also retries eligible temporary HTTP statuses. By default,
+each generation can make two additional attempts; a repair generation receives
+its own budget. Set `Config.TransientRetries` to a pointer to zero for the
+previous single-attempt behavior, or set `RunRequest.TransientRetries` to
+override one execution. A custom HTTP transport still uses this policy;
+injected `LLMClient` implementations retain their own retry policy. Application
+semantic retries are separate runs and can multiply provider traffic further.
+The caller context covers capacity and retry waiting as well as attempts;
+without a deadline a valid provider delay can be long. Promptkit does not
+infer exactly-once or complete cost accounting from a retry outcome.
+See the [provider integration contract](../integrations/openai-compatible-chat.md#transient-status-retries)
+for eligible statuses, waits, timeouts, and provider timing.
+
+### Append Already-Rendered Messages
+
+An application can include an earlier assistant response and its own corrective
+instruction in a fresh request without changing the configured prompt:
+
+```go
+request.AppendedMessages = []promptkit.RenderedMessage{
+	{Role: promptkit.RoleAssistant, Content: previousResponse},
+	{Role: promptkit.RoleUser, Content: correction},
+}
+result, err := engine.Run(ctx, request)
+```
+
+These messages are already rendered: Promptkit does not template or resolve
+files in them, and they can contain sensitive model output or application
+feedback. Every `Run` call uses the engine's captured configuration; the
+application owns any semantic retry budget. When a
+pre-execution equality check is required, use `PrepareExecution` and compare
+its opaque rendered-prompt hash before invoking `RunPrepared`.
+
 ## Inputs, Profiles, And Overrides
 
 Use `File`, `Inline`, or `InlineWithURI` to construct request inputs. A request
@@ -90,12 +510,146 @@ replace execution settings or the complete output contract.
 The [public value GoDoc](../../types.go) defines nil, empty, zero, replacement,
 copy, and credential behavior. The
 [framework format reference](../formats.md) defines how those request values
-interact with prompt definitions, file-backed profiles, built-ins, schemas,
-and framework defaults.
+interact with prompt definitions, layered profiles, built-ins, schemas, and
+framework defaults.
 
-For programmatic profiles,
-[`OpenAICompatibleProfile`](../../profiles.go) converts ordinary
-OpenAI-compatible settings into a value accepted by `WithProfiles`.
+For programmatic profiles, use [`ProfileSource`](../../profiles.go) with keyed
+`ProfileConfig` values. Nil fields are omitted; pointers preserve explicit
+zero, false, and blank values. Raw `configkit.Mapping` expresses explicit null.
+
+### Migrate From The Previous Configuration API
+
+Replace `Config.PromptDir` and `Config.ProfileDir` with selected ConfigKit
+inputs in `Config.Sources`. The removed `WithPromptFS`, `WithPromptFile`,
+`WithProfileFS`, `WithProfileFile`, `WithFallbackProfileFS`, `WithProfiles`,
+and `WithBackend` options have no compatibility wrappers. Use
+`DirectoryInputs` or `FilesystemInputs` for a strict-composition file group,
+ConfigKit file inputs for individual files, and `ProfileSource`,
+`BackendSource`, or `configkit.Mapping` for programmatic values. Backend IDs
+are keys in configuration, not registrations. Choose caller underlays in
+fallback-to-operator-to-programmatic order; use `Sources.Main` for one final
+highest-priority input.
+
+Move embedded `id` and `version` fields into keyed `profiles` and `prompts`
+maps; convert `base_profile` to ConfigKit `extends`. IDs and references are
+exact literal keys, without trimming. A supplied zero, false, or empty string
+is now distinguishable from omission; null can clear or suppress an inherited
+field according to the [field table](../formats.md#defaults-and-overrides).
+Maps merge recursively across inheritance and layers. `{}` does not clear an
+inherited map: use null to clear the whole map. At request time, a literal
+`extra_params["extends"]` is ordinary provider data, not an inheritance
+instruction; a nonempty request ExtraParams map replaces configured params.
+Changing a model or backend does not clear unrelated explicit fields, though
+backend-derived defaults are applied after inheritance to the final backend.
+
+Invalid final definitions, including unused ones, now fail `NewEngine`; only
+request-dependent failures remain at preparation or execution. Prompt text
+and static templates are captured at construction, not automatically reloaded
+between runs. Replace the engine to pick up changed definitions or message
+files. The [format reference](../formats.md) owns exact field and presence
+rules; the maintained examples show complete runnable workflows.
+
+### Choose A Profile Ownership Model
+
+Choose the profile source that matches the application's continuity needs:
+
+- Select `promptkit-openrouter-standard` or `promptkit-openrouter-auto` when
+  the maintained literal name is the desired consumer contract. Their current
+  selections are in the [format reference](../formats.md#built-in-profile-catalog).
+- Select an external catalog profile directly when the application accepts that
+  profile's catalog volatility. The selected external inventories are linked
+  from the [format reference](../formats.md#external-profile-inventories).
+- Define a complete application-owned profile with an explicit backend or
+  endpoint, model, and settings when the application owns those choices.
+
+An application profile that inherits an external base follows later changes to
+that base and fails if the base disappears; it is not a frozen copy. Promptkit
+tests the catalog versions selected in its module graph. Independently advancing
+a catalog dependency can remove an inherited base and produce an untested
+combination. Pinning the resolved module graph preserves the embedded data for
+a build, but cannot preserve provider availability. Running binaries neither
+download catalog updates nor automatically replace withdrawn models. See the
+[release policy](../policy/release.md#catalog-profile-compatibility) for the exact
+cross-version guarantees.
+
+### Alias A Built-In Profile
+
+Give an application-owned profile ID a built-in base when prompts should select
+the application ID while inheriting the built-in target. The child can override
+only the setting it owns:
+
+```go
+profile := promptkit.ProfileSource("application profiles", map[string]promptkit.ProfileConfig{
+	"weather-light": {
+		Extends: &standardID,
+		ReasoningEffort: &high,
+	},
+})
+```
+
+Place `profile` in a ConfigKit underlay, with `standardID` set to
+`"promptkit-openrouter-standard"` and `high` to `"high"`. Select
+`weather-light` in a prompt or `RunRequest.ProfileID`; it remains the
+reported selected profile. Definition IDs and inheritance references are exact
+keys. If the base is an external profile, apply the volatility guidance
+above when deciding whether to use an application-owned complete definition
+instead.
+
+### Use The Rakestrawhome Built-In Profile
+
+Set `RAKESTRAWHOME_INFERENCE_API_KEY` in the application environment, then
+select `rakestrawhome-gemma-4-31b` as an ordinary profile ID. For example, a
+prepared result identifies the selected built-in through
+`BackendRakestrawHome`:
+
+```go
+prepared, err := engine.Prepare(ctx, promptkit.RunRequest{
+	PromptID:  "meeting.summary",
+	ProfileID: "rakestrawhome-gemma-4-31b",
+	Inputs:    inputs,
+})
+if err != nil {
+	return err
+}
+if prepared.SelectedBackendID != promptkit.BackendRakestrawHome {
+	return fmt.Errorf("unexpected backend %q", prepared.SelectedBackendID)
+}
+```
+
+The maintained catalog supplies this backend, and callers may overlay its
+effective configuration through a later ConfigKit source. Direct request credentials
+and runtime endpoint overrides remain supported under their ordinary GoDoc and
+format contracts.
+
+### Inspect A Profile Before Prompt Work
+
+Use [`Engine.InspectProfile`](../../engine.go) to validate one configured
+profile without constructing a synthetic prompt or placeholder inputs. It
+resolves the profile's effective target but does not prepare or execute a
+prompt:
+
+```go
+inspection, err := engine.InspectProfile(ctx, profileID)
+if err != nil {
+	return err
+}
+
+target := inspection.EffectiveModelParams
+if target.APIKeyEnv != "" {
+	// This is a configured optional environment lookup source.
+} else if inspection.APIKeyRequired {
+	// Arrange a direct credential before later execution.
+}
+```
+
+Use this configuration-time boundary when only the profile and its target need
+checking. Use `Prepare` when the application also needs prompt, input, schema,
+or rendering work; use prepared execution when that work must remain tied to a
+later execution. A reported `APIKeyEnv` is a configured optional source, while
+`APIKeyRequired` is the explicit local requirement. The
+[credential format reference](../formats.md#credentials) and the method's
+[GoDoc](../../engine.go) own the exact precedence, timing, result, and error
+contracts.
 
 ### Set A Per-Run Session And Reasoning
 
@@ -124,38 +678,70 @@ providers. The
 [`RunRequest` and `ExecutionTargetOverride` GoDoc](../../types.go) owns the
 exact normalization, precedence, error, copying, and exposure contract.
 
-### Register A Custom Backend
+### Configure A Local OpenAI-Compatible Endpoint
 
-Register a reusable OpenAI-compatible connection once, then select it from a
-profile. This local backend limits model generation to two simultaneous calls;
-because `QueueCapacity` is omitted, the engine admits up to 1024 additional
-calls waiting behind them:
+Choose the smallest configuration that fits how the endpoint will be reused.
+
+#### Use An Endpoint-Only Profile
+
+Put the endpoint directly on a programmatic profile when only that profile needs
+it and shared backend identity or capacity policy is unnecessary:
 
 ```go
-engine, err := promptkit.NewEngine(promptkit.Config{
-	PromptDir: "prompts",
-},
-	promptkit.WithBackend(promptkit.Backend{
-		ID:        "local",
-		Endpoint:  "http://localhost:8000/v1",
-		APIKeyEnv: "LOCAL_LLM_API_KEY",
-		ConcurrencyLimit: 2,
-	}),
-	promptkit.WithProfiles(promptkit.Profile{
-		ID:        "local-summary",
-		BackendID: "local",
-		Model:     "example-model",
-	}),
-)
+endpoint, model := "http://localhost:8000/v1", "example-model"
+profile := promptkit.ProfileSource("local profile", map[string]promptkit.ProfileConfig{
+	"local-summary": {Endpoint: &endpoint, Model: &model},
+})
+// Add configkit.Input{Source: profile} to a caller underlay.
 ```
 
-Registrations belong to one engine and custom IDs cannot replace built-ins.
-The [`Backend` and `WithBackend` GoDoc](../../backends.go) defines validation,
-copying, uniqueness, exact concurrency-field semantics, and request-default
-behavior.
+Endpoint-only profiles have an empty backend ID and remain unrestricted by
+backend capacity policy.
 
-Both file-backed and in-memory profiles select a registration through
-`backend` or `Profile.BackendID`. Profile and request endpoint overrides retain
+#### Use The Conventional Local Backend
+
+Use `LocalBackend` when profiles should share the conventional `local`
+identity, endpoint, and concurrency limit:
+
+```go
+backendID, model := promptkit.BackendLocal, "example-model"
+backend := promptkit.BackendSource("local backend", map[string]promptkit.BackendConfig{
+	backendID: promptkit.LocalBackend("http://localhost:8000/v1", 2),
+})
+profile := promptkit.ProfileSource("local profile", map[string]promptkit.ProfileConfig{
+	"local-summary": {Backend: &backendID, Model: &model},
+})
+// Add both sources as inputs in a caller underlay.
+```
+
+The helper does not choose an ID or read environment variables. A positive
+concurrency limit with omitted queue capacity selects the default waiting
+capacity of 1024.
+
+#### Configure A Complete Backend
+
+Use a keyed `BackendConfig` value for authentication, extra request parameters, an
+explicit queue capacity, a custom ID, or multiple local endpoints:
+
+```go
+endpoint, keyName, limit, noWaiting := "http://gpu-host:8000/v1", "LOCAL_GPU_API_KEY", 2, 0
+backend := promptkit.BackendSource("GPU backend", map[string]promptkit.BackendConfig{
+	"local-gpu": {
+		Endpoint: &endpoint, APIKeyEnv: &keyName,
+		ConcurrencyLimit: &limit, QueueCapacity: &noWaiting,
+		ExtraParams: map[string]any{"provider_option": "enabled"},
+	},
+})
+```
+
+Place `backend` in a caller underlay and select `"local-gpu"` from a
+`ProfileConfig.Backend` pointer or a keyed file profile. All backend
+contributions use the same ConfigKit layering rules, including overlays of
+maintained definitions. The [`BackendConfig` and `LocalBackend` GoDoc](../../backends.go)
+defines exact omission and construction behavior.
+
+Both file-backed and programmatic profiles select a backend through
+`backend` or `ProfileConfig.Backend`. Profile and request endpoint overrides retain
 that routing and capacity identity. `PreparedRun.SelectedBackendID`,
 `RunResult.SelectedBackendID`, and the effective `ExecutionTarget.BackendID`
 expose it to consumers and injected model clients. Endpoint-only profiles
@@ -163,7 +749,7 @@ remain supported and expose an empty backend ID.
 
 ### Limit Backend Concurrency
 
-Set `Backend.ConcurrencyLimit` when a backend needs protection from too many
+Set `BackendConfig.ConcurrencyLimit` when a backend needs protection from too many
 simultaneous model calls. Leaving `QueueCapacity` nil, as in the local-backend
 example above, selects the default waiting capacity of 1024.
 
@@ -172,24 +758,23 @@ zero:
 
 ```go
 noWaiting := 0
-backend := promptkit.Backend{
-	ID:               "local",
-	Endpoint:         "http://localhost:8000/v1",
-	ConcurrencyLimit: 2,
-	QueueCapacity:    &noWaiting,
+backend := promptkit.BackendConfig{
+	Endpoint: &endpoint,
+	ConcurrencyLimit: &limit,
+	QueueCapacity: &noWaiting,
 }
 ```
 
-The pointer distinguishes an explicit zero from omission. Keep using keyed
-`Backend` literals so additive configuration fields remain source-compatible.
+The pointer distinguishes an explicit zero from omission. Use keyed
+`BackendConfig` literals inside `BackendSource` for programmatic definitions.
 Capacity belongs to one engine and the selected backend ID; endpoint-only
 profiles and custom backends without a configured limit remain unrestricted.
 Exact validation, defaulting, ownership, and concurrency semantics belong to
-the [`Backend` GoDoc](../../backends.go).
+the [`BackendConfig` GoDoc](../../backends.go).
 
 ## Credentials
 
-File-backed profiles name an environment variable; in-memory profiles can
+File-backed and programmatic profiles can name an environment variable or
 require a direct request key. Direct keys are request-scoped and are excluded
 from supported JSON values and the package's `String` and `GoString`
 summaries. The exact precedence and redaction guarantees belong to
@@ -223,12 +808,13 @@ the application policy that motivated the injection.
 
 Use `errors.Is` with the
 [public error sentinels and operation GoDoc](../../engine.go). The declarations
-distinguish invalid construction, invalid requests, absent sources,
-source-loading failures, collaborator failures, and operational validation
+distinguish invalid construction, invalid requests, absent selections,
+collaborator failures, and operational validation
 failures. Specific request conditions may also match the broader
 `ErrInvalidRequest`, and injected collaborator identities are preserved where
-documented. Invalid or duplicate backend registrations match
-`ErrInvalidConfig`; selecting an unknown backend matches `ErrProfileLoad`.
+documented. Invalid effective definitions, including unknown backend
+references, match `ErrInvalidConfig` during construction. ConfigKit load
+errors remain available through `errors.As`.
 
 When a limited backend has admitted all active and waiting calls, handle
 `ErrCapacityExceeded` separately from request errors and provider failures:
@@ -236,6 +822,11 @@ When a limited backend has admitted all active and waiting calls, handle
 ```go
 result, err := engine.Run(ctx, request)
 if errors.Is(err, promptkit.ErrCapacityExceeded) {
+	var capacityErr *promptkit.CapacityError
+	if errors.As(err, &capacityErr) {
+		// Record capacityErr.BackendID using application-owned diagnostics.
+	}
+
 	// Apply application policy: shed work, report overload, or retry later.
 }
 ```
@@ -243,8 +834,41 @@ if errors.Is(err, promptkit.ErrCapacityExceeded) {
 A rejected call returns no partial result and does not invoke the model
 client. Promptkit does not prescribe retries or map this error to an HTTP
 status; those choices remain with the consuming application. The
-[`Engine.Run` and error GoDoc](../../engine.go) owns exact error and
-cancellation identities.
+[`CapacityError` GoDoc](../../capacity_error.go) owns the exact typed-error
+contract, while the [`Engine.Run` and error GoDoc](../../engine.go) owns broad
+error and cancellation identities.
+
+For a non-2xx response from the built-in OpenAI-compatible client, inspect the
+status and deliberately selected provider diagnostic when useful:
+
+```go
+var generationErr *promptkit.GenerationError
+if errors.As(err, &generationErr) {
+	status := generationErr.StatusCode()
+	message := generationErr.ProviderMessage()
+	_, _ = status, message // Apply application presentation policy.
+}
+```
+
+All provider fields are untrusted and can contain sensitive request or schema
+fragments. Do not log, display, or return them without an application-specific
+disclosure policy. Promptkit does not assign presentation behavior.
+The [`GenerationError` GoDoc](../../generation_error.go) owns the exact typed
+error contract.
+
+Generation failures also match `ErrTransport` when the built-in HTTP request
+fails, or `ErrMalformedResponse` when a successful provider response cannot be
+decoded or an injected client returns a nil successful response. Both retain
+`ErrLLMGenerate`; transport failures retain their original causes for
+`errors.Is` and `errors.As`. Caller cancellation and deadlines retain their
+standard context identities. Arbitrary injected-client errors retain their own
+identities without Promptkit guessing a failure category from their text.
+An eligible built-in status that consumes a positive retry budget additionally
+matches `ErrRetriesExhausted`; use `errors.As` to read
+`RetryExhaustedError.Attempts()` and the final `GenerationError` status. With
+explicit zero retries, the first eligible status is an ordinary generation
+error, not exhaustion. Completed structural rejection remains a normal result
+with failed validation, while an operational error returns no partial result.
 
 ## Application Boundary
 

docs/development.md

@@ -1,8 +1,7 @@
 # Development
 
-This is the contributor entry point for Promptkit, a reusable Go library. All
-contributors must read the
-[architecture policy](policy/architecture.md) before making changes.
+This is the contributor entry point for Promptkit, a reusable Go library.
+Use the Policy Reading Guide below to select policies applicable to the task.
 
 ## Initial Orientation
 
@@ -15,17 +14,27 @@ Before starting work:
 4. keep documentation limited to implemented behavior unless an accepted
    decision or temporary roadmap explicitly owns future work.
 
-Start with:
+## Policy Reading Guide
 
-- the [architecture policy](policy/architecture.md) for library boundaries,
-  dependency direction, invariants, and non-goals;
-- the [internal component overview](internal/overview.md) for the current
-  package and component inventory;
-- the [documentation policy](policy/documentation.md) before changing
-  documentation;
-- the [testing policy](policy/testing.md) before adding, rewriting, or deleting
-  tests; and
-- the [release procedure](release.md) for version and publication work.
+| Policy | Read before… |
+| --- | --- |
+| [Policy set provenance](policy/meta.md) | Adopting policies, checking upstream revisions, or updating the adopted policy set. |
+| [Agents](policy/agents.md) | Recommending, planning, or implementing repository changes. |
+| [Architecture](policy/architecture.md) | Planning or changing library code or its structure. |
+| [Testing](policy/testing.md) | Selecting validation for code changes, or adding, changing, removing, or evaluating tests. |
+| [Documentation](policy/documentation.md) | Writing, revising, or reorganizing documentation. |
+| [OSS](policy/oss.md) | Adding or updating dependencies or incorporated third-party material, or changing licensing or attribution. |
+| [Release](policy/release.md) | Preparing or publishing a release, changing release procedures, or making versioning and compatibility decisions. |
+
+- Read applicable policies before making decisions they govern; multiple policies
+  may apply.
+- Select policies using the task triggers above. Follow cross-references to
+  relevant sections when needed to resolve a concrete question.
+- Reuse policies already available in context; reread when changed or no longer
+  available.
+- Expand reading when the task's scope expands.
+- When revising a policy itself, read that policy and any provisions elsewhere
+  affected by the revision.
 
 ## Task-Specific Reading Guide
 
@@ -34,108 +43,61 @@ Start with:
 | Root public API | The [architecture policy](policy/architecture.md), [consumer guide](consumers/pkg-promptkit.md), [testing policy](policy/testing.md), and existing GoDoc. |
 | Prompt, profile, or schema formats | The [framework format reference](formats.md), owning parser or validator package, and [documentation policy](policy/documentation.md). |
 | Source loading or validation | The [framework format reference](formats.md), [internal source document](internal/sources.md), and owning package tests. |
+| Maintained external catalogs or built-ins | The [internal source document](internal/sources.md), [framework format reference](formats.md), [testing policy](policy/testing.md), both catalog module repositories, each catalog's release procedure, and the selected module versions in `go.mod`. |
 | Model-client behavior | The [OpenAI-compatible integration contract](integrations/openai-compatible-chat.md), [internal model-client document](internal/llm.md), and owning package tests. |
 | Internal package implementation | The [architecture policy](policy/architecture.md), [internal component overview](internal/overview.md), and focused internal document listed for that package. |
 | Tests or test fixtures | The [testing policy](policy/testing.md), owning package, and focused internal document listed by the component overview. |
-| Maintained example | The [example](../examples/go-library/prepare/main.go), [consumer guide](consumers/pkg-promptkit.md), [framework format reference](formats.md), and [documentation policy](policy/documentation.md). |
+| Maintained example | The [preparation](../examples/go-library/prepare/main.go), [execution](../examples/go-library/run/main.go), and [observation](../examples/go-library/observe/main.go) examples, [consumer guide](consumers/pkg-promptkit.md), [framework format reference](formats.md), and [documentation policy](policy/documentation.md). |
 | Documentation | The [documentation policy](policy/documentation.md) and canonical owner of every affected contract. |
-| Release preparation or publication | The [release procedure](release.md). |
+| Licensing, dependencies/catalog updates, copied material, or attribution | The [OSS policy](policy/oss.md), [project grant](../COPYRIGHT), [license texts](../LICENSE), and [attribution entries](oss/attribution.md); inspect the selected upstream source's notices. |
+| Release preparation or publication, including release notes | The [release policy](policy/release.md), [release procedure](release.md), and [OSS discoverability rules](policy/oss.md#discoverability). |
 
 For cross-cutting changes, follow every applicable row. Do not create
 placeholder documents for packages, APIs, or integrations that do not yet
 exist.
 
-## Maintainer-Run Validation
+## Maintainer Validation
 
-Promptkit does not currently use hosted CI. Maintainers are responsible for
-running the documented checks before accepting changes. Run the default Go
-validation from the Promptkit repository root:
+For ordinary work, select validation proportionate to the change. Do not repeat
+successful checks for unchanged content merely to create a commit. During
+development, run relevant Go tests and vet, format changed Go files, and execute
+the maintained examples when they are affected:
 
 ```sh
 go test ./...
-go test -race ./...
 go vet ./...
-go build ./...
-go run ./examples/go-library/prepare
-```
-
-Check formatting across every tracked Go file:
-
-```sh
-gofmt -l $(git ls-files '*.go')
-```
-
-The formatting command must produce no paths. Follow every added or changed
-Markdown link and confirm its target exists. Finally, check whitespace:
-
-```sh
+gofmt -w path/to/changed.go
+python3 scripts/check-examples.py
 git diff --check
 ```
 
-Documentation-only work does not require unrelated new tests, but it still
-requires link validation and `git diff --check`. Run the Go validation whenever
-documentation changes commands, examples, generated output, or another
-behavior checked by the module.
+Release candidates use `./scripts/release.sh VERSION`, which runs the selected
+profile checks and publishes the tag. The [release procedure](release.md)
+describes its requirements and recovery. For an optional diagnostic of a
+committed candidate without publication, run `./scripts/releasekit.sh check
+VERSION`; this is not a required pre-publication step.
 
-## Focused Validation
+The standalone `python3 scripts/check-examples.py` command checks semantic
+expectations for the three offline consumer examples. The focused
+`python3 scripts/test-release.py` command exercises the release tooling against
+disposable local repositories; use it when changing that tooling, not as a
+recurring release gate. Tests and examples require no provider credentials or
+live inference calls.
 
-Use focused checks while iterating, then run the complete validation sequence
-before accepting the change. The root package supports:
+Ordinary validation can run on intended uncommitted work. Release validation
+uses committed candidate contents and the selected local Go installation.
 
-```sh
-go test .
-go vet .
-go build .
-```
+### Change Review
 
-Filter tests by name without assuming a fixed internal package layout:
+Inspect `git status --short --untracked-files=all` and the complete diff before
+accepting a change. Reject unintended files, credentials, private data,
+generated binaries, test output, downloaded assets, and template residue.
 
-```sh
-go test ./... -run 'TestName'
-```
+For releases, follow the [release procedure](release.md). Do not repeat its
+candidate checks as a separate sign-off. Roadmap retirement remains preparation
+housekeeping.
 
-Replace `TestName` with a useful regular expression. Target only paths that
-exist, and consult the internal component overview for their owning
-documentation. A filtered or package-specific run does not replace the
-complete repository validation.
-
-## Coordinated Work With Scriptorium
-
-Promptkit and Scriptorium must remain independently valid. For temporary local
-integration, use either a Go workspace outside both repositories or an
-uncommitted replacement in the consuming module.
-
-If the repositories are sibling directories, run the workspace commands from
-their parent directory:
-
-```sh
-go work init ./promptkit ./scriptorium
-go work sync
-```
-
-Use the workspace only for coordinated local checks. From the same parent
-directory, remove it when finished:
-
-```sh
-rm -f go.work go.work.sum
-```
-
-Alternatively, from the Scriptorium repository root, temporarily point its
-Promptkit dependency at the sibling checkout:
-
-```sh
-go mod edit -replace gitea.maximumdirect.net/eric/promptkit=../promptkit
-```
-
-After coordinated checks, remove the replacement and reconcile module
-metadata:
-
-```sh
-go mod edit -dropreplace gitea.maximumdirect.net/eric/promptkit
-go mod tidy
-```
-
-Never commit `go.work`, `go.work.sum`, or a local filesystem `replace`
-directive. Before committing in either repository, inspect its module files and
-working tree independently. Published consumer versions must depend on a tagged
-Promptkit version, not a workspace, local replacement, or unpublished commit.
+Follow the [agent commit policy](policy/agents.md#git-commit-policy) for local
+commits. Ordinary commits may leave unrelated work in the workspace; preserve
+and report it. A clean worktree and index are required for release candidates,
+as enforced by the publication command, not for every ordinary commit.

docs/formats.md

@@ -2,299 +2,316 @@
 
 ## Purpose
 
-This document is the canonical contract for Promptkit prompt-definition,
-profile, and schema files. The [Go package consumer guide](consumers/pkg-promptkit.md)
-explains how to select these sources and invoke the engine. The
+This is the canonical reference for Promptkit's backend, profile, prompt, and
+schema definitions. The [Go consumer guide](consumers/pkg-promptkit.md) shows
+how to select sources and use an engine; the
 [OpenAI-compatible integration contract](integrations/openai-compatible-chat.md)
-owns the resulting outbound wire behavior.
+owns outbound wire behavior. [ConfigKit](https://gitea.maximumdirect.net/eric/ConfigKit)
+owns the generic YAML grammar, imports, layering, inheritance, and diagnostics.
+This reference describes what Promptkit does with the final effective values.
 
-Prompt and profile sources recursively discover files ending in `.yaml` or
-`.yml`. YAML decoding is strict: unknown fields are errors for the selected
-definition. Definitions are selected by their YAML `id`, not their file name
-or directory.
+A configuration document has keyed top-level `backends`, `profiles`, and
+`prompts` mappings. Their keys are literal IDs, not filename-derived identities.
+Prompt versions are a second keyed mapping. Keys and references must be
+nonblank and remain exact: Promptkit does not trim or re-key them. Old embedded
+`id`, `version`, and `base_profile` fields are invalid. Unknown fields are
+errors.
+
+```yaml
+backends:
+  local:
+    endpoint: http://localhost:8000/v1
+    concurrency_limit: 2
+profiles:
+  local-summary:
+    backend: local
+    model: example-model
+prompts:
+  meeting.summary:
+    "1.0.0":
+      default_profile: local-summary
+      messages:
+        - role: user
+          content: Summarize this note.
+      output:
+        format: text
+        validation_mode: none
+```
+
+Every effective backend, profile, and prompt definition is validated when
+`NewEngine` constructs the engine, even if it is never selected. A profile may
+be partial: an endpoint and model need only be complete after request overrides
+are applied. A prompt's default profile similarly need not be runnable without
+an overriding request. Missing request data, artifacts, schemas, credentials,
+and provider work remain runtime concerns.
+
+The engine captures effective definitions, referenced message-file contents,
+and static template syntax at construction. It does not reread configuration or
+message files on ordinary runs; construct a new engine to observe edits.
+Schema and request artifact reads still happen during preparation. Use
+`InspectConfiguration` for an explicit effective-configuration or provenance
+report, `InspectProfile` for backend-completed profile settings, and preparation
+for request-specific rendered details. For identity discovery before inspection,
+see the consumer guide's [profile and prompt listing](consumers/pkg-promptkit.md#discover-available-profiles-and-prompts).
 
 ## Prompt Definitions
 
-A prompt definition describes inputs, Go-template messages, an optional
-default profile, and an output contract.
+The `prompts` mapping selects a literal prompt ID, then a string version key:
 
 ```yaml
-id: meeting.summary
-version: "1.0.0"
-default_profile: local-summary
-description: Summarize a synthetic meeting note.
-session_id: '{{.session}}'
-inputs:
-  - name: note
-    required: true
-    content_type: text/plain
-    description: Meeting note to summarize.
-messages:
-  - role: system
-    content: Return a concise summary.
-    cache_control:
-      type: ephemeral
-      ttl: 1h
-  - role: user
-    content_file: ./summary.user.md
-output:
-  format: markdown
-  validation_mode: basic
-  repair_attempts: 0
+prompts:
+  meeting.summary:
+    "1.0.0":
+      default_profile: local-summary
+      description: Summarize a synthetic meeting note.
+      session_id: '{{.session}}'
+      inputs:
+        - name: note
+          required: true
+          content_type: text/plain
+          description: Meeting note to summarize.
+      messages:
+        - role: system
+          content: Return a concise summary.
+          cache_control:
+            type: ephemeral
+            ttl: 1h
+        - role: user
+          content_file: ./summary.user.md
+      output:
+        format: markdown
+        validation_mode: basic
+        repair_attempts: 0
 ```
 
-| Field | Required | Meaning |
-| --- | --- | --- |
-| `id` | yes | Non-empty prompt identifier used by `RunRequest.PromptID`. |
-| `version` | yes | Non-empty version selected by an optional `RunRequest.PromptVersion`. |
-| `default_profile` | no | Non-empty profile ID used when the request omits `ProfileID`. |
-| `description` | no | Human-readable description. |
-| `session_id` | no | Go template rendered from request variables and input helpers. |
-| `inputs` | no | Declared input metadata. |
-| `messages` | yes | One or more chat-message templates. |
-| `output` | yes | Output format and validation settings. |
-
-When a request omits a version, the selected prompt ID must identify exactly
-one definition. When it supplies a version, the ID and version pair must be
-unique.
+`messages` must be nonempty and `output` must be a valid mapping. Optional
+`default_profile`, `description`, `session_id`, `inputs`, and cache-control
+fields may be absent or cleared with null where appropriate. A null whole
+definition, required role, output format, or validation mode is invalid.
+`required` and `repair_attempts` default to false and zero, respectively, and
+are not nullable. A request that omits a prompt version succeeds only if that
+prompt ID has exactly one effective version. `InspectPrompt` selects from the
+same captured definitions without rendering or resolving a profile.
 
 ### Inputs
 
-Each `inputs` item has these fields:
-
-| Field | Required | Meaning |
-| --- | --- | --- |
-| `name` | yes | Non-empty name used by the request input map and `input` template helper. Names must be unique. |
-| `required` | no | When true, preparation fails if the request omits the input. The default is false. |
-| `content_type` | no | Expected media-type metadata. |
-| `description` | no | Human-readable input description. |
-
-Requests supply inputs as inline or file-backed `ArtifactRef` values. Declared
-required inputs must be present. A template reference also requires the named
-input to exist, whether or not it was declared. Extra request inputs are
-allowed.
+Each `inputs` item may contain `name`, `required`, `content_type`, and
+`description`. `name` is required, nonblank, and unique within the prompt.
+Omitted `required` means false. Null `inputs` means no declarations. Requests
+supply inline or file-backed `ArtifactRef` values; declared required inputs
+must be present. A template reference also needs its named input, even when
+undeclared. Extra request inputs are allowed.
 
 ### Messages And Templates
 
-Each message has a non-empty `role` and exactly one of:
+Each message needs a role and exactly one nonnull `content` or `content_file`.
+An explicitly empty inline `content` is valid. Null clears either content
+selection. Roles are trimmed and lowercased, then must be `developer`,
+`system`, `user`, or `assistant`. Static template syntax is checked during
+construction; missing variables and named inputs fail when rendered.
+Request variables are template data (`{{.audience}}`), and
+`{{input "note"}}` renders an input body.
 
-- `content`, containing an inline Go template; or
-- `content_file`, naming a file whose contents are the Go template.
+`content_file` uses the effective field's own source context, including when
+inherited or imported. Its relative path is resolved against that source's
+base directory within its rooted filesystem. Absolute paths and lexical
+escapes from the selected root fail. Host-file sources also prevent symlink
+escapes. Promptkit preserves the path text, including meaningful whitespace,
+after checking that it is not blank. A source without a unique usable resource
+context must supply one explicitly or override the field. Content is captured
+at construction, not reopened for each request.
 
-For directory and `fs.FS` prompt sources, `content_file` resolves relative to
-the prompt file and remains within the source root. `WithPromptFile` also
-resolves it relative to that file.
+Optional `session_id` uses the same template variables and input helper. Its
+rendered value is trimmed, omitted if empty, and limited to 256 Unicode code
+points. A nonblank direct request session ID bypasses the template.
 
-Request variables are the template data, so a variable named `audience` is
-referenced as `{{.audience}}`. The `{{input "note"}}` helper renders the body
-of a named input. Missing variables and input references are errors.
-
-The optional `session_id` uses the same template data and input helper. Its
-rendered value is trimmed, omitted when empty, and limited to 256 Unicode code
-points. A nonblank direct request session ID bypasses this template completely;
-a blank direct value leaves the template behavior unchanged.
+`InspectRequirements` can enumerate recognizable dependencies from captured
+templates before artifact content is available. Its conservative syntax
+coverage does not change template validity or rendering; see the [consumer
+guide](consumers/pkg-promptkit.md#inspect-requirements-before-preparation) and
+the method's [GoDoc](../requirements.go) for that inspection boundary.
 
 ### Cache Control
 
-`cache_control` is optional and has these fields:
-
-| Field | Required | Values |
-| --- | --- | --- |
-| `type` | yes | `ephemeral` |
-| `ttl` | no | Empty or `1h` |
-
-Promptkit preserves cache-control metadata on the rendered message. The
-outbound integration determines its wire representation.
+Optional `cache_control` has required `type: ephemeral` and optional `ttl`,
+which may be empty or `1h`. Promptkit retains this metadata on rendered
+messages; the integration contract defines its wire representation.
 
 ### Output Contract
 
 | Field | Required | Values or behavior |
 | --- | --- | --- |
-| `format` | yes | `text`, `markdown`, or `json`. |
-| `validation_mode` | yes | `none`, `basic`, `json`, or `json_schema`. |
-| `schema_path` | for `json_schema` | Path to a schema in the configured schema source. |
-| `repair_attempts` | no | Integer zero or greater; omitted means zero. |
+| `format` | yes | `text`, `markdown`, or `json` |
+| `validation_mode` | yes | `none`, `basic`, `json`, or `json_schema` |
+| `schema_path` | for `json_schema` | Path in the selected schema source |
+| `repair_attempts` | no | Integer 0–3; default 0. Positive values need `basic`, `json`, or `json_schema` |
 
-The validation modes behave as follows:
+`none` skips content validation; `basic` requires nonempty content; `json`
+accepts any valid JSON value; `json_schema` checks the selected schema.
+`format` determines output artifact metadata. JSON Schema mode also supplies
+structured-output metadata to compatible clients. Plain `json` does not
+request a provider-native JSON-object constraint. A positive repair budget
+counts additional generations after failed validation; exhaustion returns the
+last candidate and its failed validation result. Operational errors remain
+errors. A request-level `OutputContract` replaces the whole configured one,
+not individual fields; an empty request format defaults to `text`.
 
-- `none` skips content validation;
-- `basic` requires non-empty generated content;
-- `json` requires valid JSON; and
-- `json_schema` requires valid JSON that satisfies the selected schema.
+## Built-In Backends
 
-`format` controls output artifact metadata. JSON Schema mode also supplies the
-schema to compatible model clients as structured-output metadata. The public
-engine does not install an output repairer, so its validation is single-pass
-even when a positive `repair_attempts` value is present.
+The selected catalog modules contribute these backend definitions as the
+earliest underlay. Their settings can be overlaid by caller sources; these IDs
+are not reserved against configuration changes.
 
-A request-level `OutputContract` replaces the complete prompt output contract.
-It does not merge individual fields. If its format is empty, Promptkit uses
-`text`.
+| ID | Catalog endpoint | API-key environment name | Active generation limit |
+| --- | --- | --- | ---: |
+| `openrouter` | `https://openrouter.ai/api/v1` | `OPENROUTER_API_KEY` | 16 |
+| `rakestrawhome` | `https://inference.ai.rakestrawhome.com/v1` | `RAKESTRAWHOME_INFERENCE_API_KEY` | 4 |
+
+A backend definition may have `endpoint`, `api_key_env`,
+`concurrency_limit`, `queue_capacity`, and `extra_params`. The final endpoint
+is required and must be an absolute HTTP(S) URL with a host, no user
+information, query, or fragment. An omitted or null API-key environment name
+means no lookup. Omitted/null concurrency is zero (unlimited); a supplied
+negative value is invalid. For positive concurrency, omitted/null queue
+capacity defaults to 1024. An explicit queue capacity, including zero,
+requires positive concurrency and must be nonnegative. Unlimited concurrency
+has no queue. Provider maps contain JSON-compatible values; null entries are
+ordinary JSON null. Unknown fields, including a raw `api_key`, fail.
 
 ## Profile Definitions
 
-A profile supplies model execution settings:
+Profiles are keyed by literal ID and may inherit a named parent with
+`extends`:
 
 ```yaml
-id: local-summary
-backend: openrouter
-model: example-model
-temperature: 0.2
-max_tokens: 500
-top_p: 0.95
-timeout_seconds: 90
-service_tier: flex
-reasoning_effort: medium
-extra_params:
-  provider_option: enabled
+profiles:
+  local-summary:
+    backend: local
+    model: example-model
+    temperature: 0.2
+    max_tokens: 500
+    timeout_seconds: 90
+    extra_params:
+      provider_option: enabled
+  local-summary-fast:
+    extends: local-summary
+    timeout_seconds: 30
+    reasoning_effort: low
 ```
 
-| Field | Required | Meaning |
-| --- | --- | --- |
-| `id` | yes | Non-empty profile identifier. IDs must be unique within one source. |
-| `backend` | unless `endpoint` is present | Backend registry ID. It is trimmed and registry membership is checked when the profile is prepared. |
-| `endpoint` | unless `backend` is present | Non-empty OpenAI-compatible base URL, including an API version path when required. When both connection fields are present, this overrides the backend endpoint without changing backend identity. |
-| `model` | yes | Non-empty provider model name. |
-| `temperature` | no | Number from 0 through 2. |
-| `max_tokens` | no | Integer zero or greater. |
-| `top_p` | no | Number from 0 through 1. |
-| `timeout_seconds` | no | Per-generation deadline in whole seconds; integer zero or greater. |
-| `service_tier` | no | Provider-specific request tier. |
-| `reasoning_effort` | no | Provider-specific reasoning setting. |
-| `api_key_env` | no | Name of an environment variable containing the API key. |
-| `extra_params` | no | JSON-compatible provider-specific outbound fields. |
-
-Raw `api_key` is prohibited in profile YAML. Store only an environment
-variable name in `api_key_env`.
-
-Promptkit does not infer a backend from a model or endpoint. Endpoint-only
-profiles remain supported and have no effective backend ID.
-The engine always provides the built-in `openrouter` ID. Consumers can add
-engine-scoped IDs with
-[`WithBackend`](../backends.go); exact registration validation belongs to its
-GoDoc.
-
-`extra_params` accepts null, booleans, finite numbers, strings, arrays, and
-objects with string keys. Keys must be non-empty. With the built-in client,
-they also cannot collide with the standard fields listed in the
-[outbound request contract](integrations/openai-compatible-chat.md#request-body).
+Profiles may also supply `endpoint`, `top_p`, `service_tier`, `api_key_env`,
+and `api_key_required`. Promptkit does not infer a backend from an endpoint or
+model. A profile with only an endpoint has no backend capacity identity.
+Nonblank endpoints follow the backend URL rules; temperature is 0–2, top_p
+is 0–1, max_tokens and timeout_seconds are nonnegative. Provider maps accept
+JSON-compatible values with nonempty keys and cannot collide with the built-in
+client's [standard request fields](integrations/openai-compatible-chat.md#request-body).
+Credential values do not belong in profiles: raw `api_key` is invalid.
 
 ### Defaults And Overrides
 
-Execution settings resolve in this order:
+After effective profile inheritance, Promptkit applies the selected backend's
+defaults once, then request `ExecutionTargetOverride` values. Backend-derived
+settings are not inheritance input: changing a child profile's backend does
+not carry the parent's derived endpoint or credential source into it. Changing
+only a model or backend does not clear unrelated explicitly inherited fields.
 
-1. framework defaults;
-2. the selected backend, when the profile names one;
-3. the selected profile; and
-4. request `ExecutionTargetOverride` values.
+| Profile field | Absent | Explicit value | Explicit null |
+| --- | --- | --- | --- |
+| `backend` | No backend | Exact backend ID; blank means none | No backend |
+| `endpoint` | Selected backend endpoint | Valid nonblank URL; empty clears | Clear, without backend fallback |
+| `model` | Unspecified | Nonblank model; empty clears | Unspecified |
+| `temperature`, `top_p`, `max_tokens` | Unspecified | Valid value, including outbound zero | Suppress outbound control |
+| `service_tier`, `reasoning_effort` | Unspecified | Normalized value; blank suppresses | Suppress outbound control |
+| `timeout_seconds` | 600 seconds | Nonnegative; zero disables generation deadline | Reset to 600 seconds |
+| `api_key_env` | Backend lookup name, if any | Trimmed name; blank disables | Disable backend lookup |
+| `api_key_required` | False | Boolean, including false | False |
+| `extra_params` | Backend defaults | Recursively overlay profile map on backend map | Suppress configured provider map |
 
-The framework defaults are:
-
-| Setting | Default |
-| --- | --- |
-| `temperature` | `0` |
-| `max_tokens` | `0` |
-| `top_p` | `1` |
-| `timeout_seconds` | `600` |
-
-Numeric zero in a file or in-memory profile means that the profile does not
-replace the framework default. Numeric request overrides use pointers, so an
-explicit zero is preserved. In particular, an explicit request
-`timeout_seconds` of zero disables the per-generation deadline while leaving
-the caller context and transport timeout intact.
-
-Non-empty profile strings replace backend defaults, and non-empty request
-strings replace both. Request reasoning is the exception: a nil
-`ReasoningEffort` pointer inherits the profile, a pointer to a nonblank string
-trims and replaces it, and a pointer to a blank string clears it. Backend
-identity is retained when either layer overrides the endpoint, so the override
-also retains any engine-local capacity policy configured for that backend.
-Capacity configuration belongs to the Go
-[`Backend` API](../backends.go), not prompt or profile YAML. A non-empty
-`extra_params` map at each layer replaces the entire lower-precedence map
-rather than merging keys.
-The [outbound integration contract](integrations/openai-compatible-chat.md)
-defines how the effective settings are serialized.
+An empty mapping does not clear inherited map entries; null clears the whole
+map. Nested maps merge recursively, while scalar and array values replace.
+Runtime nonempty `ExtraParams` overrides replace the entire configured map.
+Request overrides otherwise follow the [public value GoDoc](../types.go).
+For example, a runtime `extra_params["extends"]` key is ordinary provider
+data, not a configuration inheritance instruction. An explicit zero timeout
+does not disable the caller context or HTTP transport timeout.
 
 ### Source And Profile Precedence
 
-An explicit request profile ID takes precedence over the prompt's
-`default_profile`. If neither is present, preparation fails.
+`NewEngine` composes catalog definitions first, Promptkit-owned definitions
+next, then caller underlays in their given order, with an optional main input
+last. Later tiers overlay earlier ones. Inputs within one underlay group are
+strictly composed: conflicting contributions are errors, not priority
+signals. The directory helpers discover `.yaml` and `.yml` files recursively
+in sorted order, but that order does not give one file higher precedence.
+An explicitly selected missing/unreadable root fails; an empty root contributes
+nothing. Imports are disabled by these helpers unless the caller enables them
+on an input. Single files can use ConfigKit inputs directly.
 
-Profile sources resolve matching IDs in this order:
+Backends, profiles, and prompts all follow these same source rules. A later
+profile or backend can change individual fields of a lower definition. A
+request `ProfileID` takes precedence over a prompt `default_profile`; if
+neither is selected, preparation fails. Exact profile inspection selects one
+explicit ID and does not consult a prompt default.
 
-1. in-memory profiles supplied with `WithProfiles`;
-2. a profile file, `fs.FS`, or configured profile directory; and
-3. embedded built-in profiles.
+### Profile Inheritance
 
-A higher-precedence source falls back only when the profile is absent. An
-invalid matching profile is an error and does not fall back. In-memory
-`Profile` values follow the same ranges as YAML profiles. They use
-`APIKeyRequired` for request-scoped credentials instead of `api_key_env`.
+ConfigKit resolves `extends` on effective profile definitions. One profile
+may inherit another keyed profile; presence, explicit null, and recursively
+merged maps determine the final fields. References are literal IDs, including
+IDs containing punctuation. There is no Promptkit-specific depth limit or
+second base-chain resolver. Missing references and cycles fail at construction.
+For generic inheritance and layer mechanics, consult
+[ConfigKit](https://gitea.maximumdirect.net/eric/ConfigKit).
 
 ## Built-In Profile Catalog
 
-Every built-in selects the `openrouter` backend. The engine's built-in backend
-registry supplies `https://openrouter.ai/api/v1` and the environment-variable
-name `OPENROUTER_API_KEY`, so individual profiles contain only model and
-generation settings. Built-in profile files do not repeat those connection
-values. A custom or in-memory profile with the same profile ID takes
-precedence.
+Promptkit contributes two opt-in definitions; neither is selected by default:
 
-| Provider | ID | Model |
-| --- | --- | --- |
-| aion-labs | `aion-2` | `aion-labs/aion-2.0` |
-| anthropic | `claude-fable-latest` | `~anthropic/claude-fable-latest` |
-| anthropic | `claude-haiku-latest` | `~anthropic/claude-haiku-latest` |
-| anthropic | `claude-opus-latest` | `~anthropic/claude-opus-latest` |
-| anthropic | `claude-sonnet-latest` | `~anthropic/claude-sonnet-latest` |
-| deepseek | `deepseek-3-2` | `deepseek/deepseek-v3.2` |
-| deepseek | `deepseek-4-flash` | `deepseek/deepseek-v4-flash` |
-| deepseek | `deepseek-4-pro` | `deepseek/deepseek-v4-pro` |
-| google | `gemini-2-flash` | `google/gemini-2.5-flash` |
-| google | `gemini-2-flash-lite` | `google/gemini-2.5-flash-lite` |
-| google | `gemini-2-pro` | `google/gemini-2.5-pro` |
-| google | `gemini-3-flash-lite` | `google/gemini-3.1-flash-lite` |
-| google | `gemini-flash-latest` | `~google/gemini-flash-latest` |
-| google | `gemini-pro-latest` | `~google/gemini-pro-latest` |
-| google | `gemma-4-31b` | `google/gemma-4-31b-it:exacto` |
-| minimax | `minimax-m2` | `minimax/minimax-m2.5` |
-| minimax | `minimax-m3` | `minimax/minimax-m3` |
-| mistral | `mistral-large-2512` | `mistralai/mistral-large-2512` |
-| mistral | `mistral-medium-3-5` | `mistralai/mistral-medium-3-5` |
-| mistral | `mistral-small-3` | `mistralai/mistral-small-3.2-24b-instruct` |
-| mistral | `mistral-small-4` | `mistralai/mistral-small-2603` |
-| nvidia | `nemotron-3-ultra` | `nvidia/nemotron-3-ultra-550b-a55b` |
-| openai | `gpt-5-mini` | `openai/gpt-5.4-mini` |
-| openai | `gpt-5-nano` | `openai/gpt-5.4-nano` |
+| ID | Current definition |
+| --- | --- |
+| `promptkit-openrouter-standard` | Extends the selected OpenRouter catalog's `gemini-flash-latest` profile. |
+| `promptkit-openrouter-auto` | Selects backend `openrouter` and model `openrouter/auto`. |
+
+Select one through request `ProfileID` or prompt `default_profile`.
+Caller layers may overlay these definitions or their catalog bases. The
+[release policy](policy/release.md#catalog-profile-compatibility) owns their
+cross-version compatibility boundary.
+
+## External Profile Inventories
+
+The selected catalog modules, not this document, own their profile inventories:
+
+- [OpenRouter catalog v3.20260930.0 profiles](https://gitea.maximumdirect.net/eric/promptkit-backend-openrouter/src/tag/v3.20260930.0/catalog/profiles)
+- [Rakestrawhome catalog v3.20260930.0 profiles](https://gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/src/tag/v3.20260930.0/catalog/profiles)
+
+These are tested local data, not a promise of live provider availability.
+The [release policy](policy/release.md#catalog-profile-compatibility) owns
+the compatibility boundary.
 
 ## Schemas
 
-Schemas are JSON documents selected by a prompt or request
-`schema_path`. For a directory or `fs.FS` source, paths resolve within the
-configured source root. Referenced nested schemas resolve relative to the
-owning schema document. `WithSchemaFile` exposes one schema, addressed by its
-base name.
-
-An unreadable, invalid, or unresolvable schema produces an operational
-validation error. Generated content that is valid JSON but does not satisfy the
-schema produces a failed validation result.
+Schemas are JSON documents selected by a prompt or request `schema_path`.
+Directory and `fs.FS` schema paths resolve within the configured root;
+referenced nested schemas resolve relative to the owning schema document.
+`WithSchemaFile` exposes one schema under its base name. An unreadable,
+invalid, or unresolvable schema causes an operational validation error.
+Valid JSON that fails the schema instead yields a failed validation result.
 
 ## Credentials
 
-Credential values belong at the request or environment boundary, never in
-prompt, profile, schema, or example files:
+Only environment-variable names belong in backend/profile definitions, never
+credential values. A request may supply a direct `APIKey` or an `APIKeyEnv`
+override. A direct key takes precedence, followed by a request environment
+source, an explicit profile source, and an inherited backend source. A true
+profile `api_key_required` suppresses the backend-derived environment source,
+but an explicit profile or request environment source remains usable. False
+restores ordinary optional-credential behavior. Missing or blank optional
+environment values leave the built-in client's Authorization header absent;
+required availability is checked at preparation and again during prepared
+execution. Injected clients own their credential-resolution behavior.
 
-- a file profile names an environment variable with `api_key_env`;
-- an in-memory profile may set `APIKeyRequired`;
-- a request can provide a direct `APIKey` or override `APIKeyEnv`; and
-- a direct request key takes precedence over environment lookup.
-
-After a direct request key, the credential-source precedence is request
-`APIKeyEnv`, profile `api_key_env`, then the backend default. An in-memory
-profile with `APIKeyRequired` clears an inherited backend environment name and
-requires a direct key unless the request explicitly supplies `APIKeyEnv`.
-Promptkit validates required credential availability during preparation.
-Direct keys are excluded from JSON results and redacted by public string
-formatters. Environment-variable names may appear in prepared metadata, but
-their values do not.
+Credential values are excluded from JSON results and redacted by public
+string formatters. ConfigKit inspection redacts credential-shaped dynamic
+provider keys, but does not make arbitrary prompt/provider content safe to
+disclose. Applications own report disclosure and secret handling.

docs/integrations/openai-compatible-chat.md

@@ -14,10 +14,16 @@ that produce these outbound settings.
 
 Generation sends an HTTP `POST` with `Content-Type: application/json`.
 Before the client is called, the engine resolves framework, backend, profile,
-and request values into one execution target. A non-empty endpoint from that
-target overrides the client's configured base URL. After trailing slashes are
-removed, `/chat/completions` is appended. Generation fails before sending when
-neither source supplies an endpoint.
+and request values into one execution target. Endpoint configuration is trimmed
+and must be an absolute HTTP or HTTPS URL with a host and without user
+information, a query, or a fragment. A non-empty endpoint from the target
+overrides the client's configured base URL. The final selected endpoint is
+validated again before transport.
+
+The completion URL is composed through parsed URL path operations. Nested base
+paths are retained, repeated trailing slashes are normalized, and the result
+has exactly one appended `/chat/completions` suffix. Generation fails before
+sending when neither source supplies a valid endpoint.
 
 The target's backend ID is routing metadata for prepared values, results, and
 injected clients. The built-in client does not derive the URL from that ID and
@@ -25,11 +31,13 @@ does not serialize it in the provider request.
 
 ## Authentication
 
-A non-empty API key supplied directly on the execution target takes
-precedence. Otherwise, when an API-key environment-variable name is supplied,
-the client reads that variable and requires a non-empty value. The selected
-key is sent as `Authorization: Bearer <key>`. No authorization header is sent
-when neither mechanism is configured.
+A usable API key supplied directly on the execution target takes precedence.
+Otherwise, when an API-key environment-variable name is supplied, the client
+reads and trims that variable. A bearer header is sent only when the resolved
+direct or environment credential is non-empty. When neither source is usable,
+the client omits `Authorization` and handles the provider response normally.
+An explicitly required target with no usable source is rejected before
+transport.
 
 The target contains the already resolved environment-variable name: an
 explicit request override takes precedence over profile metadata, which takes
@@ -47,24 +55,31 @@ Each ordinary message contains its `role` and string `content`. A
 cache-controlled message instead uses a text content block containing `type`,
 `text`, and `cache_control`; an empty cache-control TTL is omitted.
 
+Promptkit sends only `developer`, `system`, `user`, and `assistant` roles and
+does so without provider-specific translation. Tool and deprecated function
+payloads are outside this text-message contract. A backend or model that
+rejects an otherwise supported role or context returns its ordinary provider
+error, which follows the normal generation-error path.
+
 The effective direct or prompt-rendered session ID is trimmed, limited to 256
 Unicode code points, and sent when nonempty as top-level `session_id`. It is
 never also sent as a session header.
 
 The client conditionally includes:
 
-- `temperature`, `max_tokens`, and `top_p` when non-zero or explicitly
-  present;
+- `temperature`, `max_tokens`, and `top_p` only when selected by a profile or
+  runtime override, including an explicit runtime zero; they are absent when
+  unspecified;
 - non-empty `service_tier` and effective `reasoning_effort`; an explicitly
   disabled reasoning setting is empty and therefore omitted; and
 - `response_format` for JSON Schema structured output, including its name,
-  strict flag, and schema document.
+  strict flag, and schema document. Plain JSON validation does not add an
+  object-only response constraint.
 
 The engine resolves backend, profile, and request extra-parameter maps by
 whole-map replacement rather than key merging. The resulting effective map is
-then merged directly into the top-level body after JSON serialization is
-verified. Empty keys and collisions with these reserved fields are rejected
-before any provider call:
+merged into the top-level body. Empty keys, values that cannot be encoded, and
+collisions with these reserved fields are rejected before any provider call:
 
 - `model`
 - `session_id`
@@ -81,13 +96,82 @@ request fields.
 
 ## Response Handling
 
-Any 2xx response is decoded as an OpenAI-compatible chat response. The client
-returns the first choice's non-empty message content and maps prompt,
-completion, total, cached, and cache-write token counts.
+Any 2xx response body is limited to 16 MiB (16,777,216 bytes). A larger
+declared `Content-Length` is rejected before the body is read, and streamed,
+chunked, or underreported bodies are read through the same bound with at most
+one additional byte used to detect overflow. A body exactly at the limit is
+allowed. The body is closed on every outcome and an oversized stream is not
+drained.
 
-Invalid JSON, absent choices, and empty first-choice content are malformed
-responses. For a non-2xx status, the error includes the status code but never
-the provider response body.
+The bounded body must contain exactly one OpenAI-compatible JSON response
+object followed only by JSON whitespace and EOF. The client returns the first
+choice's explicitly present string message content, including an empty or
+whitespace-only string, and maps prompt, completion, total, cached, and
+cache-write token counts. Invalid or truncated JSON, trailing non-whitespace
+data, a second JSON value, absent choices, missing content, `null` content,
+non-string content, and size overflow are malformed responses and return no
+partial result.
+
+Each of the five token fields retains whether the provider reported a
+non-null integer, including zero. Omitted and `null` fields are unknown for
+observations. A strictly decoded successful envelope can yield observed usage
+even when a later missing choice or content makes the attempt malformed.
+Invalid framing and non-2xx error bodies supply no observed usage; provider
+work may still have occurred. Reported counts are not inferred, reconciled,
+or proof of complete cost accounting.
+
+For a non-2xx status, Promptkit recognizes one JSON document with a top-level
+object-valued `error` member. Its optional `message` and `type` fields must be
+strings, and `code` may be a string or JSON number. Valid supported fields are
+handled independently, numeric codes retain their JSON number text, and
+unknown fields are ignored. Missing, invalid, malformed, or multiply framed
+envelopes contribute no provider detail.
+
+Non-success bodies have a 65,536-byte limit. A larger declared
+`Content-Length` is not read; otherwise the client reads at most one additional
+byte to detect streamed or underreported overflow. Empty, unreadable,
+oversized, malformed, and unrecognized bodies retain only the received status.
+The body is always closed and no oversized stream is drained beyond that probe.
+
+Extracted strings are made valid UTF-8, trimmed, and converted to one line by
+collapsing Unicode whitespace, control, and format-character runs. Blank
+values are omitted. Codes and types longer than 256 Unicode code points are
+omitted; messages longer than 4,096 code points are truncated at a code-point
+boundary with an ellipsis inside the limit. Promptkit never exposes raw bodies,
+headers, endpoints, credentials, request data, schemas, generated content, or
+unsupported provider metadata through this handling.
+
+An outbound `http.Client.Do` failure retains both Promptkit's request-failure
+identity and the exact transport error for `errors.Is` and `errors.As` checks.
+The rendered error does not include the selected endpoint, request headers,
+request content, credentials, or provider body.
+
+## Transient Status Retries
+
+The built-in path retries only HTTP 429, 500, 502, 503, and 504. It makes two
+additional attempts by default; engine configuration or an individual run can
+choose a different nonnegative count, including zero for single-attempt
+behavior. The same budget applies separately to initial generation and each
+structural repair generation. An injected `LLMClient` is dispatched once by
+Promptkit regardless of this setting. A custom `http.Client` on the built-in
+path remains subject to these retries.
+
+Each retry uses equal jitter between half and all of an exponential local
+delay, starting at 500 milliseconds and capped at 8 seconds. The first
+`Retry-After` header value sets a minimum not-before time measured from
+response receipt if it is valid unsigned decimal seconds or an HTTP date.
+Invalid values use local delay; past dates impose no extra delay. The provider
+minimum is not capped at 8 seconds. Retry waiting is cancellable and holds no
+active backend permit. A caller without a deadline can wait a long time for
+provider timing.
+
+The request body, selected URL, and resolved authorization are fixed across
+attempts of one generation. Each attempt creates a fresh HTTP request and
+receives its own configured generation and HTTP-client timeout after capacity
+admission. The caller context covers every attempt and wait without a reset.
+Transport failures, malformed responses, noneligible statuses, and invalid
+generated content are not transient retries. A positive budget ending in
+eligible statuses reports retry exhaustion with the final provider cause.
 
 ## Timeout And Cancellation
 
@@ -102,5 +186,7 @@ Timeouts are layered:
   timeout when the supplied value is not positive.
 
 The earliest applicable caller, generation, or transport deadline controls the
-request. Constructing the internal client does not mutate a supplied
-`http.Client`.
+request. Caller cancellation retains `context.Canceled`; caller, generation,
+and whole-request timeout failures retain `context.DeadlineExceeded`, together
+with the request-failure identity. Constructing the internal client does not
+mutate a supplied `http.Client`.

docs/internal/capacity.md

@@ -14,7 +14,7 @@ stable JSON values.
 
 ## Construction And Pool Lifecycle
 
-Each root `NewEngine` call obtains a normalized capacity-policy snapshot from
+Each internal engine construction obtains a normalized capacity-policy snapshot from
 its immutable backend registry and constructs a new `Manager`. The manager
 creates one pool for each limited backend ID. It has no package-global mutable
 state, background workers, shutdown protocol, or persistence, so engines with
@@ -29,51 +29,73 @@ One pool owns immutable active and total limits plus mutex-protected admission
 count, active count, and ordered waiter list. Pool state exists only for the
 lifetime of its engine.
 
-## Bounded Run Admission
+## Bounded Execution Admission
 
-The runner asks the manager to admit a run after resolving the prompt, profile,
-selected backend, effective execution target, credentials, and output contract,
-but before schema loading, artifact loading, or rendering. Admission is
-immediate: a limited pool either reserves a slot or returns the internal
-`ErrCapacityExceeded` identity. The root facade maps that identity to the
-public error without treating it as an invalid request or generation failure.
+For ordinary `Run`, the runner asks the manager to admit after resolving the
+prompt, profile, selected backend, effective execution target, credentials, and
+output contract, but before schema loading, artifact loading, or rendering.
+`PrepareExecution` performs no admission. `RunPrepared` claims its handle,
+rechecks credential availability, and then asks the manager to admit the
+frozen backend before generation.
+
+Admission is immediate: a limited pool either reserves a slot or returns only
+the internal `ErrCapacityExceeded` identity. The runner attaches the selected
+backend identity at its use-case boundary, and the root facade translates that
+typed value without treating it as an invalid request or generation failure.
 
 The total admitted bound is the active-generation limit plus its configured
 waiting capacity. The returned release function is idempotent. The runner
-defers it as soon as admission succeeds and holds the lease across remaining
-preparation, initial generation, validation, every repair attempt, and all
-failure or cancellation exits. A repair is part of its original admission and
-does not reserve another bounded slot.
+defers it as soon as admission succeeds. An ordinary run holds the lease across
+remaining preparation, initial generation, validation, every repair attempt,
+and all failure or cancellation exits. Prepared execution holds the normal
+lease across generation, validation, every internal repair attempt, and all
+execution exits. A repair is part of its original admission and does not
+reserve another bounded slot.
 
 ## FIFO Generation Permits
 
-`NewClient` wraps the engine's selected internal model client after public
-client adaptation or built-in client construction. Initial generation and the
-default repairer receive the same wrapper.
+The engine supplies one [generation coordinator](generation.md) to initial
+generation and the default repairer. Their requests retain the same effective
+backend, credential, numeric-presence metadata, structured-output settings,
+and retry budget. Scheduling does not change provider omission semantics
+between calls.
 
-For each `Generate` call, the wrapper selects a pool from the request's
-effective backend ID. An unlimited call passes directly to the next client. A
-limited call acquires an active permit, invokes the next client, and defers
-permit release so ordinary returns and panic unwinding both restore capacity.
+For each model attempt, the coordinator selects a pool from the request's
+effective backend ID. An unlimited call uses the unrestricted path. A limited
+call acquires an active permit and defers release around that attempt, so
+ordinary returns and panic unwinding both restore capacity.
 Preparation and validation never hold an active permit.
 
+The manager's `Acquire` operation owns that active-permit acquisition and
+returns an idempotent release function. The coordinator uses it around
+individual attempts. The earlier one-shot wrapper still delegates to the same
+operation for focused internal callers and tests, but engine assembly no
+longer uses it. There is one queue and one active limit per limited backend pool.
+
 When all active permits are occupied, calls join a mutex-protected FIFO waiter
 list. Releasing a permit transfers it directly to the oldest remaining waiter
 before making it generally available. Pools do not order work relative to
 other backend IDs.
 
-The wrapper passes generation requests, responses, and collaborator errors
-through unchanged. It owns scheduling only; the concrete model client remains
-responsible for provider transport behavior.
+The coordinator does not retry injected clients; it normalizes a nil successful
+response as malformed generation. The built-in HTTP client remains responsible
+for provider transport behavior. The runner, rather than the
+capacity layer, sums all five usage fields from the initial response and every
+completed repair response into the successful run result.
+
+An invocation-local observer receives attempt events after active permit
+release, including canceled acquisition without dispatch. A retry wait holds
+the logical admission lease but no active permit. Observer callbacks must not
+block on work that requires the same run's admission slot.
 
 ## Cancellation And Release
 
 Admission checks the caller context before reserving a slot. A call canceled
 while waiting for an active permit removes its waiter under the same pool lock
-used to grant permits. If cancellation removes the waiter first, the wrapped
-client is not invoked. If a concurrent grant wins first, the call owns the
-permit and invokes the client with the original context, allowing the client
-to observe cancellation normally.
+used to grant permits. If cancellation removes the waiter first, no model
+attempt is dispatched. If a concurrent grant wins first, the call owns the
+permit; the coordinator checks the context before dispatch and releases the
+permit if cancellation has already occurred.
 
 This grant-or-cancel decision prevents lost and double-released permits.
 Admission leases and active permits are released after success, collaborator
@@ -90,11 +112,19 @@ unlimited admission. The
 FIFO transfer, canceled-waiter removal, grant/cancel races, independent pools,
 unlimited calls, passthrough behavior, and panic release.
 
-The [runner tests](../../internal/usecase/runner_test.go) own early admission,
-lease lifetime, failure release, and shared initial/repair scheduling. The
+The [runner tests](../../internal/usecase/runner_test.go) own ordinary early
+admission, lease lifetime, failure release, and shared initial/repair
+scheduling. The
+[prepared-execution use-case tests](../../internal/usecase/prepared_execution_test.go)
+own deferred admission, credential ordering, and prepared-execution lease
+release. The
 [external package capacity tests](../../capacity_contract_test.go) own the
 assembled public-engine behavior for configured limits, capacity errors,
 endpoint identity, engine independence, and injected clients. The
+[reliable execution contract tests](../../reliable_execution_contract_test.go)
+cover shared and independent backend pools during retry and repair. The
+[prepared-execution contract tests](../../prepared_execution_contract_test.go)
+own the public prepared-capacity boundary. The
 [root error-boundary tests](../../errors_internal_test.go) own preservation of
 the public generation category and context identity when generation is
 canceled.

docs/internal/generation.md

@@ -0,0 +1,50 @@
+# Internal Generation Coordination
+
+`internal/generation` coordinates one model-generation request through the
+existing engine-local [capacity manager](capacity.md) and the
+[built-in HTTP client](llm.md). The engine supplies the same coordinator to
+initial generation and the default output repairer.
+
+The coordinator has explicit built-in and injected-client constructors. The
+built-in path prepares the provider request once, then acquires and releases an
+active backend permit around each actual HTTP attempt. Only HTTP 429, 500, 502,
+503, and 504 can lead to another attempt. The injected-client path acquires a
+permit and invokes its client once, regardless of the internal retry budget.
+The runner continues to own the logical execution's admission lease. A request
+can override the engine's retry budget; the runner freezes that choice for
+prepared execution and passes it to each repair generation.
+
+Retry waits hold no active permit. The local delay uses equal jitter between
+half and all of a capped exponential base, beginning at 500 milliseconds and
+capping at 8 seconds. A valid provider not-before time can extend that delay;
+the caller context interrupts it. The next attempt reacquires the ordinary
+FIFO permit. Preparation failures dispatch nothing. When a positive retry
+budget ends with eligible failures, the internal exhaustion error retains the
+final cause and actual dispatched-attempt count.
+
+When an invocation-local [observation recorder](observation.md) is present in
+the context, each acquisition emits an attempt event, including canceled
+capacity waits without dispatch. Each actual retry wait emits a delay event,
+including interrupted waits. Active permits are released before callbacks;
+generation duration ends before callback delivery. Built-in successful-envelope
+usage retains reported-field presence, while injected clients use explicit
+presence when supplied and otherwise treat zero fields as unknown.
+Malformed successful envelopes can retain strictly decoded usage; error
+response bodies are never mined for it. Arbitrary injected failures carry
+unknown usage. Transport, timeout, malformed, and noneligible status failures
+stop the sequence without retry exhaustion. Context is checked before each
+wait, acquisition, and dispatch.
+
+The public client adapter translates explicitly supplied public transport and
+malformed-response identities into internal categories while retaining the
+original error chain. This keeps injected attempt classifications consistent
+with terminal observations without parsing diagnostic text or retrying injected
+clients.
+
+Focused coordinator tests use controlled transport and wait functions. They
+cover eligibility, exhaustion, cancellation, backoff, request freezing, permit
+release, and the injected-client path. The capacity manager tests own permit
+acquisition and idempotent release. Public
+[integration tests](../../reliable_execution_contract_test.go) cover retry,
+repair, observation, frozen prepared execution, and shared-backend behavior
+through the assembled facade.

docs/internal/llm.md

@@ -10,7 +10,7 @@ owns the observable outbound HTTP contract. The
 [framework format reference](../formats.md) owns the profile and prompt
 settings consumed by the client.
 
-The concrete client remains under `internal/llm`. The root engine assembles it
+The concrete client remains under `internal/llm`. The internal engine assembles it
 as the default implementation behind Promptkit's public client boundary.
 
 ## Components And Flow
@@ -21,38 +21,113 @@ uses internal domain values for rendered prompts, execution targets,
 structured output, responses, and token usage.
 
 The runner supplies a fully resolved target after applying backend, profile,
-and request precedence. The client uses its endpoint, credential metadata,
-generation fields, and extra parameters. `BackendID` remains routing metadata
+and request precedence, plus canonical provider-bound text messages. The
+client uses its endpoint, credential metadata, generation fields, and extra
+parameters. `BackendID` remains routing metadata
 for the generation boundary and is not mapped into the provider payload.
 
-Construction validates the configured base URL and clones any supplied
-`http.Client` so Promptkit can apply its timeout default without mutating the
-caller's client. Generation then:
+Construction trims and validates a nonempty configured base URL and clones any
+supplied `http.Client` so Promptkit can apply its timeout default without
+mutating the caller's client. An empty configured base remains valid because a
+resolved request target may supply the endpoint. Generation preparation then:
 
-1. validates request-level timeout and endpoint requirements;
+1. validates shared execution-setting invariants and the final selected base
+   endpoint;
 2. maps the internal request into the OpenAI-compatible chat payload;
-3. validates and merges extra parameters;
-4. resolves authentication;
-5. performs the outbound request under the applicable deadlines; and
-6. decodes the first response choice and token usage.
+3. checks extra-parameter keys against empty and reserved names, then merges
+   their values into the payload;
+4. composes `/chat/completions` through parsed URL path operations;
+5. encodes the complete body once on success. If encoding fails, extra values
+   are checked individually to retain invalid-request classification; otherwise
+   the complete-request failure remains a request failure;
+6. resolves authentication and freezes the URL, body, and optional credential.
+
+Each execution of that prepared request creates a fresh HTTP request, body
+reader, and attempt timeout context. It closes the response body and returns
+the HTTP status, decoded response or error, reported token values and presence,
+and any valid provider `Retry-After` not-before time. A strictly decoded
+successful envelope can retain usage even if its first choice or content is
+missing. Invalid framing and non-success error bodies supply no usage. The
+ordinary `Generate` method prepares and executes once and returns no partial
+response on failure.
 
 `internal/llm` owns the set of reserved OpenAI-compatible request fields used
 when validating extra parameters. Backend registration consumes the same rule
 without making the model client depend on registry configuration.
 
-The implementation has no retry loop, tool-call support, provider catalog,
-inbound HTTP behavior, or durable session store.
+This client does not schedule retries; the [generation coordinator](generation.md)
+does so around its prepared attempts. The client has no tool-call support,
+provider catalog, inbound HTTP behavior, or durable session store.
+
+## Prepared Generation
+
+For [`RunPrepared`](../../engine.go), the runner supplies the model client with
+the target, rendered messages, and structured-output constraint retained by
+executable preparation. Execution does not reopen or rerender consumer
+sources.
+
+Before backend admission, the runner rechecks a frozen credential
+environment-variable name only when the target explicitly requires a
+credential. The handle does not retain the environment value; the model client
+resolves the value visible when generation begins. For optional sources with no
+usable value, the built-in client omits `Authorization` and continues to the
+provider. A direct request key remains in private execution state only until
+the claimed execution finishes or an unclaimed handle is discarded. Exact
+public ownership and redaction semantics belong to the
+[`PreparedExecution` GoDoc](../../prepared_execution.go).
 
 ## Failure Categories
 
 The package preserves distinct error identities for invalid client
 configuration, invalid generation requests, request execution failures,
 non-success provider statuses, and malformed successful responses. Provider
-response bodies are not included in non-success errors.
+response bodies are never exposed in raw form through non-success errors.
 
-Caller cancellation and deadline failures during the outbound request are
-reported as request execution failures. The runner classifies these identities
-without depending on HTTP status mapping.
+Invalid nonempty configured endpoints are configuration failures. A missing or
+invalid final selected endpoint is an invalid generation request and is
+rejected before transport.
+
+Authentication resolves a trimmed direct key before a trimmed configured
+environment value. Optional missing, empty, or whitespace-only sources do not
+block transport and produce no `Authorization` header. An explicitly required
+target with no usable source is rejected before transport with the existing
+invalid-request diagnostics.
+
+Successful response bodies have a fixed 16 MiB limit enforced by declared
+length and by reading at most one byte beyond the boundary. The decoder accepts
+exactly one JSON object plus trailing whitespace and EOF. Size overflow,
+truncation, malformed JSON, trailing data, and a second value are malformed
+responses with no partial result or provider content in the error. Every body
+is closed, and an unbounded oversized stream is not drained.
+
+After framing succeeds, the first choice must contain an explicitly present
+string `message.content`. The string is returned exactly, including empty or
+whitespace-only content. Missing choices, missing or `null` content, and
+non-string content are malformed responses. Output validation and correction
+eligibility remain outside this package.
+
+Successful-envelope usage is decoded with independent presence bits for each
+token field: omission and `null` are unknown, while a reported integer zero is
+known. Strictly decoded usage can still be observed if later choice/content
+checks fail; invalid framing or a non-success error body contributes none.
+Existing successful-result accounting remains value-based.
+
+For a non-success response, `ProviderHTTPError` retains the HTTP status and
+only normalized detail from the bounded recognized envelope. It retains
+`ErrUnexpectedStatus` through unwrapping. The client owns response closure;
+its bounded reader and parser never close or drain a body themselves. The root
+facade converts this concrete internal error into the public
+[`GenerationError`](../../generation_error.go), while arbitrary injected-client
+errors continue through the ordinary generation-error mapping unchanged.
+
+An `http.Client.Do` failure is represented by a redacting multi-cause error:
+the package request-failure sentinel and the exact returned transport error are
+both available through `errors.Is` and `errors.As`, while the rendered text
+does not expose the endpoint, headers, request content, credential, transport
+detail, or provider body. Caller cancellation retains `context.Canceled`;
+caller deadlines, generation deadlines, and whole-request client timeouts
+retain `context.DeadlineExceeded`. The runner adds its generation category
+without discarding those identities or depending on HTTP status mapping.
 
 ## Test Ownership
 
@@ -60,7 +135,15 @@ The
 [OpenAI-compatible client tests](../../internal/llm/openai_compatible_client_test.go)
 own configuration, client cloning, deterministic deadline precedence,
 authentication, request and response mapping, malformed data, error identity,
-cancellation, and response-body suppression. The root transport contract test
-also verifies that resolved backend settings reach this client without
-serializing backend identity. All use local test servers or test transports;
-the default suite makes no live or paid provider requests.
+cancellation, endpoint selection and composition, pre-transport rejection, and
+bounded single-document successful-response framing, closure, and
+response-body suppression. The focused
+[provider HTTP error tests](../../internal/llm/provider_http_error_test.go)
+own envelope parsing, normalization, and bounded-reader cases; their
+[transport tests](../../internal/llm/provider_http_error_transport_test.go)
+own non-success response closure and integration. Root transport contract tests
+own public `GenerationError` conversion, while also verifying that resolved
+backend settings reach this client without serializing backend identity and
+that ordinary-run cancellation retains its public generation and context
+identities. All use local test servers or controlled test transports; the
+default suite makes no live or paid provider requests.

docs/internal/observation.md

@@ -0,0 +1,48 @@
+# Internal Execution Observation
+
+`internal/observation` owns one invocation's run ID, presence-aware token
+aggregation, measured phase totals, and synchronous metadata event delivery.
+Its event values and callback type live in `internal/domain` so internal
+requests can carry them without depending on the recorder package. The
+recorder supplies run IDs to the runner and receives attempt and
+retry-delay measurements from the generation coordinator. Ordinary `Run`
+lifecycle instrumentation spans the facade and runner boundaries.
+Prepared execution starts only after its private payload is atomically claimed;
+the facade receives that invocation's recorder privately and emits the terminal
+event after public error mapping. Failed claims and preparation emit nothing.
+
+The facade masks inherited recorder and prepared-claim state when entering
+preparation or execution. Delegated calls from an injected collaborator retain
+the caller's cancellation, deadline, and unrelated context values, but do not
+reuse the enclosing invocation's accounting. Only the new request's observer
+(or the claimed handle's retained observer) enables its observation lifecycle.
+
+A recorder keeps only counters, resolved identity fields, and the selected
+callback. It has no global history, queue, goroutine, or mutex held across a
+callback. The caller supplies normalized failure metadata; the recorder never
+parses an error string or receives provider bodies, credentials, headers, or
+other content. Events are values with no mutable containers. Callback panics
+propagate to the caller.
+Per-operation callbacks are ordered and synchronous; different operations may
+invoke the same callback concurrently. The recorder marks each measured phase's
+end before delivery, so callback work does not inflate provider duration.
+Terminal phase durations sum measured phases, while wall duration may also
+include preparation, validation, and callbacks. Panics do not promise a
+terminal event.
+
+Attempt usage is known when its explicit presence bit is true. For legacy
+injected clients without presence metadata, nonzero counts are known and
+zeros are unknown. Aggregate values sum known contributions; `Known` means
+at least one reported value, while `Complete` means every dispatched attempt
+reported the field. Zero dispatched attempts produce known, complete zero
+totals. Non-dispatched acquisitions contribute capacity wait but no usage or
+generation duration. Terminal totals count dispatched attempts, transient
+attempts above the first in a generation, and repair generations with at least
+one dispatch. A terminal event is an aggregate and must not be added to its
+attempt events.
+
+The [recorder tests](../../internal/observation/recorder_test.go) own usage
+presence, phase sums, event ordering, empty-run accounting, identity snapshots,
+and callback panic behavior. The [generation coordinator](generation.md) owns
+attempt and retry-delay measurements and emits their completion events after
+releasing active permits.

docs/internal/overview.md

@@ -11,25 +11,30 @@ contributor workflow and validation.
 
 | Component | Implemented responsibility | References |
 | --- | --- | --- |
-| Root `promptkit` package | Provides the supported engine facade, source, backend-registration, and injection options, public request and result values, profile construction, extension interfaces, value conversion, redacted formatting, public error mapping, and engine-local assembly. | [Package GoDoc](../../doc.go), [backend API](../../backends.go), [engine assembly](../../engine.go) |
-| `examples/go-library/prepare` | Demonstrates an offline downstream consumer using a prompt file, in-memory profile, inline input, and `Prepare`. It is not a public library package. | [Example program](../../examples/go-library/prepare/main.go) |
-| `examples/go-library/run` | Demonstrates an offline downstream consumer using a prompt file, in-memory profile, inline input, an injected deterministic model client, and `Run`. It is not a public library package. | [Example program](../../examples/go-library/run/main.go) |
-| `internal/backend` | Constructs each engine's immutable registry from the built-in OpenRouter definition and consumer additions, validates and defensively copies definitions through the shared JSON-value package, and consumes the LLM-owned OpenAI-compatible reserved request-field rule. | [Backend registry](../../internal/backend/registry.go) |
-| `internal/capacity` | Owns engine-local bounded run admission and FIFO model-generation permits for limited backend IDs, including cancellation-safe waiter removal and client wrapping. | [Internal capacity management](capacity.md) |
-| `internal/domain` | Defines internal framework values for requests, artifacts, prompt definitions, profiles, execution targets, rendering, generation, and validation. | [Domain declarations](../../internal/domain/domain.go) |
+| Root `promptkit` package | Provides the supported engine facade, ConfigKit source helpers, profile and backend mapping conveniences, injection options, public requests and results, prompt/profile/requirement inspection and identity listing, concrete-request validation, observation values, opaque prepared-execution handles, value conversion, redacted formatting, and typed capacity and generation error mapping. | [Package GoDoc](../../doc.go), [observations](../../observations.go), [prepared execution](../../prepared_execution.go), [backend API](../../backends.go), [engine facade](../../engine.go), [identity listing](../../listing.go), [requirement facade](../../requirements.go) |
+| `examples/go-library/prepare` | Demonstrates offline preparation and explicit configuration inspection with a prompt file, programmatic profile, and inline input. It is not a public library package. | [Example program](../../examples/go-library/prepare/main.go) |
+| `examples/go-library/run` | Demonstrates an offline downstream consumer using a prompt file, programmatic profile, inline input, an injected deterministic model client, and `Run`. It is not a public library package. | [Example program](../../examples/go-library/run/main.go) |
+| `examples/go-library/observe` | Demonstrates offline built-in HTTP retry and synchronous metadata observation across structured proposal and assessment runs using embedded sources, two profiles on one finite backend, and an application decision. It is not a public library package. | [Example program](../../examples/go-library/observe/main.go) |
+| `internal/backend` | Holds a defensive, immutable lookup and capacity-policy index for final backend definitions validated by the configuration adapter. | [Backend registry](../../internal/backend/registry.go) |
+| `internal/catalog` | Selects released keyed catalog assets and PromptKit-owned profile definitions as ConfigKit underlays; effective validation belongs to the configuration loader. | [Catalog adapter](../../internal/catalog/catalog.go), [internal sources](sources.md#profiles-and-built-ins) |
+| `internal/configuration` | Loads keyed backend, profile, and prompt configuration through ConfigKit; records effective field presence, validates final values, captures indexed prompt definitions, referenced message content, compiled templates, input metadata snapshots, sorted identity-only profile and prompt inventories, backend and presence-aware profile snapshots, and an explicit read-only inspection view. | [Typed loader](../../internal/configuration/load.go), [identity listing](../../internal/configuration/listing.go), [internal sources](sources.md#typed-configuration-loading) |
+| `internal/capacity` | Owns engine-local bounded execution admission and FIFO model-generation permits for limited backend IDs, including cancellation-safe waiter removal and client wrapping. | [Internal capacity management](capacity.md) |
+| `internal/domain` | Defines internal framework values for requests, artifacts, prompt definitions, requirement inspection, profiles, execution targets, rendering, generation, and validation, and owns source-neutral invariants for shared execution settings, OpenAI-compatible base endpoints, session identifiers, and output contracts. Source parsing, required fields, other source-specific normalization, defaulting, and boundary-specific error classification remain with their callers. | [Domain declarations](../../internal/domain/domain.go), [requirement values](../../internal/domain/requirements.go), [endpoint invariant](../../internal/domain/endpoint.go) |
+| `internal/engine` | Assembles selected catalog and caller configuration into one snapshot, then constructs capacity, rendering, validation, generation, repair, and the use-case runner for each engine. It delegates identity listing directly to the captured configuration inventory rather than the runner. | [Engine assembly](../../internal/engine/engine.go), [internal sources](sources.md#profiles-and-built-ins), [internal runner](runner.md) |
+| `internal/generation` | Coordinates frozen built-in HTTP attempts and one-shot injected calls through active backend permits, including bounded status retries, cancellable backoff, and attempt observations. | [Internal generation coordination](generation.md) |
 | `internal/defaults` | Defines application-neutral framework constants and constructs the default execution target. It contains no CLI, server, or inbound HTTP limits. | [Framework defaults](../../internal/defaults/defaults.go) |
 | `internal/filecatalog` | Provides deterministic YAML discovery and path helpers for operating-system filesystems and `fs.FS` sources. | [File catalog](../../internal/filecatalog/catalog.go) |
-| `internal/jsonvalue` | Validates and deeply copies JSON-compatible extra-parameter trees while preserving supported concrete value types. | [JSON values](../../internal/jsonvalue/jsonvalue.go) |
-| `internal/promptdef` | Loads strictly decoded, validated prompt definitions from filesystem and `fs.FS` sources, including version selection and contained file-backed message content. | [Framework formats](../formats.md), [prompt-definition repository](../../internal/promptdef/filesystem_repository.go) |
-| `internal/profile` | Loads strictly decoded, validated execution profiles, including backend selection, from filesystem and `fs.FS` sources and composes repositories with error-preserving fallback. | [Framework formats](../formats.md), [profile repositories](../../internal/profile/filesystem_repository.go) |
-| `internal/profile/builtin` | Embeds the built-in profile catalog, whose entries select OpenRouter, and combines it with an optional primary repository. | [Built-in catalog](../formats.md#built-in-profile-catalog), [repository](../../internal/profile/builtin/repository.go) |
-| `internal/prompt` | Renders prompt messages from Go templates with artifact, variable, session, and cache-control data. | [Go-template renderer](../../internal/prompt/go_renderer.go) |
+| `internal/jsonvalue` | Validates and deeply copies bounded JSON-compatible extra-parameter and prepared-schema trees while preserving supported concrete value types and rejecting cycles or excessive depth and work. | [JSON values](../../internal/jsonvalue/jsonvalue.go) |
+| `internal/promptdef` | Provides a narrow captured-prompt lookup contract, not-found identity, and execution-ready definition hash. | [Prompt lookup](../../internal/promptdef/repository.go), [framework formats](../formats.md) |
+| `internal/profile` | Provides a narrow captured-profile lookup contract and not-found identity. | [Profile lookup](../../internal/profile/repository.go), [framework formats](../formats.md) |
+| `internal/prompt` | Renders prompt messages from Go templates with artifact, variable, session, and cache-control data, and conservatively enumerates dependencies from captured template syntax. | [Go-template renderer](../../internal/prompt/go_renderer.go), [requirement discovery](../../internal/prompt/requirements.go) |
 | `internal/artifact` | Resolves ordinary inline and unrestricted caller-selected file references into copied artifacts with metadata and hashes. | [Internal sources and validation](sources.md) |
-| `internal/validate` | Validates basic, JSON, and JSON Schema output using operating-system filesystem or `fs.FS` schema sources. | [Framework formats](../formats.md#schemas), [internal sources and validation](sources.md) |
-| `internal/llm` | Defines the internal generation boundary and implements outbound OpenAI-compatible chat requests from resolved execution targets, including response decoding, authentication, deadline handling, and ownership of the OpenAI-compatible reserved request-field policy. | [Internal model client](llm.md) |
-| `internal/usecase` | Resolves backend, profile, and request settings and coordinates preparation and execution across internal sources, rendering, artifact loading, generation, validation, and optional repair. | [Internal runner](runner.md) |
+| `internal/validate` | Validates basic, JSON, and JSON Schema output using operating-system filesystem or `fs.FS` schema sources and creates operation-local validation plans with canonical contained schema resources. | [Framework formats](../formats.md#schemas), [internal sources and validation](sources.md) |
+| `internal/llm` | Defines the internal generation boundary and implements outbound OpenAI-compatible chat requests from resolved execution targets, including bounded structured non-success response decoding, successful-response decoding, authentication, deadline handling, and ownership of the OpenAI-compatible reserved request-field policy. | [Internal model client](llm.md) |
+| `internal/observation` | Owns invocation-local run IDs, presence-aware usage and phase accounting, and synchronous metadata event delivery for ordinary and claimed prepared runs and generation attempts. | [Internal execution observation](observation.md) |
+| `internal/usecase` | Selects captured prompts and profiles, enumerates requirements from captured prompt syntax, applies request overrides and completion checks, and coordinates ordinary and single-claim prepared execution across rendering, artifact loading, operation-local validation plans, generation, capacity, and bounded repair. | [Internal runner](runner.md), [requirement inspection](../../internal/usecase/requirements_inspection.go), [prepared-execution implementation](../../internal/usecase/prepared_execution.go) |
 
-The root package assembles these internal components without exposing their
+The internal engine assembles these internal components without exposing their
 representations. Consumers depend only on the root facade.
 
 ## Maintenance

docs/internal/runner.md

@@ -4,30 +4,72 @@
 
 This document describes Promptkit's implemented internal orchestration. The
 [architecture policy](../policy/architecture.md) owns dependency and consumer
-boundaries. The [source and validation document](sources.md) owns repository,
+boundaries. The [source and validation document](sources.md) owns configuration,
 artifact, rendering, and validation behavior, while the
 [model-client document](llm.md) owns generation behavior and failure
 categories.
 
-The runner remains under `internal/usecase` and is assembled by the root
-Promptkit engine. Its concrete type is not part of the public API.
+The runner remains under `internal/usecase` and is assembled by
+`internal/engine`. Its concrete type is not part of the public API.
 The [framework format reference](../formats.md) owns prompt, profile, schema,
 and override semantics consumed by the runner.
 
 ## Collaborators
 
-`Runner` coordinates narrow internal interfaces for prompt definitions,
-profiles, backend resolution, artifacts, rendering, model generation, and
-validation. The root engine supplies one immutable registry containing the
-built-in backend and validated consumer additions, one engine-local run
-admitter, and a model client wrapped by the same capacity manager. Schema
-documents are loaded through the validator's optional schema-loader interface.
-An output repairer can be injected internally, but the ordinary runner
-constructor does not enable one.
+`Runner` coordinates narrow internal interfaces for captured prompt definitions
+and resolved profiles, artifacts, rendering, model generation, and validation.
+The internal engine supplies one engine-local run admitter and a generation
+coordinator using the same capacity manager. Validation
+plans and provider-facing schema metadata come from the validator's preparation
+interface.
+The internal engine supplies one default output repairer through the explicit
+runner constructor, using the same coordinator as initial
+generation. The no-repair runner constructor remains available for focused
+internal callers and tests.
 
 Each invocation carries its state in request, prepared-run, and result values.
 The runner has no durable run or session store.
 
+## Shared Prompt Selection
+
+Ordinary preparation selects the full prompt definition and construction-time
+hash. It retains its early request-ID check before direct-session
+normalization. Referenced message content and static template syntax were
+captured when the engine was built.
+
+`InspectPrompt` selects captured metadata through an optional repository
+capability, falling back to the full definition and existing hash resolution
+for internal repositories without it. It returns owned metadata without
+rendering templates or touching profile, artifact, schema, renderer, validator,
+admission, or model collaborators. The root
+[`Engine.InspectPrompt`](../../engine.go) GoDoc owns the public operation's
+exact contract.
+
+`InspectRequirements` uses the same metadata selection to resolve the
+request-profile/default identity, then obtains the exact captured requirement
+snapshot by the selected prompt ID and version. The analyzer receives only
+those identity fields rather than an execution definition. It analyzes the
+snapshot with the supplied caller variables and normalized direct-session
+bypass; it does not resolve target readiness or enter the preparation
+pipeline. Missing snapshot capability is a prompt-load failure rather than a
+source fallback. The root
+[`Engine.InspectRequirements`](../../requirements.go) GoDoc owns the public
+operation's exact contract.
+
+## Shared Profile Selection
+
+The runner uses one profile-selection and target-resolution boundary for
+ordinary preparation and exact profile inspection. Preparation first selects a
+request profile or a prompt default; inspection begins with its required
+explicit profile ID. Both use the captured, presence-aware target constructed
+from the final effective profile and backend configuration. Request overrides
+apply afterward; preparation checks endpoint and model completeness only after
+those overrides. Inspection can report a partial profile without requiring a
+runnable target. Neither path checks credential availability or performs prompt,
+artifact, schema, rendering, admission, or model-client work. The root
+[`Engine.InspectProfile`](../../engine.go) GoDoc owns the public operation's
+exact contract.
+
 ## Shared Preparation Pipeline
 
 `Prepare` and `Run` share one private preparation pipeline split at the point
@@ -35,23 +77,25 @@ where a run can be assigned to its selected backend pool. The resolution phase
 performs only the work needed to validate routing and admission:
 
 1. validate the required prompt selection and normalize any direct session ID;
-2. load the prompt definition and hash the original definition;
+2. select the captured prompt definition and hash;
 3. select the request profile or the prompt's default profile;
-4. resolve the profile's backend ID, when present;
-5. resolve application-neutral defaults, backend defaults, profile values,
-   and explicit request overrides in that order;
+4. retain the profile's selected backend identity, when present;
+5. apply explicit request overrides to its captured backend-completed target;
 6. validate endpoint, model, numeric overrides, and credential requirements;
-7. resolve the effective output contract without loading its schema; and
+7. resolve and validate the effective output contract without loading its
+   schema; and
 8. retain the definition, source identities, effective settings, output
    contract, and preparation start time in invocation-local state.
 
 The completion phase consumes that state without reloading the prompt,
 profile, or backend:
 
-1. load structured-output schema metadata when required;
+1. create one operation-local validation plan and derive structured-output
+   schema metadata from it when required;
 2. load and hash input artifacts;
 3. render messages and the prompt-defined session;
-4. apply any direct session ID;
+4. apply any direct session ID, then append already-normalized request messages
+   after the rendered definition messages;
 5. hash the effective rendered prompt; and
 6. construct the prepared value and preparation timing.
 
@@ -59,29 +103,44 @@ profile, or backend:
 `Run` performs backend admission between the phases. This structure preserves
 one execution-precedence and error-ordering implementation while allowing a
 full backend pool to reject work before expensive schema, artifact, and
-rendering operations.
+rendering operations. `Prepare` discards the plan after returning its public
+metadata. `Run` retains the plan through initial and repaired-output validation
+and discards it when the operation ends. Prepared execution transfers the
+completed preparation's owned message, input-hash, and copied-schema values
+into its private payload, while deep-copying mutable provider extras and
+creating an independent details snapshot. Claim completion or discard clears
+the execution payload without removing the details snapshot.
+
+The root facade's `Engine.ValidateRequest` uses the same complete preparation
+path as `Prepare` and discards the resulting internal prepared value without
+converting it into a public snapshot. It does not add a runner operation or
+validator component; request conversion and public error mapping remain at the
+facade boundary.
 
 Pointer-based numeric overrides preserve an explicit zero. Invalid negative or
 out-of-range values fail as invalid requests. Endpoint overrides do not change
 the selected backend identity. Non-empty extra-parameter maps replace whole
 lower-precedence maps. A direct API key takes precedence over environment
-lookup; otherwise request, profile, and backend environment-variable names
-apply in that order. A profile requiring a direct key clears an inherited
+lookup; otherwise request, explicit profile, and backend environment-variable
+names apply in that order. A profile requiring a direct key suppresses an inherited
 backend environment name unless the request supplies its own name. Secret
 values remain excluded from serialized metadata.
 
 Reasoning overrides are tri-state: nil inherits the profile, a pointer to a
 nonblank string trims and replaces it, and a pointer to a blank string clears
-it. A nonblank direct session is normalized before source loading, bypasses
+it. A nonblank direct session is normalized before prompt selection, bypasses
 the prompt session template, and is applied after ordinary message rendering.
 A blank direct value retains prompt-template behavior. The runner clears the
 template only on a value copy of the definition, so the definition hash always
-describes the original source while the rendered-prompt hash includes the
-effective direct or rendered session.
+describes captured effective content while the rendered-prompt hash includes the
+effective direct or rendered session and the complete effective message
+sequence. The rendered-prompt hash uses a versioned, length-framed SHA-256
+encoding of that session, every role and content value, and cache-control
+presence and values; its hexadecimal value is opaque.
 
-The registry is read-only after engine construction. Concurrent `Prepare` and
-`Run` calls resolve independent defensive backend values and keep all
-invocation state local.
+The configuration snapshot is read-only after engine construction. Concurrent
+`Prepare` and `Run` calls copy their resolved targets and keep all invocation
+state local.
 
 ## Run Flow
 
@@ -90,31 +149,67 @@ its `RunAdmitter` to reserve capacity for the effective backend ID. A nil
 admitter is an internal unlimited fallback. After successful admission, `Run`
 immediately defers the returned release function, performs the completion
 phase, makes one initial generation call, builds the named output artifact,
-and validates that artifact. Invalid generated content remains a validation
-result; an inability to perform validation is an operational error.
+and validates that artifact with the plan compiled during completion. Invalid
+generated content remains a validation result; an inability to generate or
+perform validation is an operational error.
+
+When the public facade installs an invocation-local observation recorder,
+ordinary `Run` reuses its run ID and carries it to the generation coordinator
+through context. Prompt and profile identities are enriched only after their
+respective resolutions succeed, before admission. The facade starts observation
+before request conversion and emits its terminal event after public error
+mapping and after the runner's deferred admission release. Preparation and
+prepared execution follows the same terminal mapping after a successful claim;
+invalid or discarded handles do not acquire an observer.
+
+Validation preparation and execution honor cancellation at every
+Promptkit-controlled boundary and do not publish a partial plan or result.
+Schema reads are bounded and context-checked between chunks; JSON decoding,
+schema compilation, and schema execution are checked immediately before and
+after their synchronous calls. Promptkit does not move arbitrary filesystem or
+JSON Schema work to background goroutines, so an already-blocked dependency
+method must return before cancellation can take precedence over its outcome.
 
 The admission lease covers completion-phase preparation, initial generation,
 validation, every repair, and every exit. It bounds accepted work without
 serializing preparation or validation behind the active-generation limit.
-The wrapped model client separately acquires a FIFO active permit only around
-each actual generation call.
+The coordinator separately acquires a FIFO active permit only around each
+actual model attempt. It releases the permit before a transient retry wait.
 
-When an internal repairer is present, a JSON or JSON Schema content failure can
-trigger bounded repair attempts. Repair receives the effective execution
-target and session ID, validation errors, prior output, and structured-output
-specification. The default repairer uses the same wrapped client as initial
+After a failed `basic`, JSON, or JSON Schema validation with a positive frozen
+budget, the installed repairer can make a bounded corrective call. Each request
+starts with a fresh copy of the complete effective message sequence (the
+configured prefix followed by the request suffix), includes only the latest
+nonempty candidate as an assistant message, and appends one corrective user
+message. Empty candidates omit that assistant message. The
+correction carries validation diagnostics as JSON data bounded to 64 KiB; the
+full diagnostics remain in the validation result.
+
+Repair receives the effective execution target, explicit numeric-presence bits,
+credential, backend identity, session ID, structured-output specification, and
+frozen transient retry budget.
+The same request constructor supplies those common fields to initial and repair
+generation. The default repairer uses the same coordinator as initial
 generation, so each repair reacquires the selected backend's active permit
 while remaining inside its original admission lease. Repair never performs a
-second bounded admission. This capability remains internal and is not a public
-option.
+second bounded admission, and repaired outputs use the operation's existing
+validation plan. The runner stops at the first valid candidate, sums completed
+generation usage, reports calls actually started, and returns the final failed
+validation result on exhaustion. A repair generation failure follows the
+ordinary generation-error category rather than becoming a validation error.
 
 A successful result includes the output artifact and raw output, validation
 state, effective session ID, prompt and rendered-prompt hashes, selected
 profile and backend, effective settings, input hashes, token usage, a generated
 run identifier, and UTC timing. The same effective session reaches initial
 generation and any repair attempt through the rendered prompt. The same
-effective target, including backend identity, reaches generation and any
-repair attempt.
+effective target and presence metadata, including backend identity and direct
+credential during execution, reaches generation and every repair attempt.
+Result usage is the field-wise sum of all five usage values from the initial
+response and every completed repair response. Final raw output, artifact, and
+validation state still come from the last candidate. A repair error returns no
+partial run result or partial result usage, although the invocation-local
+observer retains completed-attempt usage and reports it in the terminal event.
 
 ## Failure Categories
 
@@ -124,19 +219,22 @@ validation failures. Wrapping preserves the package identities mapped by the
 public facade and retains collaborator identities where they are part of the
 internal contract.
 
-Admission capacity exhaustion retains the internal capacity identity and adds
-the selected backend ID as context. It is not recategorized as an invalid
-request or generation failure, and no partial result is returned. A context
-already done at admission retains its context identity directly. Cancellation
-while waiting for an active generation permit prevents client invocation when
-it wins the grant race; the model-client boundary then preserves the context
-error through the generation-failure category. Deferred release restores the
-admission lease on preparation, generation, validation, repair, and
-cancellation failures.
+Admission capacity exhaustion retains the internal capacity identity. At the
+use-case boundary, the runner attaches the selected backend ID in an internal
+typed error, and the root facade copies that value into the public
+[`CapacityError`](../../capacity_error.go) without parsing diagnostic text. It
+is not recategorized as an invalid request or generation failure, and no
+partial result is returned. A context already done at admission retains its
+context identity directly. Cancellation while waiting for an active generation
+permit prevents client invocation when it wins the grant race; the model-client
+boundary then preserves the context error through the generation-failure
+category. Deferred release restores the admission lease on preparation,
+generation, validation, repair, and cancellation failures.
 
 Other context cancellation propagates through the invoked collaborator and is
-classified by the owning operation.
-An overlong direct session is an invalid request before source loading, while
+classified by the owning operation. In particular, cancellation observed by
+validation retains the context identity through the validation error category.
+An overlong direct session is an invalid request before prompt selection, while
 an invalid or overlong prompt session template remains a prompt-render failure.
 An unknown selected backend, or a selected backend with no configured resolver,
 is classified as a profile-load failure.
@@ -147,10 +245,12 @@ The [runner tests](../../internal/usecase/runner_test.go) own preparation order,
 selection and override precedence, the two-phase boundary, early admission,
 lease lifetime and release, direct-session resolution, schema-before-generation
 behavior, hashing, generation and validation outcomes, backend propagation,
-bounded repair, shared initial/repair capacity, credentials and redaction,
-error categories, artifact metadata, usage, and timing. The
-[capacity subsystem document](capacity.md) identifies the focused pool,
-waiter, and wrapped-client tests.
+bounded repair progression, initial/repair request parity, cumulative usage,
+shared initial/repair capacity, credentials and redaction, error categories,
+artifact metadata, and timing. The
+[capacity subsystem document](capacity.md) identifies the focused pool and
+waiter tests. The [generation document](generation.md) owns attempt and retry
+coordination tests.
 
 Changes to orchestration should continue to use the existing package
 interfaces, keep request state local to an invocation, and preserve the shared

docs/internal/sources.md

@@ -2,50 +2,160 @@
 
 ## Purpose
 
-This document describes Promptkit's implemented internal source, artifact,
+This document describes Promptkit's implemented configuration, artifact,
 rendering, and output-validation behavior. The
-[architecture policy](../policy/architecture.md) owns the library boundary and
-dependency rules. None of these internal packages is a supported consumer API,
-and the root engine assembles them behind its public source options and values.
-The [framework format reference](../formats.md) owns the exact file fields,
-validation modes, built-in catalog, and source precedence.
+[architecture policy](../policy/architecture.md) owns dependency rules, while
+the [framework format reference](../formats.md) owns fields, defaults, and
+source semantics. Internal packages are not consumer APIs.
+
+## Typed Configuration Loading
+
+`internal/configuration` is the single adapter from ConfigKit's final
+effective configuration to execution-owned definitions. It performs one
+layered load, captures effective presence (including explicit null), records
+normalization and defaults through resolution, and validates all effective
+backend, profile, and prompt values. Replaced lower-layer domain values are
+not separately validated. It builds an immutable backend index and
+presence-aware profile targets, applying final backend defaults only after
+profile inheritance. It retains ConfigKit's result privately for explicit
+inspection, not for per-run resolution. The runtime/domain packages do not
+depend on ConfigKit.
+
+The `ConfigurationInspection.Diff` facade method compares two retained results
+through a private wrapper and a narrow `internal/configuration.Diff` bridge.
+The wrapper keeps `Result[document]` inside the adapter; the bridge delegates
+once to ConfigKit v0.7.0 and returns its public difference records. Comparison
+uses only frozen effective values and does not reopen sources, resolve runtime
+credentials, or enter the execution path. ConfigKit applies sensitivity from
+both results; reports may still contain ordinary application data. See the
+[consumer comparison guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration)
+and ConfigKit's [v0.7.0 comparison contract](https://gitea.maximumdirect.net/eric/ConfigKit/src/tag/v0.7.0/docs/config.md#effective-configuration-diffs).
+
+After prompt materialization and effective-profile snapshot construction, the
+adapter captures lexically ordered, identity-only inventories from the final
+prompt and indexed-profile maps. Its listing methods return caller-owned copies
+of those inventories without consulting ConfigKit or copying definitions.
+
+The root `DirectoryInputs` and `FilesystemInputs` helpers discover sorted YAML
+files under an explicit root, provide resource context, and leave imports
+disabled. Each returned slice belongs to one strict-composition group; file
+order does not confer priority. The engine prepends selected catalog and
+Promptkit-owned definitions, then caller underlays and optional main. The
+[configuration tests](../../internal/configuration/load_test.go) exercise the
+real loader and final-value handoff.
 
 ## Prompt Definitions
 
-`internal/promptdef` discovers YAML deterministically, decodes and validates
-definitions, selects an ID and optional version, and resolves file-backed
-message content within the selected operating-system or `fs.FS` source.
+The configuration adapter indexes each prompt by exact ID and version,
+materializes file-backed message content using the exact effective field's
+origin and resource context, and checks static message/session template syntax
+at construction. It stores captured content, compiled templates, declared
+input metadata, and a definition hash. Alongside the unchanged hashed
+definition, it retains content-type and description presence for each declared
+input and provides an internal requirement snapshot with those declarations
+and the captured parses. Metadata values come from the final normalized
+declaration; absent and null fields remain distinguishable from explicit
+strings, including empty strings. It closes content-file handles before
+returning and retains no source capability for public inspection.
 
-Its package tests own prompt selection, strict decoding, definition validation,
-duplicate detection, and source containment:
-[prompt-definition repository tests](../../internal/promptdef/repository_test.go).
+Role and cache-control normalization is applied to execution-owned message
+copies. ConfigKit v0.7.0 retains effective input message metadata and its
+resource origins; recording a sequence-wide normalization in the retained
+configuration would replace origins of unchanged content-file fields.
+
+The runner selects full definitions from this index for preparation and uses
+captured prompt metadata for `InspectPrompt` and `InspectRequirements`; the
+internal requirement snapshot lookup selects by exact prompt ID and version.
+These paths do not rescan configuration or reopen message files. `InspectPrompt`
+returns declared metadata without rendering or resolving profile, input, or
+schema resources, and its public shape and prompt hashes do not include the
+captured presence state. Runtime rendering still checks request variables and
+input helpers. The [prompt adapter tests](../../internal/configuration/prompts_test.go)
+cover resource origins, context containment, capture, and selection.
+
+The configuration adapter shares exact ID/version selection across its
+internal prompt lookups. The execution snapshot lookup returns an owned prompt
+definition and cached hash; metadata inspection returns owned input metadata
+without copying execution templates. Profile selection likewise has an
+identity-only lookup for consumers that do not need the execution target,
+while the full profile lookup continues to return an owned target and deep copy
+of provider extras. These lookups use captured indexes and do not query
+ConfigKit again.
+
+The `internal/prompt` requirement walker uses the captured parse trees and
+declaration metadata without executing templates or reading artifacts. It
+collects literal `input` names, direct root caller keys, and literal-key
+`index` lookups. For `if` and `with`, it classifies branches when their
+conditions are simple supported values: direct values, literals, negation, or
+two-string equality/inequality. It uses Go template truthiness, so a supplied
+nonempty string such as `false` is true. Missing decision values and
+content-dependent conditions leave dependency-bearing branches conditional
+and report incomplete discovery; unresolved conditions with no branch
+dependencies do not by themselves make discovery incomplete. Recognizable
+references on known skipped paths remain optional inventory entries with an
+inactive-use explanation, unless another applicable use establishes a
+stronger classification. Skipped expressions are scanned only for
+syntactically recognizable names and are not evaluated. A direct caller value
+or literal-key lookup can also supply an input name, including through simple
+parentheses or a direct pipeline to `input`; names are used exactly as
+supplied. Unresolved names and unsupported dependency-bearing expressions
+produce structured limitations while retaining recognizable references.
+Where an unresolved name or scope could affect existing optional entries, those
+entries are reported as conditional rather than inactive-only or definitively
+optional. Direct root-map `with` forms retain caller-map identity. Known scalar
+values remain separate from unknown scopes; accesses through untracked local
+maps report incomplete discovery rather than silently losing caller keys.
+Unsupported name expressions reuse value analysis's conservative syntax scan
+instead of recursively rescanning the same subtree.
+Invoked named templates are scanned conservatively, and
+dependency-bearing range scopes make discovery incomplete. Uncalled definitions
+and bypassed session templates are not traversed. This internal walker does not
+change rendering behavior.
 
 ## Profiles And Built-Ins
 
-`internal/profile` loads and validates execution profiles from an
-operating-system filesystem or an `fs.FS`. It supports a primary repository
-with fallback only when the primary reports that a profile is absent. Strict
-YAML decoding recognizes the optional `backend` field, trims its value, and
-requires a model plus at least one non-blank backend or endpoint. Loading does
-not check registry membership because the available registry belongs to the
-assembled engine; the runner checks membership during preparation.
+`internal/catalog` selects the released keyed OpenRouter and Rakestrawhome
+assets as one ConfigKit underlay and PromptKit-owned profiles as the next.
+It discovers YAML inputs and supplies passive filesystem context without
+parsing or pre-validating catalog definitions. The configuration loader
+validates only the final effective backend and profile values, including
+consumer overlays of maintained bases or backends. There is no catalog-only
+preflight, reserved source namespace, or replay of backend registration.
 
-`internal/profile/builtin` embeds the maintained built-in profile catalog and
-can place a caller-selected repository ahead of that catalog. Every embedded
-profile selects `openrouter` and inherits its endpoint and credential
-environment-variable name from the built-in backend registry rather than
-repeating those values. Profile behavior is owned by the
-[profile repository tests](../../internal/profile/repository_test.go), while
-catalog completeness, the backend-selection invariant, duplicate IDs, and
-overlay behavior are owned by the
-[built-in repository tests](../../internal/profile/builtin/repository_test.go).
+ConfigKit resolves profile inheritance before the handoff. The configuration
+adapter validates effective references and field values, then builds target
+snapshots with backend defaults and presence/suppression state. A profile can
+remain partial until runtime overrides complete it. `InspectProfile` selects
+that captured target without prompt, artifact, schema, credential lookup, or
+provider work. Preparation applies runtime overrides and then checks target
+completeness. Identity-only profile selection can identify a partial captured
+profile without copying its target or requiring readiness. The
+[target tests](../../internal/configuration/targets_test.go) cover effective
+target interpretation, while the [catalog tests](../../internal/catalog/catalog_test.go)
+cover selected released assets.
+
+The internal engine delegates listing through a narrow identity-inventory
+interface directly to the configuration snapshot. The execution runner has no
+listing responsibility.
 
 ## Ordinary Artifacts
 
-`internal/artifact` resolves inline references and unrestricted,
-caller-selected file paths. It copies content into an artifact, records
-metadata and a content hash, applies a content-type fallback, and honors
-context cancellation.
+`internal/artifact` accepts explicitly typed inline references even when their
+body is empty. It also resolves unrestricted, caller-selected paths only when
+they identify regular operating-system files, checking that condition before
+and after opening the file. It copies content into an artifact, records
+metadata and an opaque content-equality value, and applies a content-type
+fallback.
+
+Regular files are read synchronously in bounded chunks. The opened file's size
+is only an initial-capacity hint, capped at 1 MiB; reads continue to EOF, and
+the returned body determines artifact size and hash. If an overreported hint
+leaves unused capacity greater than the actual body length, the completed body
+is copied to release the excess backing storage, including when the body is
+empty. Cancellation is checked before opening, before and after every read,
+and before publishing the artifact, so a
+canceled read never publishes partial content. The ordinary reader does not
+detach file reads into background goroutines.
 
 This ordinary reader does not implement an inbound HTTP security boundary. In
 particular, it does not constrain files to an application root or impose an
@@ -56,9 +166,22 @@ implemented reader behavior and failures.
 
 ## Rendering
 
-`internal/prompt` renders definition messages as Go templates using named
-artifacts and variables. It carries message roles, session IDs, and cache
-control into the rendered prompt. The
+`internal/prompt` renders captured, pre-parsed definition templates using named
+artifacts and variables. It clones templates to bind request-specific input
+helpers without mutating shared state. Within one render, each referenced
+artifact body is converted to text lazily and cached by input name for reuse
+across the session and every message; the cache is not shared across renders.
+Conversion uses bounded chunks and preserves the artifact bytes exactly.
+
+Template parsing occurs during engine construction. Runtime execution remains
+synchronous. The renderer checks cancellation around execution boundaries,
+between artifact conversion chunks, around each message, and before publishing
+the complete prompt. It cannot interrupt template work already in progress and
+never publishes a partial prompt after observing cancellation. It validates and
+canonicalizes message roles before carrying roles, session IDs, and cache
+control into the rendered prompt. Message and session templates execute into
+per-call string builders, so their rendered strings do not require a final
+buffer-to-string copy. The
 [renderer tests](../../internal/prompt/renderer_test.go) own rendering behavior.
 
 ## Schemas And Output Validation
@@ -66,8 +189,39 @@ control into the rendered prompt. The
 `internal/validate` provides validators backed by an operating-system
 filesystem or an `fs.FS`. Invalid generated content is returned as a validation
 result; inability to load, register, or compile a schema is an operational
-error.
+error. Basic validation checks body emptiness on the byte slice directly
+instead of first making a response-sized string copy.
+
+Every preparation operation creates one operation-local validation plan. None,
+basic, and JSON modes retain the effective output contract without source
+access. JSON Schema mode loads the root document once, resolves and compiles
+each transitive reference, and retains the compiled validator. Schema compiler
+resources use canonical escaped file or private-scheme URLs; loaders decode
+their paths once and enforce the configured source boundary. The
+provider-facing structured-output metadata uses the root document captured by
+the same plan.
+
+Schema preparation and execution remain synchronous. Promptkit checks
+cancellation before and after source resolution, JSON decoding, compilation,
+and validation, and between bounded schema-read chunks. Once cancellation is
+observed it returns the context error without publishing a partial plan or
+validation result, even when a compiler or validator has just returned a
+different error or a successful result. An `fs.FS` method or JSON Schema
+dependency call already in progress cannot be preempted; Promptkit waits for
+that call to return and then gives cancellation precedence. Validation does
+not detach dependency work into background goroutines.
+
+`Prepare` discards its validation plan after returning metadata. `Run` retains
+its plan for initial and repaired-output validation, then discards it with the
+operation. `PrepareExecution` retains the plan in its private frozen payload;
+`RunPrepared` uses that plan without reopening configuration, message, input,
+or schema resources or rerendering the request. A later ordinary `Run` uses
+the same captured configuration but prepares request artifacts and schemas
+afresh.
 
 The [validator tests](../../internal/validate/standard_validator_test.go) own
-basic, JSON, JSON Schema, source resolution, schema loading, compilation, and
-content-failure behavior.
+basic, JSON, JSON Schema, source resolution, schema loading, compilation,
+frozen-reference behavior, content-failure behavior, and the synchronous
+cancellation boundary. Prepared execution
+orchestration is owned by the
+[use-case tests](../../internal/usecase/prepared_execution_test.go).

docs/oss/attribution.md

@@ -0,0 +1,134 @@
+# Open Source Attribution Notices
+
+## Standard Repository Policies
+
+Material: policy documents under `docs/policy/` and the Policy Reading Guide in
+`docs/development.md`, adapted for this repository; root `AGENTS.md`, copied
+without modification from upstream `agents/AGENTS.md`.
+
+The release policy uses upstream commit
+`e833faddf120c789a9b45348edaf75720de16348`; the remaining material uses
+`d84a38ebab26f048cafbe59a2246700c35d29843`.
+
+Copyright (C) 2026 Eric Rakestraw
+
+SPDX-License-Identifier: CC-BY-SA-4.0
+
+License and notices: [CC BY-SA 4.0](licenses/CC-BY-SA-4.0.txt) |
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/COPYRIGHT) |
+[LICENSE](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/LICENSE) |
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/policy/src/commit/e833faddf120c789a9b45348edaf75720de16348/COPYRIGHT) |
+[LICENSE](https://gitea.maximumdirect.net/eric/policy/src/commit/e833faddf120c789a9b45348edaf75720de16348/LICENSE).
+
+[Source](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843).
+
+[Source](https://gitea.maximumdirect.net/eric/policy/src/commit/e833faddf120c789a9b45348edaf75720de16348).
+
+## ConfigKit
+
+Selected module: `gitea.maximumdirect.net/eric/ConfigKit` at `v0.7.1`.
+
+Copyright (C) 2026 Eric Rakestraw
+
+SPDX-License-Identifier: LGPL-3.0-or-later
+
+License and notices: [LGPL 3.0](../../LICENSE) |
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/ConfigKit/src/commit/6fffde4af8f31f99d3ee295c426eb3234cb0fb77/COPYRIGHT) |
+[LICENSE](https://gitea.maximumdirect.net/eric/ConfigKit/src/commit/6fffde4af8f31f99d3ee295c426eb3234cb0fb77/LICENSE).
+
+[Source](https://gitea.maximumdirect.net/eric/ConfigKit/src/commit/6fffde4af8f31f99d3ee295c426eb3234cb0fb77).
+
+## Promptkit OpenRouter Catalog v3
+
+Selected module: `gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v3`
+at `v3.20260930.0`.
+
+Copyright (C) 2026 Eric Rakestraw
+
+SPDX-License-Identifier: MIT
+
+License and notices: [MIT](licenses/MIT.txt) |
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit-backend-openrouter/src/commit/f17d208cb7ad1fb043cda2266cd3011d11b40aa0/COPYRIGHT) |
+[LICENSE](https://gitea.maximumdirect.net/eric/promptkit-backend-openrouter/src/commit/f17d208cb7ad1fb043cda2266cd3011d11b40aa0/LICENSE).
+
+[Source](https://gitea.maximumdirect.net/eric/promptkit-backend-openrouter/src/commit/f17d208cb7ad1fb043cda2266cd3011d11b40aa0).
+
+## Promptkit Rakestrawhome Catalog v3
+
+Selected module: `gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v3`
+at `v3.20260930.0`.
+
+Copyright (C) 2026 Eric Rakestraw
+
+SPDX-License-Identifier: MIT
+
+License and notices: [MIT](licenses/MIT.txt) |
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/src/commit/993443d88c451901f4c917fa00c204fd3b54d28d/COPYRIGHT) |
+[LICENSE](https://gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/src/commit/993443d88c451901f4c917fa00c204fd3b54d28d/LICENSE).
+
+[Source](https://gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/src/commit/993443d88c451901f4c917fa00c204fd3b54d28d).
+
+## jsonschema
+
+Selected module: `github.com/santhosh-tekuri/jsonschema/v6` at `v6.0.2`.
+
+SPDX-License-Identifier: Apache-2.0
+
+License and notices: [Apache 2.0](licenses/Apache-2.0.txt) |
+[LICENSE](https://github.com/santhosh-tekuri/jsonschema/blob/29cbed948d24a04700eb94436416b18a07953b71/LICENSE).
+
+[Source](https://github.com/santhosh-tekuri/jsonschema/tree/29cbed948d24a04700eb94436416b18a07953b71).
+
+## Go Text
+
+Selected module: `golang.org/x/text` at `v0.14.0` (transitive dependency).
+
+Copyright (c) 2009 The Go Authors. All rights reserved.
+
+SPDX-License-Identifier: BSD-3-Clause
+
+License and notices: [BSD 3-Clause](licenses/BSD-3-Clause.txt) |
+[LICENSE](https://github.com/golang/text/blob/6c97a165dd661335ff7bce6104a008558123c353/LICENSE) |
+[PATENTS](https://github.com/golang/text/blob/6c97a165dd661335ff7bce6104a008558123c353/PATENTS).
+
+[Source](https://github.com/golang/text/tree/6c97a165dd661335ff7bce6104a008558123c353).
+
+## Go YAML v3
+
+Selected module: `gopkg.in/yaml.v3` at `v3.0.1` (transitive dependency).
+
+Copyright (c) 2006-2010 Kirill Simonov
+
+Copyright (c) 2006-2011 Kirill Simonov
+
+Copyright (c) 2011-2019 Canonical Ltd
+
+Copyright 2011-2016 Canonical Ltd.
+
+SPDX-License-Identifier: MIT AND Apache-2.0
+
+License and notices: [MIT](licenses/MIT.txt) |
+[Apache 2.0](licenses/Apache-2.0.txt) |
+[LICENSE](https://github.com/go-yaml/yaml/blob/f6f7691b1fdeb513f56608cd2c32c51f8194bf51/LICENSE) |
+[NOTICE](https://github.com/go-yaml/yaml/blob/f6f7691b1fdeb513f56608cd2c32c51f8194bf51/NOTICE).
+
+[Source](https://github.com/go-yaml/yaml/tree/f6f7691b1fdeb513f56608cd2c32c51f8194bf51).
+
+## Go YAML v4
+
+Selected module: `go.yaml.in/yaml/v4` at `v4.0.0-rc.6` (transitive dependency).
+
+Copyright 2006-2010 Kirill Simonov
+
+Copyright 2011-2019 Canonical Ltd
+
+Copyright 2025 The go-yaml Project Contributors
+
+SPDX-License-Identifier: MIT AND Apache-2.0
+
+License and notices: [MIT](licenses/MIT.txt) |
+[Apache 2.0](licenses/Apache-2.0.txt) |
+[LICENSE](https://github.com/yaml/go-yaml/blob/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125/LICENSE) |
+[NOTICE](https://github.com/yaml/go-yaml/blob/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125/NOTICE).
+
+[Source](https://github.com/yaml/go-yaml/tree/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125).

docs/oss/licenses/Apache-2.0.txt

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

docs/oss/licenses/BSD-3-Clause.txt

@@ -0,0 +1,11 @@
+Copyright (c) <year> <owner>.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

docs/oss/licenses/CC-BY-SA-4.0.txt

@@ -0,0 +1,170 @@
+Creative Commons Attribution-ShareAlike 4.0 International
+
+ Creative Commons Corporation (“Creative Commons”) is not a law firm and does not provide legal services or legal advice. Distribution of Creative Commons public licenses does not create a lawyer-client or other relationship. Creative Commons makes its licenses and related information available on an “as-is” basis. Creative Commons gives no warranties regarding its licenses, any material licensed under their terms and conditions, or any related information. Creative Commons disclaims all liability for damages resulting from their use to the fullest extent possible.
+
+Using Creative Commons Public Licenses
+
+Creative Commons public licenses provide a standard set of terms and conditions that creators and other rights holders may use to share original works of authorship and other material subject to copyright and certain other rights specified in the public license below. The following considerations are for informational purposes only, are not exhaustive, and do not form part of our licenses.
+
+Considerations for licensors: Our public licenses are intended for use by those authorized to give the public permission to use material in ways otherwise restricted by copyright and certain other rights. Our licenses are irrevocable. Licensors should read and understand the terms and conditions of the license they choose before applying it. Licensors should also secure all rights necessary before applying our licenses so that the public can reuse the material as expected. Licensors should clearly mark any material not subject to the license. This includes other CC-licensed material, or material used under an exception or limitation to copyright. More considerations for licensors.
+
+Considerations for the public: By using one of our public licenses, a licensor grants the public permission to use the licensed material under specified terms and conditions. If the licensor’s permission is not necessary for any reason–for example, because of any applicable exception or limitation to copyright–then that use is not regulated by the license. Our licenses grant only permissions under copyright and certain other rights that a licensor has authority to grant. Use of the licensed material may still be restricted for other reasons, including because others have copyright or other rights in the material. A licensor may make special requests, such as asking that all changes be marked or described.
+
+Although not required by our licenses, you are encouraged to respect those requests where reasonable. More considerations for the public.
+
+Creative Commons Attribution-ShareAlike 4.0 International Public License
+
+By exercising the Licensed Rights (defined below), You accept and agree to be bound by the terms and conditions of this Creative Commons Attribution-ShareAlike 4.0 International Public License ("Public License"). To the extent this Public License may be interpreted as a contract, You are granted the Licensed Rights in consideration of Your acceptance of these terms and conditions, and the Licensor grants You such rights in consideration of benefits the Licensor receives from making the Licensed Material available under these terms and conditions.
+
+Section 1 – Definitions.
+
+     a.	Adapted Material means material subject to Copyright and Similar Rights that is derived from or based upon the Licensed Material and in which the Licensed Material is translated, altered, arranged, transformed, or otherwise modified in a manner requiring permission under the Copyright and Similar Rights held by the Licensor. For purposes of this Public License, where the Licensed Material is a musical work, performance, or sound recording, Adapted Material is always produced where the Licensed Material is synched in timed relation with a moving image.
+
+     b.	Adapter's License means the license You apply to Your Copyright and Similar Rights in Your contributions to Adapted Material in accordance with the terms and conditions of this Public License.
+
+     c.	BY-SA Compatible License means a license listed at creativecommons.org/compatiblelicenses, approved by Creative Commons as essentially the equivalent of this Public License.
+
+     d.	Copyright and Similar Rights means copyright and/or similar rights closely related to copyright including, without limitation, performance, broadcast, sound recording, and Sui Generis Database Rights, without regard to how the rights are labeled or categorized. For purposes of this Public License, the rights specified in Section 2(b)(1)-(2) are not Copyright and Similar Rights.
+
+     e.	Effective Technological Measures means those measures that, in the absence of proper authority, may not be circumvented under laws fulfilling obligations under Article 11 of the WIPO Copyright Treaty adopted on December 20, 1996, and/or similar international agreements.
+
+     f.	Exceptions and Limitations means fair use, fair dealing, and/or any other exception or limitation to Copyright and Similar Rights that applies to Your use of the Licensed Material.
+
+     g.	License Elements means the license attributes listed in the name of a Creative Commons Public License. The License Elements of this Public License are Attribution and ShareAlike.
+
+     h.	Licensed Material means the artistic or literary work, database, or other material to which the Licensor applied this Public License.
+
+     i.	Licensed Rights means the rights granted to You subject to the terms and conditions of this Public License, which are limited to all Copyright and Similar Rights that apply to Your use of the Licensed Material and that the Licensor has authority to license.
+
+     j.	Licensor means the individual(s) or entity(ies) granting rights under this Public License.
+
+     k.	Share means to provide material to the public by any means or process that requires permission under the Licensed Rights, such as reproduction, public display, public performance, distribution, dissemination, communication, or importation, and to make material available to the public including in ways that members of the public may access the material from a place and at a time individually chosen by them.
+
+     l.	Sui Generis Database Rights means rights other than copyright resulting from Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases, as amended and/or succeeded, as well as other essentially equivalent rights anywhere in the world.
+
+     m.	You means the individual or entity exercising the Licensed Rights under this Public License. Your has a corresponding meaning.
+
+Section 2 – Scope.
+
+     a.	License grant.
+
+          1. Subject to the terms and conditions of this Public License, the Licensor hereby grants You a worldwide, royalty-free, non-sublicensable, non-exclusive, irrevocable license to exercise the Licensed Rights in the Licensed Material to:
+
+               A. reproduce and Share the Licensed Material, in whole or in part; and
+
+               B. produce, reproduce, and Share Adapted Material.
+
+          2. Exceptions and Limitations. For the avoidance of doubt, where Exceptions and Limitations apply to Your use, this Public License does not apply, and You do not need to comply with its terms and conditions.
+
+          3. Term. The term of this Public License is specified in Section 6(a).
+
+          4. Media and formats; technical modifications allowed. The Licensor authorizes You to exercise the Licensed Rights in all media and formats whether now known or hereafter created, and to make technical modifications necessary to do so. The Licensor waives and/or agrees not to assert any right or authority to forbid You from making technical modifications necessary to exercise the Licensed Rights, including technical modifications necessary to circumvent Effective Technological Measures. For purposes of this Public License, simply making modifications authorized by this Section 2(a)(4) never produces Adapted Material.
+
+          5. Downstream recipients.
+
+               A. Offer from the Licensor – Licensed Material. Every recipient of the Licensed Material automatically receives an offer from the Licensor to exercise the Licensed Rights under the terms and conditions of this Public License.
+
+               B. Additional offer from the Licensor – Adapted Material. Every recipient of Adapted Material from You automatically receives an offer from the Licensor to exercise the Licensed Rights in the Adapted Material under the conditions of the Adapter’s License You apply.
+
+               C. No downstream restrictions. You may not offer or impose any additional or different terms or conditions on, or apply any Effective Technological Measures to, the Licensed Material if doing so restricts exercise of the Licensed Rights by any recipient of the Licensed Material.
+
+          6. No endorsement. Nothing in this Public License constitutes or may be construed as permission to assert or imply that You are, or that Your use of the Licensed Material is, connected with, or sponsored, endorsed, or granted official status by, the Licensor or others designated to receive attribution as provided in Section 3(a)(1)(A)(i).
+
+     b.	Other rights.
+
+          1. Moral rights, such as the right of integrity, are not licensed under this Public License, nor are publicity, privacy, and/or other similar personality rights; however, to the extent possible, the Licensor waives and/or agrees not to assert any such rights held by the Licensor to the limited extent necessary to allow You to exercise the Licensed Rights, but not otherwise.
+
+          2. Patent and trademark rights are not licensed under this Public License.
+
+          3. To the extent possible, the Licensor waives any right to collect royalties from You for the exercise of the Licensed Rights, whether directly or through a collecting society under any voluntary or waivable statutory or compulsory licensing scheme. In all other cases the Licensor expressly reserves any right to collect such royalties.
+
+Section 3 – License Conditions.
+
+Your exercise of the Licensed Rights is expressly made subject to the following conditions.
+
+     a.	Attribution.
+
+          1. If You Share the Licensed Material (including in modified form), You must:
+
+               A. retain the following if it is supplied by the Licensor with the Licensed Material:
+
+                    i.	identification of the creator(s) of the Licensed Material and any others designated to receive attribution, in any reasonable manner requested by the Licensor (including by pseudonym if designated);
+
+                    ii.	a copyright notice;
+
+                    iii. a notice that refers to this Public License;
+
+                    iv.	a notice that refers to the disclaimer of warranties;
+
+                    v.	a URI or hyperlink to the Licensed Material to the extent reasonably practicable;
+
+               B. indicate if You modified the Licensed Material and retain an indication of any previous modifications; and
+
+               C. indicate the Licensed Material is licensed under this Public License, and include the text of, or the URI or hyperlink to, this Public License.
+
+          2. You may satisfy the conditions in Section 3(a)(1) in any reasonable manner based on the medium, means, and context in which You Share the Licensed Material. For example, it may be reasonable to satisfy the conditions by providing a URI or hyperlink to a resource that includes the required information.
+
+          3. If requested by the Licensor, You must remove any of the information required by Section 3(a)(1)(A) to the extent reasonably practicable.
+
+     b.	ShareAlike.In addition to the conditions in Section 3(a), if You Share Adapted Material You produce, the following conditions also apply.
+
+          1. The Adapter’s License You apply must be a Creative Commons license with the same License Elements, this version or later, or a BY-SA Compatible License.
+
+          2. You must include the text of, or the URI or hyperlink to, the Adapter's License You apply. You may satisfy this condition in any reasonable manner based on the medium, means, and context in which You Share Adapted Material.
+
+          3. You may not offer or impose any additional or different terms or conditions on, or apply any Effective Technological Measures to, Adapted Material that restrict exercise of the rights granted under the Adapter's License You apply.
+
+Section 4 – Sui Generis Database Rights.
+
+Where the Licensed Rights include Sui Generis Database Rights that apply to Your use of the Licensed Material:
+
+     a.	for the avoidance of doubt, Section 2(a)(1) grants You the right to extract, reuse, reproduce, and Share all or a substantial portion of the contents of the database;
+
+     b.	if You include all or a substantial portion of the database contents in a database in which You have Sui Generis Database Rights, then the database in which You have Sui Generis Database Rights (but not its individual contents) is Adapted Material, including for purposes of Section 3(b); and
+
+     c.	You must comply with the conditions in Section 3(a) if You Share all or a substantial portion of the contents of the database.
+For the avoidance of doubt, this Section 4 supplements and does not replace Your obligations under this Public License where the Licensed Rights include other Copyright and Similar Rights.
+
+Section 5 – Disclaimer of Warranties and Limitation of Liability.
+
+     a.	Unless otherwise separately undertaken by the Licensor, to the extent possible, the Licensor offers the Licensed Material as-is and as-available, and makes no representations or warranties of any kind concerning the Licensed Material, whether express, implied, statutory, or other. This includes, without limitation, warranties of title, merchantability, fitness for a particular purpose, non-infringement, absence of latent or other defects, accuracy, or the presence or absence of errors, whether or not known or discoverable. Where disclaimers of warranties are not allowed in full or in part, this disclaimer may not apply to You.
+
+     b.	To the extent possible, in no event will the Licensor be liable to You on any legal theory (including, without limitation, negligence) or otherwise for any direct, special, indirect, incidental, consequential, punitive, exemplary, or other losses, costs, expenses, or damages arising out of this Public License or use of the Licensed Material, even if the Licensor has been advised of the possibility of such losses, costs, expenses, or damages. Where a limitation of liability is not allowed in full or in part, this limitation may not apply to You.
+
+     c.	The disclaimer of warranties and limitation of liability provided above shall be interpreted in a manner that, to the extent possible, most closely approximates an absolute disclaimer and waiver of all liability.
+
+Section 6 – Term and Termination.
+
+     a.	This Public License applies for the term of the Copyright and Similar Rights licensed here. However, if You fail to comply with this Public License, then Your rights under this Public License terminate automatically.
+
+     b.	Where Your right to use the Licensed Material has terminated under Section 6(a), it reinstates:
+
+          1. automatically as of the date the violation is cured, provided it is cured within 30 days of Your discovery of the violation; or
+
+          2. upon express reinstatement by the Licensor.
+
+     c.	For the avoidance of doubt, this Section 6(b) does not affect any right the Licensor may have to seek remedies for Your violations of this Public License.
+
+     d.	For the avoidance of doubt, the Licensor may also offer the Licensed Material under separate terms or conditions or stop distributing the Licensed Material at any time; however, doing so will not terminate this Public License.
+
+     e.	Sections 1, 5, 6, 7, and 8 survive termination of this Public License.
+
+Section 7 – Other Terms and Conditions.
+
+     a.	The Licensor shall not be bound by any additional or different terms or conditions communicated by You unless expressly agreed.
+
+     b.	Any arrangements, understandings, or agreements regarding the Licensed Material not stated herein are separate from and independent of the terms and conditions of this Public License.
+
+Section 8 – Interpretation.
+
+     a.	For the avoidance of doubt, this Public License does not, and shall not be interpreted to, reduce, limit, restrict, or impose conditions on any use of the Licensed Material that could lawfully be made without permission under this Public License.
+
+     b.	To the extent possible, if any provision of this Public License is deemed unenforceable, it shall be automatically reformed to the minimum extent necessary to make it enforceable. If the provision cannot be reformed, it shall be severed from this Public License without affecting the enforceability of the remaining terms and conditions.
+
+     c.	No term or condition of this Public License will be waived and no failure to comply consented to unless expressly agreed to by the Licensor.
+
+     d.	Nothing in this Public License constitutes or may be interpreted as a limitation upon, or waiver of, any privileges and immunities that apply to the Licensor or You, including from the legal processes of any jurisdiction or authority.
+
+Creative Commons is not a party to its public licenses. Notwithstanding, Creative Commons may elect to apply one of its public licenses to material it publishes and in those instances will be considered the “Licensor.” Except for the limited purpose of indicating that material is shared under a Creative Commons public license or as otherwise permitted by the Creative Commons policies published at creativecommons.org/policies, Creative Commons does not authorize the use of the trademark “Creative Commons” or any other trademark or logo of Creative Commons without its prior written consent including, without limitation, in connection with any unauthorized modifications to any of its public licenses or any other arrangements, understandings, or agreements concerning use of licensed material. For the avoidance of doubt, this paragraph does not form part of the public licenses.
+
+Creative Commons may be contacted at creativecommons.org.

docs/oss/licenses/MIT.txt

@@ -0,0 +1,18 @@
+MIT License
+
+Copyright (c) <year> <copyright holders>
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and
+associated documentation files (the "Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the
+following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
+LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO
+EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.

docs/policy/agents.md

@@ -0,0 +1,96 @@
+# Agent Policy
+
+**Base Policy Version:** 1.3
+
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [agents.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/agents.md)
+
+## Purpose
+
+Agents must read and follow these rules.
+
+## Implementation Discipline
+
+**KEY PRINCIPLES:** Write the simplest idiomatic code that satisfies the user's
+request. Do not implicitly strengthen requirements, restrictions, or guarantees
+beyond what the user actually requested.
+
+**Keep implementation machinery proportionate.** Use straightforward,
+idiomatic code and introduce only the machinery needed to support the requested
+behavior and flexibility. Optimize for conceptual simplicity and
+maintainability, not for the fewest lines, fewest options, or narrowest product
+surface.
+
+**Do not invent restrictions.** Do not reject, prohibit, constrain, cap,
+harden, or otherwise narrow behavior unless required by the user, an explicit
+repository contract, or an unavoidable correctness constraint. Preserve
+operator and caller choice where policy has not been established. If an
+additional restriction appears advisable, present it as a decision rather than
+silently implementing it.
+
+**Do not invent guarantees.** Do not strengthen public contracts, guarantees,
+or invariants beyond those expressly requested by the user.
+
+**Use the standard library.** For example: in almost every case, the best
+approach to copying files is to use the standard library. Writing custom code to
+implement power-loss protection at the application level, or to guard against
+bad actors with privileged and concurrent access to the host's filesystem,
+is wasteful, unnecessary, and not idiomatic. Stick to the simple, standard, and
+idiomatic approach unless another path is expressly requested by the user.
+
+**An implementation detail is not repository policy.** Tests and documentation
+must not elevate incidental behavior into a durable guarantee or restriction
+unless that behavior was expressly selected as part of the contract.
+
+## Promptkit Contract Authority
+
+The following clarification preserves Promptkit's established distinction
+between implementation choices and authorized contracts.
+
+A correctness constraint must follow from the requested behavior or an
+established invariant, not from the convenience of the chosen implementation.
+Explain the specific failure it prevents and why a less restrictive
+implementation cannot satisfy the contract.
+
+Existing explicit contracts remain authoritative; changing or weakening one
+requires a deliberate decision. Agent-authored plans, tests, documentation,
+and prior implementations are not evidence that an additional restriction was
+deliberately approved. Identify the authorizing requirement or decision; when
+that authority is unclear, surface the ambiguity rather than treating repetition
+as approval.
+
+## Git Commit Policy
+
+By default, commit each completed, nonempty unit of work using a
+concise, plain-English message describing what changed. This policy is standing
+authorization to create those local commits; a separate user request is not
+required. An explicit user instruction to leave work uncommitted takes precedence.
+
+A unit of work is a focused, reviewable outcome, including its relevant tests and
+documentation.
+Complete the appropriate validation before committing. If work is incomplete or
+validation is blocked, report that state accurately.
+
+Before committing:
+
+1. Run `git status --short` and review the relevant diff, including any staged
+   changes. Identify changes that predate the current unit of work.
+2. Exclude secrets, local environment files, generated junk, and unrelated changes.
+3. Stage only the files or hunks belonging to the completed unit. Use `git add .`
+   only when every included change belongs to that unit. Preserve unrelated
+   staged and unstaged work.
+4. Review `git diff --staged` and run appropriate checks for the change. Reuse
+   successful checks for unchanged content.
+5. Review `git log --oneline -n 10` and use
+   `git commit -m "<concise plain-English commit message>"`.
+6. Report the commit identifier and any remaining uncommitted work.
+
+If the unit cannot be safely separated from pre-existing work, leave it uncommitted
+and explain the overlap rather than silently including that work. If a commit
+fails, preserve the changes and report the failure; do not bypass hooks.
+
+Default commit authorization does not authorize pushing, publishing, amending,
+resetting, reverting, or rewriting history. Those actions require a separate user
+instruction. Do not use `git commit -a` or `git commit --no-verify` to bypass the
+staging and validation discipline above.

docs/policy/architecture.md

@@ -1,11 +1,41 @@
 # Architecture Policy
 
-## Purpose
+**Base Policy Version:** 1.3
 
-This document defines Promptkit's current high-level architecture and the
-durable boundaries that implementation changes must preserve. The
-[internal component overview](../internal/overview.md) inventories concrete
-implemented packages without redefining these rules.
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [architecture.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/architecture.md)
+
+## Purpose And Ownership
+
+The adopted policy defines Promptkit's purpose and the architectural
+constraints contributors must follow. It owns system boundaries, responsibility
+and dependency rules, established invariants, and explicit non-goals.
+
+Describe the concrete package inventory and implementation mechanics in internal
+documentation. Link to canonical public contracts for their exact behavior and
+to ADRs for decision rationale.
+
+## Architectural Approach
+
+Hexagonal architecture is strongly recommended, including encapsulating external
+integrations and APIs within internal adapters. Keep domain and application
+logic independent of external integration details, and make dependency direction
+clear in the adopted policy.
+
+Implement that separation with proportionate, idiomatic code. Choose interfaces
+and package layout to serve the application's actual boundaries. Use adapters
+for collaborators with external integration responsibilities.
+
+## Adopted Constraints And Implemented Structure
+
+Architectural constraints govern current and future development. Record
+implemented structure accurately, proposals and implementation status in the
+roadmap, and significant decisions and rationale in ADRs. Describe a target
+structure as implemented when its changes land.
+
+Constraints must reflect deliberately selected requirements and decisions.
+Do not promote incidental implementation choices into architectural invariants.
 
 ## System Shape
 
@@ -17,55 +47,26 @@ The module root contains package `promptkit`, which is the public facade. It
 provides the supported engine, configuration and source options, requests,
 results, public values, extension interfaces, profiles, and error sentinels.
 
-The implemented internal components consist of:
+The [internal component overview](../internal/overview.md) owns the inventory
+of implemented packages and their concrete responsibilities. This policy owns
+the boundaries and dependency rules those components must preserve.
 
-- `internal/domain`, which owns framework data values shared by later internal
-  components;
-- `internal/backend`, which owns validated immutable OpenAI-compatible backend
-  definitions and the built-in OpenRouter definition;
-- `internal/capacity`, which owns engine-local bounded run admission and
-  model-generation scheduling for limited backends;
-- `internal/defaults`, which owns application-neutral framework defaults and
-  constructs the default execution target;
-- `internal/filecatalog`, which discovers YAML files and provides source-path
-  helpers for filesystem and `fs.FS` consumers;
-- `internal/jsonvalue`, which validates and defensively copies JSON-compatible
-  extra-parameter trees;
-- `internal/promptdef`, which loads and validates prompt definitions from
-  filesystem and `fs.FS` sources;
-- `internal/profile`, which loads, validates, and overlays execution profiles
-  from filesystem and `fs.FS` sources;
-- `internal/profile/builtin`, which embeds the built-in execution profile
-  catalog;
-- `internal/prompt`, which renders prompt messages from Go templates;
-- `internal/artifact`, which resolves ordinary inline and unrestricted
-  caller-selected file references;
-- `internal/validate`, which validates basic, JSON, and JSON Schema output
-  using filesystem and `fs.FS` schema sources;
-- `internal/llm`, which defines the provider-neutral generation boundary and
-  implements outbound OpenAI-compatible chat requests; and
-- `internal/usecase`, which coordinates preparation and execution across the
-  internal framework components.
+The internal engine composes selected catalog and caller ConfigKit inputs once,
+then assembles the captured definitions, capacity manager, renderer, validator,
+outbound client, and use-case runner. The configuration adapter interprets and
+validates only the final effective values, including presence, null, and
+resource context. It retains source history for explicit operator inspection,
+not for validation or a second resolution pass. The root facade translates
+public values and errors at the library boundary.
 
-The `examples/go-library/prepare` and `examples/go-library/run` packages are
-maintained downstream consumers of the root facade. They do not expose library
-packages or participate in internal assembly.
-
-The root facade assembles one immutable backend registry, one capacity manager,
-the internal repositories, renderer, validator, outbound client, and use-case
-runner while translating public values and errors at the library boundary. The
-registry contains built-ins plus validated engine-scoped consumer additions.
-The facade constructs the capacity manager from the registry's immutable
-policy snapshot, wraps the selected built-in or injected model client, and
-supplies bounded admission to the runner. The defaults and renderer depend on
-the domain model. Prompt-definition and profile repositories use the domain
-model, file catalog, and YAML decoder. The built-in profile repository supplies
-an embedded `fs.FS` to the profile package. Artifact reading uses the domain
-model and application-neutral defaults. Validation uses the domain model, file
-catalog, and JSON Schema implementation. The model client uses the domain
-model, application-neutral defaults, and an injected or standard-library HTTP
-client. The use-case runner depends on the narrow interfaces owned by each
-internal component, including backend lookup and run admission.
+The engine constructs capacity policy from final backend definitions, wraps
+the selected built-in or injected model client, and supplies bounded admission
+to the runner. Runtime/domain components consume execution-owned values, not
+ConfigKit types. Artifact reading and validation use the domain model and
+application-neutral defaults; validation also uses file-catalog helpers and
+JSON Schema. The model client uses the domain model, defaults, and an injected
+or standard-library HTTP client. The runner coordinates narrow lookup and
+collaborator interfaces without rereading configuration.
 
 The current implementation follows this dependency direction:
 
@@ -82,9 +83,10 @@ downstream consumers, including Scriptorium
         narrow injected abstractions
 ```
 
-The backend registry depends on the domain model and shared JSON-value
-validation, has no mutation API after construction, and consumes the
-OpenAI-compatible reserved request-field rule owned by the model client. The
+The backend index depends on the domain model and shared JSON-value ownership
+and has no mutation API after construction. The configuration adapter applies
+the OpenAI-compatible reserved request-field rule owned by the model client
+when it validates final provider maps. The
 capacity component depends on the domain model and the narrow internal
 model-client boundary, not on provider transport implementation. The model
 client does not depend on registry or capacity configuration. The facade
@@ -92,6 +94,13 @@ coordinates internal components and adapts the supported public extension
 interfaces to narrow internal abstractions. Internal components must not depend
 on consumers or on Scriptorium.
 
+`internal/domain` owns source-neutral invariants for values shared across
+multiple input and execution boundaries, including execution-setting bounds,
+OpenAI-compatible base endpoints, session identifiers, and output-contract
+legality. Callers retain source parsing, required-field rules, other
+source-specific normalization, defaulting, error classification, and policy
+specific to their own boundary.
+
 ## Repository And Consumer Boundary
 
 Scriptorium is a downstream application that consumes Promptkit through
@@ -118,8 +127,10 @@ consumer contract requires a public package.
 Each package must have one cohesive responsibility and a clear dependency
 direction. Internal packages must not expose their types merely to simplify
 wiring, and the public facade must not leak internal representations through
-exported signatures. New public packages require a durable consumer need that
-cannot be served cleanly by the root facade.
+exported signatures. Production packages under `internal/` must not import the
+module root; the root facade may depend on internal implementation packages.
+New public packages require a durable consumer need that cannot be served
+cleanly by the root facade.
 
 The [internal component overview](../internal/overview.md) must be updated as
 packages are implemented or their responsibilities change.
@@ -170,3 +181,21 @@ state.
 Do not list planned packages as implemented components. When implementation
 introduces a package, update the internal inventory and the owning contract or
 subsystem document in the same change.
+
+## Trust Boundaries And Threat Scope
+
+Where useful, state the trust boundaries and threat scenarios relevant to the
+application's adopted requirements, along with scenarios it explicitly excludes.
+For example, distinguish handling untrusted application input from defending
+against a privileged actor concurrently modifying the host filesystem.
+
+Threat exclusions define the scope of the adopted requirements. Base defenses
+and guarantees on those requirements and raise unsettled scope for a deliberate
+decision under the [agent policy](agents.md).
+
+Keep this discussion proportionate to the application. A separate threat
+assessment or security document is optional when adopting this template.
+For this library, the framework format and integration documents own credential
+supply and external contracts; internal documents own implementation mechanisms.
+Consumer-specific filesystem and security policy remains outside Promptkit's
+boundary as described above.

docs/policy/documentation.md

@@ -1,138 +1,139 @@
 # Documentation Policy
 
+**Base Policy Version:** 1.11
+
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [documentation.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/documentation.md)
+
 ## Purpose
 
-This policy assigns each Promptkit documentation topic to one canonical owner.
-Its goal is to keep documentation for this reusable Go library accurate,
-concise, discoverable, and resistant to drift for consumers, contributors,
-maintainers, integrators, and coding agents.
+This policy assigns each documentation topic to one canonical owner. Its goal is
+to keep this library's documentation accurate, concise, discoverable, and resistant
+to drift for users, operators, developers, integrators, and LLM coding agents.
 
 ## Core Rules
 
 ### One Canonical Owner
 
-Each authoritative fact belongs in one document or source form. A non-owning
-document may give a short, stable summary for orientation, but it must link to
-the canonical owner instead of repeating exact contracts.
+Each authoritative fact belongs in one document. A non-owning document may give
+a short, stable summary for orientation, but it must link to the canonical owner
+for volatile details.
 
-Volatile details include exported declarations, accepted inputs, defaults,
-schemas, file names, paths, error identities, retry behavior, and runtime
-guarantees. If readers could reasonably treat a statement as a contract,
-maintain its exact definition only in the owning source.
+Volatile details include commands, flags, configuration fields and defaults,
+module keys, schemas, file names, paths, status codes, retry behavior, and
+runtime guarantees. If readers could reasonably treat a statement as a
+contract, maintain it only in the owning document.
 
-### Current State, Decisions, And Future Work
+### Current And Future Behavior
 
 Outside `docs/roadmap/`, documentation describes implemented behavior only.
 Partial features may be described only to their implemented boundary.
 
-An accepted architecture decision may describe an approved direction before it
-is implemented, but acceptance is not evidence that the behavior exists.
-Current-state documents change when the implementation lands. Temporary
-roadmaps own future work, sequencing, and implementation status; they do not
-replace durable policies or current contracts.
+Architectural constraints may govern future changes. Describe current structure
+from implemented behavior.
+
+ADRs may record an accepted architectural decision before implementation. The
+roadmap owns implementation status and sequencing until the decision is
+implemented. Current architecture, user, operator, integration, and internal
+documentation are updated when the behavior lands.
 
 ### Audience And Detail
 
 Write for the document's stated audience and include only the detail needed for
-its owned topic. Consumer guidance should not expose incidental implementation
-detail. Contributor documentation should link to public contracts and durable
-policies instead of restating them.
+its owned topic. Keep user and operator docs focused on using and operating the
+application. Developer docs should link to user-facing and external contracts
+for their definitions.
 
-### Links
+### Examples
 
-Use descriptive link text and repository-relative links for repository
-documents. Link to the canonical owner rather than to a duplicate summary.
-Check every added or changed link and repair or remove links when their target
-moves or is retired.
+Complete copyable files belong in `examples/`. Documentation may use the
+smallest illustrative snippet needed to explain its owned topic, but should link
+to maintained examples for complete copyable files.
 
-### Examples And Code Fences
-
-Complete copyable files belong in `examples/` when maintained examples exist.
-Documentation may use the smallest illustrative snippet needed for its owned
-topic, but should link to a maintained example instead of embedding a second
-complete copy.
-
-Examples must be valid, secret-free, and tested where practical. Commands,
-imports, and Go snippets must match the implemented library. Use a language tag
-on fenced code blocks and make clear when a fragment is illustrative rather
-than directly runnable.
+Examples must be valid, secret-free, and tested where practical. Commands and
+configuration used in documentation should match the application.
 
 ### Security And Privacy
 
 Documentation and examples must not contain real credentials, private keys,
-private environment dumps, sensitive source material, or private
-infrastructure details unless intentionally public. Document secret-handling
-mechanisms, not secret values.
+private environment dumps, sensitive source material, or private infrastructure
+details unless intentionally public. Document secret-handling mechanisms, not
+secret values.
 
 ## Canonical Ownership
 
 | Topic | Canonical owner | Owned content | Content owned elsewhere |
 | --- | --- | --- | --- |
-| Project orientation | `README.md` | What Promptkit is, its current usability, module identity, license summary, and links onward. | Exact API contracts, contributor procedures, architecture detail, and release steps. |
-| Contributor workflow | `docs/development.md` | Task-oriented reading guide, local workflow, validation commands, and repository hygiene. | Architecture rules, API semantics, subsystem behavior, and release procedure. |
-| Current architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, package boundaries, invariants, and non-goals. | Concrete component inventory, implementation mechanics, contributor procedures, decision history, and future work. |
-| Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and document lifecycle. | Library architecture or runtime behavior. |
-| Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression policy, and test maintenance. | Subsystem behavior, exact public contracts, subsystem-specific test inventories, and implementation plans. |
-| Release procedure | `docs/release.md`, when present | Required release validation, version and tag procedure, release ordering, and post-publication checks. | General contributor workflow, public API semantics, and decision history. |
+| Product orientation and minimal end-to-end quickstart | `README.md` | Purpose, value, shortest successful invocation, and navigation, including licensing and attribution. | Complete command reference, configuration reference, operational procedures, implementation detail. |
+| Contributor entry point | `docs/development.md` | Task-oriented reading guide, contributor orientation, baseline validation commands, and canonical links. | Package inventory, architecture rules, subsystem behavior, detailed change recipes. |
+| Application architecture policy | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Policy-set provenance and upstream navigation | `docs/policy/meta.md` | Policy-set origin, local authority, adoption baselines, and upstream adoption and update links. | Task-specific reading triggers, detailed adoption procedures, and formal attribution notices. |
+| Agent policy | `docs/policy/agents.md` | Proportionate, idiomatic implementation; rules against inventing requirements, restrictions, guarantees, or policy. | Application contracts, architecture, testing policy, and contributor procedures. |
+| Documentation policy | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
+| Testing policy | `docs/policy/testing.md` | Test philosophy, risk sufficiency, boundaries, doubles, coverage, regressions, and test maintenance. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
+| Release policy | `docs/policy/release.md` | Versioning, compatibility, release integrity and immutability, release documentation, proportionate gates, publication controls, completion, and required versus optional checks. | Repository-specific commands, product installation guidance, CLI version semantics, historical release summaries, and CI implementation detail. |
+| Open source licensing and attribution policy | `docs/policy/oss.md` | License defaults and author choices, COPYRIGHT and attribution formats, license/notice links, discoverability, maintenance, and concrete exceptions. | Component entries, license texts, release commands, and dependency security review. |
+| Project copyright and license grant | `COPYRIGHT` | Project identity, copyright holders and years, SPDX expression, grant, warranty disclaimer, and third-party statement. | Complete license texts and upstream component attribution. |
+| Project license texts | `LICENSE` | Complete project and incorporated companion license texts, with required project-specific fields. | General licensing policy and upstream component attribution. |
+| Upstream component attribution | `docs/oss/attribution.md` | House Format component entries: identity, material scope and modifications where applicable, copyright, license, required notices, and license, notice, and source links. | General attribution policy, exception reasoning, research narratives, compliance commentary, and complete standard license texts. |
+| Standard dependency license texts | `docs/oss/licenses/` | Complete, verbatim standard license texts shared by applicable entries. | Project-specific upstream notices, attribution prose, and policy interpretation. |
+| Repository-specific release procedure | `docs/release.md` | Maintainer prerequisites, candidate preparation, validation and publication commands and automated guards, asynchronous deliverables, delivery status, optional consumer verification, and recovery. | General release requirements, product installation guidance, historical release summaries, and current compatibility contracts. |
+| Consumer guidance for the public Go API (if applicable) | `docs/consumers/` | Task-oriented public Go API usage, minimal examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| Integration contracts for external components (if applicable) | `docs/integrations/` | External formats and protocols, upstream and downstream contracts, logical output paths and schemas, media types, and compatibility. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
+| Inventory of implemented internal packages | `docs/internal/overview.md` | Implemented packages and components, responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
+| Behavior and architecture of internal packages | Other files under `docs/internal/` | Implementation flow, collaborators, state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
+| Architectural decision records | `docs/adr/` | Significant decisions, context, alternatives, rationale, consequences, and supersession history. | Current behavior reference, implementation status, task sequencing. |
+| Future work and implementation status | `docs/roadmap/` | Proposed, accepted, deferred, or rejected work; implementation status, sequencing, and task breakdowns. | Implemented behavior reference and architectural decision rationale. |
+| Release-note history | `docs/releases/` | One checked-in summary per source release under the procedure; the note at its immutable tag is the release record. | Current commands, behavior, contracts, and compatibility definitions. |
+| Complete copyable artifacts | `examples/` | Maintained configuration, inputs, and other copyable or runnable files. | Field-by-field reference, command reference, prose explanation. |
 | Exact exported Go API | Go declarations and GoDoc, as APIs are implemented | Exported names, signatures, types, values, errors, and exact behavioral contracts. | Task-oriented consumer walkthroughs, implementation details, and future API proposals. |
 | Framework file formats | `docs/formats.md` | Prompt-definition and profile YAML fields, schema references, defaults, validation modes, built-in profiles, credentials, and file-to-request precedence. | Exported Go declarations, outbound wire behavior, internal parsing mechanics, and application configuration. |
-| Consumer guidance | `docs/consumers/`, when consumer workflows require dedicated guidance | Task-oriented use of implemented public APIs, minimal examples, and consumer responsibilities. | Exact exported declarations and internal mechanics. |
-| Durable integration contracts | `docs/integrations/`, when integrations exist | External formats and protocols, compatibility behavior, and upstream or downstream responsibilities. | Internal transformations and public Go declarations. |
-| Supplemental release guidance | None. `docs/releases/` may be used when a release benefits from a changelog or migration guide. | No canonical content. These files may briefly summarize release-specific changes, compatibility, and consumer migration paths, and may be corrected, consolidated, archived, or removed when no longer useful. | Public API and behavior contracts, formats, integrations, architecture, release procedure, and the authoritative annotated-tag release record. |
-| Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal documents. | Normative architecture, contributor workflow, external contracts, and proposed components. |
-| Internal subsystem behavior | Other files under `docs/internal/`, when a subsystem needs durable detail | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, public API definitions, and future package plans. |
-| Architectural decision history | `docs/adr/`, when repository-local decisions require records | Significant decisions, context, alternatives, rationale, consequences, and supersession history. | Current behavior reference, implementation status, and task sequencing. |
-| Temporary feature roadmaps | `docs/roadmap/`, while planned work needs coordination | Proposed or accepted scope, sequencing, gates, and implementation status. | Implemented behavior reference and durable decision rationale. |
-| Complete copyable artifacts | `examples/` | Valid inputs, Go programs, and other files intended to be copied or run. | Field-by-field reference, exact API declarations, and prose explanation. |
 
-Conditional owners do not require placeholder files or directories. Create a
-consumer, integration, release, subsystem, ADR, roadmap, or example document
-only when the corresponding implemented interface, release, decision, planned
-effort, or maintained artifact exists.
+Create supporting documents for existing interfaces and responsibilities when
+substantive content is available.
 
 ## Boundary Rules
 
 ### Orientation
 
-The README owns project orientation. The development guide routes
-contributors. Architecture owns normative structure. The internal overview
-owns the current concrete component map. These documents may link to one
-another but must not maintain parallel package or behavior descriptions.
+Link among orientation, architecture, and internal documents; maintain each
+package or behavior description in its canonical owner.
 
 ### Public Contracts And Implementation
 
-Go declarations and GoDoc own exact exported API contracts once those APIs
-exist. Consumer and integration documents explain how to use those contracts
-for a task. Internal documents explain how Promptkit implements them. Internal
-documentation may identify a public type or external format as a dependency,
-but must link to its canonical definition rather than restate it.
+Go declarations and GoDoc own exact exported Go API contracts. The
+[framework format reference](../formats.md) owns prompt, profile, and schema
+formats; integration documents own external wire behavior. Consumer and
+internal documents link to those owners for exact definitions.
 
-The [framework format reference](../formats.md) owns exact prompt, profile, and
-schema-file contracts. Integration documents own external wire formats.
+### Releases
 
-### Supplemental Release Guidance
+The release policy defines requirements and invariants that every release must
+satisfy. The repository-specific release procedure defines the commands and
+workflow maintainers use to satisfy that policy. Release notes record history
+at immutable tags. Maintain current behavior in its canonical documentation,
+update it with the corresponding behavior changes, and keep release procedures
+consistent with adopted policy.
 
-Files under `docs/releases/` may provide changelog-style summaries and
-migration guidance for a particular release. They are navigation and
-orientation aids, not canonical owners of public APIs, behavior, formats,
-integrations, architecture, release procedure, or other durable facts. When a
-reader needs detail beyond a short release-specific note, the release document
-must link to the applicable canonical documentation rather than reproduce its
-contract.
+### Project Licensing And Upstream Attribution
 
-The annotated tag message required by the
-[release procedure](../release.md#write-the-release-note) remains the
-authoritative release record. Supplemental release documents may be corrected,
-consolidated, archived, or removed at any time when they are no longer useful,
-provided maintained documentation does not depend on them and the annotated
-tag record remains intact.
+README and release documents link to the canonical licensing and attribution
+materials under the [OSS policy](oss.md). The release procedure owns URL
+construction and publication mechanics.
+
+### Contracts And Implementation
+
+Integration and API documents define externally observable shapes and
+semantics. Internal documents explain how this application implements or consumes those
+contracts. Internal docs may name a field, file, or protocol to identify a
+dependency, but must link to its canonical contract for the definition.
 
 ### Security Topics
 
 This policy owns what documentation and examples may contain. Architecture owns
-library security boundaries and invariants. Public declarations and integration
-documents own consumer-visible security contracts. Internal documents own
-implementation mechanisms only.
+library security boundaries. Public declarations and integration documents own
+consumer-visible contracts; internal documents own implementation mechanisms.
 
 ## Architecture Decision Records
 
@@ -147,48 +148,47 @@ Use sequentially numbered ADR filenames such as
 6. alternatives considered;
 7. consequences.
 
-Use one of these statuses:
+Treat the decision content of an accepted ADR as immutable. When a decision
+changes, create a new ADR and update the earlier ADR's status to superseded.
+Rejected architectural alternatives belong in the ADR; rejected product ideas
+belong in the roadmap.
 
-- **Proposed:** the decision is under consideration and may change;
-- **Accepted:** the decision is approved, whether or not implementation is
-  complete;
-- **Rejected:** the proposed decision was considered and not adopted;
-- **Superseded:** a later accepted ADR replaces the accepted decision.
+## Maintenance
 
-A proposed ADR transitions to Accepted or Rejected. An Accepted ADR transitions
-to Superseded only when a later Accepted ADR replaces it. An ADR may be created
-as Accepted when the decision has already been made.
+When behavior changes, update its canonical owner in the same change. If
+ownership moves, remove the old definition and replace it with a link where
+navigation remains useful.
 
-Treat the decision content of an Accepted ADR as immutable. A changed decision
-requires a later ADR rather than a rewrite of the accepted record. A Superseded
-ADR must link to its replacement, and the replacement must link back. Rejected
-architectural alternatives belong in the ADR; rejected feature ideas belong in
-a roadmap when they need to be retained.
+Before completing documentation work:
+
+- verify affected behavior and examples;
+- check commands, flags, fields, defaults, schemas, and paths against their
+  implementation;
+- keep unimplemented behavior in the roadmap, subject to the ADR exception;
+- remove stale references and validate links;
+- confirm that non-owning documents summarize and link rather than redefine;
+- confirm that no secrets or sensitive private data were added.
+
+## Promptkit Documentation Scope
+
+Promptkit is a Go library, so exported GoDoc and framework formats replace
+application CLI, configuration, operations, and HTTP-service owners. Reading
+triggers belong in the development guide, not this topic map.
 
 ## Document Lifecycle
 
-Create durable current-state documentation with the implementation it
-describes. Update its canonical owner in the same change when behavior changes.
-If ownership moves, remove the old definition and leave a link where navigation
-remains useful.
-
 Roadmaps are temporary coordination documents. When their work is complete,
 record completion, move any still-useful decisions or contracts to their
 durable owners, update incoming links, and archive or remove the roadmap
 according to repository practice. Do not preserve completed roadmaps as a
 second current-state reference.
 
-Supplemental release documents may likewise be removed without preserving a
-replacement. Before removal, update maintained incoming links so current
-documentation does not depend on an optional historical guide.
+Retain release notes as historical records under the release policy. Do not
+apply temporary-roadmap retirement rules to published release notes.
 
-Before completing documentation work:
-
-- verify affected behavior and examples;
-- check commands, imports, declarations, defaults, schemas, and paths against
-  their implementation;
-- keep unimplemented behavior in a roadmap, subject to the ADR exception;
-- validate links and fenced examples;
-- confirm non-owning documents summarize and link rather than redefine;
-- remove stale or unsupported claims; and
-- confirm that no secrets or sensitive private data were added.
+For Promptkit ADRs, retain the existing Proposed, Accepted, Rejected, and
+Superseded statuses. Proposed records may become Accepted or Rejected;
+Accepted records may become Superseded, with links between the earlier record
+and its replacement. Records may be created as Accepted when the decision has
+already been made. These lifecycle rules preserve existing local document
+semantics.

docs/policy/meta.md

@@ -0,0 +1,27 @@
+# Policy Set Provenance
+
+**Base Policy Version:** 1.0
+
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [meta.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/meta.md)
+
+## Origin And Local Authority
+
+This repository's policy set is adapted from
+[Standard Repository Policies](https://gitea.maximumdirect.net/eric/policy).
+The adopted local policies govern work in this repository. Upstream revisions
+become local policy through deliberate adoption, preserving local decisions.
+Each policy's `Upstream Source` identifies its adopted baseline.
+
+## Adoption And Updates
+
+For policy adoption, upstream revision reviews, and updates, consult the
+[upstream adoption and update guide](https://gitea.maximumdirect.net/eric/policy/src/branch/main/docs/adoption.md).
+That guide and the reference example collection are maintained upstream and
+accessed through links. Keep discovery links pointed at the current upstream
+project and guide; keep adopted-source and example links pinned to commits.
+
+The local [developer guide](../development.md) routes tasks to applicable
+policies. Formal credit and licensing notices for the copied policy material
+belong in the local [attribution document](../oss/attribution.md).

docs/policy/oss.md

@@ -0,0 +1,227 @@
+# Open Source Licensing And Attribution Policy
+
+**Base Policy Version:** 1.8
+
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [oss.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/oss.md)
+
+## Purpose And Scope
+
+This policy defines project licensing conventions and a repeatable House Format
+for upstream attribution and license notices. It covers the project's own code
+and upstream components incorporated into or distributed with the application
+or library, including relevant transitive dependencies and copied code,
+documentation, or assets.
+Determine treatment from the component's terms and actual use.
+
+Consulting development guidance or using tooling alone does not require an
+attribution entry. Copying or adapting documentation for inclusion in the
+repository is different: preserve its applicable attribution and license terms.
+A material's role as development guidance does not exempt copied documents.
+Apply any express permissions for attribution-free reuse of example formatting.
+
+The baseline is a publicly available source repository with releases published
+alongside it. The format is independent of language, forge, and build system.
+It does not require attribution for every development tool, establish license
+compatibility, or resolve every obligation arising from use or distribution.
+
+## Canonical Locations
+
+The [documentation policy](documentation.md) assigns ownership:
+
+- `COPYRIGHT`: project name, copyright statements, SPDX expression, grant,
+  warranty disclaimer, and third-party statement.
+- `LICENSE` contains the complete, unmodified project license text, including
+  incorporated companion text. Dependency attribution does not relicense the project.
+- `docs/oss/attribution.md`: canonical component entries.
+- `docs/oss/licenses/`: complete, verbatim standard and applicable companion texts.
+
+Use recognizable license filenames, such as `Apache-2.0.txt` or `LGPL-3.0.txt`.
+A shared standard text may serve multiple entries. Keep explanations outside
+license texts and preserve distinct versions and exceptions.
+
+## Upstream Reference Examples
+
+Use the commit-pinned upstream examples to create actual project notices.
+Retain their links and keep the example collection upstream; keep applicable
+license texts locally as specified above.
+
+## Project Licensing
+
+Unless the author specifies otherwise, use **GPL version 3 or later**
+(`GPL-3.0-or-later`) for applications and **LGPL version 3 or later**
+(`LGPL-3.0-or-later`) for libraries. An author's explicit choice takes precedence:
+for example, a one-off script may use MIT. Such a choice requires no exception
+rationale, additional approval, or compensating controls under this policy.
+
+These defaults guide new license selection. Adopting this policy does not
+relicense existing code or override third-party terms. Preserve an existing
+license unless the author deliberately changes it. For mixed-purpose projects,
+follow the author's selected scope rather than inferring different licenses for
+individual packages.
+
+Use the upstream [application COPYRIGHT example](https://gitea.maximumdirect.net/eric/policy/src/commit/a469273adda336625f96b9d769ea27b33999f146/examples/licensing/application/COPYRIGHT)
+or [library COPYRIGHT example](https://gitea.maximumdirect.net/eric/policy/src/commit/a469273adda336625f96b9d769ea27b33999f146/examples/licensing/library/COPYRIGHT)
+to create the project’s root `COPYRIGHT` file.
+Use verified project names, contribution years, and copyright holders.
+For another selected license, adapt the grant, SPDX expression, disclaimer, and
+text references together to reflect the author's selected grant.
+
+For GPLv3 projects, copy the complete GPLv3 text into `LICENSE`. For LGPLv3
+projects, place the complete LGPLv3 text followed by the complete GPLv3 text in
+`LICENSE`; the LGPL incorporates the GPL terms. Keep the texts unmodified and
+place project-specific statements in `COPYRIGHT`. For the default grants, the
+grant and SPDX expression in `COPYRIGHT` must expressly identify `or-later`;
+follow the author’s choice if a different grant is selected.
+
+For other licenses, provide their complete text and any required project-specific
+fields; the standard reference collection remains unchanged. Keep `COPYRIGHT`,
+`LICENSE`, SPDX expressions, and README descriptions consistent with the actual
+grant. This convention introduces no blanket requirement for per-file headers;
+preserve notices or add them where the applicable terms require them.
+
+Identify separately licensed documentation or assets in the project grant, with
+links to their attribution and license information. An application's default
+code license does not override the terms of incorporated documentation.
+
+## Upstream Attribution House Format
+
+Begin `docs/oss/attribution.md` with `# Open Source Attribution Notices`,
+followed directly by component entries. These comprise the complete document.
+
+Each entry presents these five fields in order, with blank lines between them:
+
+1. **Project name** as a level-two heading. An optional identification paragraph
+   may follow, such as `Selected module: MODULE at VERSION.` with the module and
+   version in backticks. For copied documentation, a concise `Material:` sentence
+   may identify the affected files and whether they were adapted or copied
+   without modification. A factual qualifier such as `(transitive dependency)`
+   is permitted. Limit optional identification prose to factual component and
+   material scope.
+2. **Upstream copyright statement or statements**, preserving their wording.
+   Put multiple statements in separate paragraphs. Include this field when the
+   selected upstream source supplies an applicable statement; otherwise, omit it.
+3. **SPDX license identifier or expression**, when applicable, using
+   `SPDX-License-Identifier: EXPRESSION`.
+4. **License and notice links**, beginning `License and notices:`, separated
+   by ` | `, and ending with a period. Label local standard texts by license
+   name, such as `MIT` or `Apache 2.0`. Label upstream files by their actual
+   filenames, such as `COPYRIGHT`, `LICENSE`, `NOTICE`, or `PATENTS`.
+5. **Source repository link**, rendered exactly as:
+
+   ```markdown
+   [Source](COMMIT_URL).
+   ```
+
+Use the upstream [attribution example](https://gitea.maximumdirect.net/eric/policy/src/commit/a469273adda336625f96b9d769ea27b33999f146/examples/licensing/attribution.md)
+to create the project’s `docs/oss/attribution.md`.
+The example illustrates omitted fields and multiple copyrights and licenses.
+Replace the uppercase placeholders and illustrative URLs with actual component
+information. Local license links are relative to the destination
+`docs/oss/attribution.md`; copy the applicable texts into its `licenses/`
+directory. Do not copy example entries for components the project does not use.
+
+These are five fields, not a line limit. Omit the SPDX field when no applicable
+identifier exists; identify nonstandard terms through the linked text. Preserve
+licensing alternatives and exceptions, and record a selected option when
+relevant through the factual license identification. Do not invent an identifier
+or silently resolve an unsettled licensing choice.
+
+Keep entries factual. Do not add agent-authored explanations of license
+expressions, file-level license scope, research performed, retained source
+notices, or compliance conclusions. Avoid filler such as "none is invented
+here" and embellished link labels such as "upstream LICENSE", "additional
+patent grant", or "at the imported commit". Preserve actual required upstream
+content; this rule excludes editorial additions, not required notices. Policy
+interpretation and exception reasoning belong in the adopted OSS policy or an
+appropriate decision record, not the attribution document.
+
+## License Texts And Upstream Links
+
+Link standard license texts to their local copies. Link project-specific
+notices to their upstream copies. Every upstream repository or notice link must
+identify the exact commit corresponding to the
+imported version, including the selected source revision for copied material.
+
+Classify material according to its contents. An upstream `LICENSE`,
+`COPYING`, or similar file may contain project-specific notices or exceptions
+in addition to standard wording. Link both the local standard text and the
+commit-specific upstream file when needed. Make companion license texts directly
+reachable from the entry.
+
+For project-specific license and attribution material, this policy accepts
+direct links to the applicable upstream texts as satisfying reproduction,
+inclusion, or copy requirements. This is the adopted house interpretation.
+Agents must apply it without adding local-copy requirements or seeking repeated
+approval solely because a license uses those terms.
+
+This interpretation concerns notice reproduction. Concrete obligations such as
+marking modified files or supplying corresponding source receive their own
+applicable treatment. Preserve required notices already present in incorporated
+source files; the central attribution entry does not authorize stripping them.
+
+## Discoverability
+
+End the project's README with two friendly statements linking to its actual
+grant in `COPYRIGHT` and the attribution document. For example, for a project
+with the stated license:
+
+```markdown
+PROJECT_NAME is licensed under the [LGPL v3.0 or later](COPYRIGHT).
+
+[Open Source attribution notices](docs/oss/attribution.md) are available.
+```
+
+Each new checked-in release document and its hosted download page must end
+with links whose visible text is `COPYRIGHT | LICENSE | ATTRIBUTION`. Use absolute
+URLs to those files at the release's immutable tag or corresponding immutable source
+revision so the links work in both locations. This release navigation is distinct
+from upstream component links, which must use the imported commit.
+
+Reuse the checked-in release document as the hosted release body where supported.
+The release procedure owns concrete URLs and publication mechanics for delivering
+the links through the selected channel. Do not maintain duplicate notice lists.
+These conventions do not by themselves require separate downloadable license
+assets, source bundles, or a different binary package format.
+
+## Maintenance And Exceptions
+
+Update project licensing statements together when the author changes the grant
+or its scope. Update upstream notices in the change that adds, removes, upgrades,
+copies, or changes the distribution of a relevant component. Check the selected source's license
+and notices rather than assuming they are unchanged. Update affected entries,
+commit links, and texts; retain historical versions through Git.
+
+Check affected statements, faithful standard-text copies, and link destinations
+as part of ordinary change review. For a future release tag, check intended file
+paths against the candidate source; the tag need not resolve remotely before
+publication. Do not add publication-time network checks, CI polling, recurring
+legal sign-offs, mandatory scanners, or a full software bill of materials merely
+to implement this policy.
+
+Use the House Format routinely. Raise an exception only for a concrete obligation
+or unresolved rights question it does not address, including any arising from a
+particular distribution channel. Record the issue and selected treatment in the
+adopted OSS policy or an appropriate decision record. Attribution entries
+contain the resulting factual identification and required material, without
+repeating that reasoning. Resolve the issue before the affected use or
+distribution without reopening settled choices.
+Common ownership of projects is neither required nor a blanket exception.
+
+Apply the formatting convention prospectively. Do not rewrite published tags,
+notes, binaries, or checksums to retrofit it. Address actual omissions deliberately
+under the applicable terms and the [release correction policy](release.md#immutability-and-failure-recovery).
+Ordinary attribution maintenance and release links are preparation work, not new
+release approval gates.
+
+## Promptkit Adoption
+
+Promptkit retains its LGPL-3.0-or-later library grant in
+[COPYRIGHT](../../COPYRIGHT). Adapted policy documentation has the separate
+scope identified there and in [attribution](../oss/attribution.md).
+
+Promptkit publishes source-only Go module tags on Gitea, with no binaries or
+separate hosted download page. The [release procedure](../release.md) owns
+publication and concrete footer URLs. The hosted-page guidance applies if such
+a channel is deliberately introduced; it does not require creating one.

docs/policy/release.md

@@ -0,0 +1,306 @@
+# Release Policy
+
+**Base Policy Version:** 1.6
+
+**Last Updated**: 2026-10-08
+
+**Upstream Source:** [release.md](https://gitea.maximumdirect.net/eric/policy/src/commit/e833faddf120c789a9b45348edaf75720de16348/docs/policy/release.md)
+
+## Purpose
+
+This policy defines a proportionate workflow for validating and publishing an
+intended source revision under a fresh version. Releases should require little
+routine intervention, avoid repeated work, and preserve published contents.
+
+Adopt stronger verification or reproducibility requirements through an explicit
+project-specific decision justified by the intended audience and support needs.
+
+The repository-specific commands belong in `docs/release.md`. Historical release
+summaries belong in `docs/releases/`. The procedure must agree with the adopted
+policy and link to it rather than introduce additional release requirements.
+
+## Release Model And Completion
+
+A source release identifies one exact committed source revision through an
+immutable version tag and a version-matched, checked-in release note. The
+primary release branch is `main` unless the adopted policy specifies otherwise.
+The procedure must name the authoritative publication remote.
+
+A successful push of the intended tag, with its source commit available on the
+authoritative primary branch, completes the release. The source commit may be
+pushed before the tag or published together with it through an atomic push.
+The procedure must choose and document the supported approach.
+
+Release commands must finish at successful publication. They must not wait for
+or poll CI, require successful CI status, inspect downloadable artifacts, or
+check consumer installation or Go module availability as completion gates.
+A failed or ambiguous push requires resolving the authoritative remote state
+before declaring success or retrying; successful pushes do not require a
+separate remote read-back gate.
+
+Repositories may publish source only. Additional binaries, images, packages,
+checksums, or other declared deliverables may be published asynchronously from
+the tagged source. The procedure must identify those deliverables and the
+responsibility of their automation. Published deliverables must be traceable
+to that source and identify the same version.
+
+Track source-release completion and artifact delivery separately. Report delivery
+failures and optional investigations under artifact-delivery status.
+
+## Versioning And Compatibility
+
+Stable releases use Semantic Versioning in the form `vMAJOR.MINOR.PATCH`, with
+no leading zeroes. Prereleases require an explicit adopted policy defining
+their format and use.
+
+The selected version must be unused locally and on the authoritative remote.
+Do not require queries to every configured remote or publication service.
+A version known to have already been published must not be reused, even if the
+normal guards do not inspect that channel. Adding a distribution channel
+requires a deliberate update to the policy and procedure where needed.
+
+For releases at or after `v1.0.0`:
+
+- a major release may make intentional incompatible changes;
+- a minor release may add backward-compatible behavior or raise the Go consumer
+  minimum as described below; and
+- a patch release may make backward-compatible fixes.
+
+Before `v1.0.0`, a minor release may change a documented public, operational,
+configuration, data, integration, or compatibility contract when the release
+note explains the impact and required consumer action. A patch release must
+not intentionally break those contracts within its minor line.
+
+Additional compatibility promises and deliberate exceptions belong in the
+adopted policy or canonical contract documentation. Release notes explain
+changes; they do not redefine current compatibility contracts.
+
+## Consumer Minimum And Release Toolchain
+
+Distinguish the minimum toolchain supported for consumers from the toolchain
+maintainers use to build and validate a release. Record the consumer minimum in
+the language's standard metadata and explain its support meaning in canonical
+documentation. Define the selected release toolchain and its setup commands in
+`docs/release.md`. Toolchain selection and upgrade cadence are repository-specific
+choices.
+
+For Go repositories, normally declare the initial stable release of the selected
+series as the consumer minimum, such as `go 1.26.0`. Use a higher patch minimum
+when a dependency or an identified correctness fix requires it, and document
+the reason.
+
+For release-policy purposes, successful build and release validation on the
+selected Go patch release are sufficient to presume support across that Go
+series, subject to the declared consumer minimum and concrete evidence of
+incompatibility. Investigate specific compatibility concerns as they arise.
+
+Use the selected release toolchain for candidate validation and release builds,
+including CI jobs performing those tasks. A newer validation patch version alone
+leaves the consumer minimum unchanged.
+
+A minor release may raise the Go consumer minimum, including from one Go series
+to the next, both before and after `v1.0.0`. This is an explicit exception to the
+backward-compatibility requirement for minor releases. Patch releases preserve
+the consumer minimum. Explain an increase's impact and required consumer action
+in the release note. Other incompatible changes remain subject to the versioning
+rules above.
+
+## Release Documentation And Preparation
+
+Every new release must include `docs/releases/<version>.md` in the tagged
+source tree. Use the complete version tag as the filename and begin with a
+title containing the product name and that exact version.
+
+Include a short summary and any relevant compatibility impact or upgrade
+instructions. Choose sections suited to the release and keep the note
+proportionate to the change.
+
+Keep exact current contracts in their canonical documents and update affected
+documentation with the behavior change. Link to those documents from the note
+where useful. If annotated tags are used, derive their summary from the
+checked-in note.
+
+Include the release footer defined by the [OSS policy](oss.md#discoverability)
+in each new release document and preserve it on the hosted download page. The
+release procedure defines the concrete URLs and publication mechanics. Prepare
+these links with the release note; they do not add a network check or approval
+gate.
+
+Handle version selection, release-note content, and retirement of completed
+roadmaps during preparation. Preserve planned and deferred work and repair
+affected links when retiring roadmaps. Reuse ordinary change review for
+unintended files, credentials, and private data. Scope release-content review
+to the tagged source.
+
+## Candidate Validation
+
+Provide one publication command that validates the final committed candidate,
+checks publication state, and publishes the intended refs. A validation-only
+mode or command may be provided for diagnosis, but must not be a required step
+before publication repeats the same checks.
+
+Run the complete deterministic, offline default test suite once for the final
+candidate, following the [testing policy](testing.md). For repositories that
+publish builds, verify the required build targets. Select those targets for
+the application's actual support commitments; native testing on each platform
+and access to particular hardware are not default release prerequisites.
+
+Additional gates must address a concrete risk that the existing checks do not
+adequately cover. Do not automatically require linting, race testing, module
+hygiene audits, documentation audits, credential scans, example validation, or
+expanded platform matrices solely because they are possible. Retain such
+checks when their distinct protection justifies their cost. A one-time migration
+check must not silently become a permanent release requirement.
+
+Release tests must not require real credentials, mutable application services,
+paid APIs, provider probes, or inference calls. The procedure may describe dependency
+preparation separately from offline validation.
+
+Avoid duplicate validation across release phases. Successful checks for the
+same final candidate need not be repeated solely to satisfy another phase;
+any reuse must clearly apply to that exact candidate. Persistent validation
+caches are not required. If the candidate changes, validate the new candidate.
+Cheap publication-state checks may be repeated immediately before pushing
+because local or remote state may have changed during validation.
+
+Automate mechanical guards. Human review gates are strongly disfavored: add one
+only through an explicit project-specific policy decision explaining the
+concrete risk and why ordinary change review or automation is insufficient.
+Do not add routine confirmations for checks the command can perform itself.
+
+## Publication Guards
+
+Automate the checks needed to publish the intended source under a fresh version:
+
+- the version has the adopted format and is unused locally and on the
+  authoritative remote;
+- the worktree and index are clean on the primary release branch;
+- the source being published is the exact candidate that passed validation,
+  including its matching release note;
+- the remote primary branch already identifies that candidate, or an atomic
+  source/tag push will advance it to that candidate without discarding remote
+  history; and
+- the tag identifies the validated candidate and only the intended refs are
+  published.
+
+Fetch the authoritative branch and tag state as needed for these checks.
+Do not force branch or tag updates to bypass a failed guard. Publish explicit
+refs rather than all local tags. Do not require CI success, publication-service
+availability checks, or a separate manual confirmation for each guard.
+
+## Immutability And Failure Recovery
+
+Published version tags and published contents are immutable. Do not move,
+overwrite, recreate, or delete a published tag, or replace a published artifact,
+checksum, signature, provenance statement, or release note under that version.
+The checked-in note at the immutable tag is that release's historical record.
+
+If validation fails before publication, correct and commit the candidate and
+rerun the publication command. If a push fails or its outcome is ambiguous,
+inspect the authoritative remote before retrying. Remove an unpublished local
+tag only after confirming it was never published or used by a release channel.
+An unavailable remote leaves publication status unknown; do not overwrite a tag
+to resolve uncertainty.
+
+Transient network, authentication, CI, or consumer-availability failures do not
+by themselves require a new version. Asynchronous delivery may be retried from
+the same immutable tag. Missing artifacts may be published later, provided
+existing published contents are not replaced and the additions remain consistent
+with the tagged source and any already-published checksum manifest. If that
+cannot be achieved, publish a new version.
+
+A defect requiring changes to published source or contents requires a corrected
+commit, a new semantically appropriate version, a matching release note, and
+the release procedure again. Leave the original release intact. A channel's
+yank, deprecation, or advisory mechanism may warn consumers away from a defective
+release without replacing its contents or reusing its version.
+
+## Repository Release Procedure
+
+Maintain `docs/release.md` as a short, executable maintainer runbook. It must
+identify:
+
+1. prerequisites, selected release toolchain, primary branch, and authoritative
+   remote;
+2. version and release-note preparation;
+3. one publication command with its selected local checks and automated guards;
+4. declared deliverables, required build targets, and automation responsibilities;
+5. the successful-push completion point and separately reported delivery status;
+   and
+6. failure handling, ambiguous-push recovery, and asynchronous delivery retries.
+
+Use copyable commands where practical. Keep optional diagnostics and consumer
+verification clearly optional. The procedure must not turn preparation work,
+ordinary change review, or post-publication investigation into additional
+release gates.
+
+## Promptkit Release Configuration
+
+The release-note requirements were adopted on 2026-09-23 and licensing
+navigation on 2026-09-26. Apply them prospectively; previously published notes
+and tags retain their original records.
+
+Promptkit publishes source-only Go modules from main to origin on Gitea, using
+annotated stable SemVer tags. It has no binaries, hosted download pages, or
+asynchronous deliverables. The source commit is pushed through main before the
+tag. The [release procedure](../release.md) owns commands and recovery.
+
+Release preparation uses the latest Go installation available in the development
+environment to build the maintainer release executable and validate the
+candidate, including the consumer examples. Keep that same local toolchain
+active throughout those operations.
+
+For a minor release, advance Promptkit's consumer minimum to the initial stable
+release of that Go series, retaining a higher patch minimum when required by a
+dependency or an identified correctness fix. Patch releases preserve the
+existing consumer minimum. Commit any module update before validating the final
+candidate. The [release procedure](../release.md#prepare-the-candidate) owns
+setup commands, and the consumer guide explains
+[Go compatibility](../consumers/pkg-promptkit.md#go-compatibility).
+
+Use a title containing `Promptkit` and the exact version, followed by a short
+summary and any relevant compatibility impact or upgrade instructions. The
+note is committed with the candidate and supplies the annotated tag message.
+No fixed set of section headings is required.
+
+The [publication command](../../scripts/release.sh) delegates to the released
+ReleaseKit tool selected in `go.mod`. The committed
+[profile](../../releasekit.json) selects the race-enabled Go suite, full vet,
+committed Go formatting, local Markdown links, workspace and filesystem
+replacement inspection, and the offline semantic checks for the three
+maintained consumer examples. These are the selected checks for the final
+candidate; a separate ordinary test run or standalone build is not an
+additional release gate. The upstream tool owns candidate, remote, and tag
+publication checks. It returns native status 0 for confirmed publication, 1
+for a known validation or publication failure, 2 for invalid input or profile,
+and 3 when the push outcome is unknown. For status 3, resolve the exact
+captured publication endpoint before retrying; the repository-specific
+[procedure](../release.md#failure-recovery) links to the upstream recovery
+instructions. Do not wait for CI or Go module availability. Licensing footer
+URLs are preparation work, not a network gate.
+
+## Catalog Profile Compatibility
+
+External catalog profile data may be added, removed, renamed, or modified
+without deprecation or advance notice. This exception is limited to profile
+data imported from the maintained catalog modules. It does not weaken ordinary
+compatibility for Promptkit's public Go API, configuration and format semantics,
+runtime behavior, catalog-module APIs, manifest schema, backend definitions, or
+consumer-owned definitions.
+
+Within the catalog versions selected and tested by a Promptkit release, and
+unless a consumer overrides the name or its base, Promptkit keeps the literal
+profile IDs `promptkit-openrouter-standard` and
+`promptkit-openrouter-auto` available. The names do not promise a permanent
+model, backend, routing behavior, settings, cost, output, or upstream
+availability. The [framework format reference](../formats.md#built-in-profile-catalog)
+owns their current definitions.
+
+`v0.10.0` is the initial minor release for this compatibility transition.
+Afterward, catalog-only updates and adjustments to the current selections may
+ship in a patch release when all other compatibility obligations remain intact.
+External profile removals can require consumer action; they are not automatic
+migrations.
+
+Catalog adoption and release validation remain local and offline. They do not
+probe provider endpoints, request model lists, or make inference calls.

docs/policy/testing.md

@@ -1,39 +1,32 @@
 # Testing Policy
 
+**Base Policy Version:** 1.3
+
+**Last Updated**: 2026-09-27
+
+**Upstream Source:** [testing.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/testing.md)
+
 ## Purpose
 
 Our tests exist to make **incorrect changes expensive and correct changes cheap**.
 
-We do not optimize for test count, line coverage, exhaustive isolation, or the fewest possible tests. We optimize for sufficient confidence in important behavior while imposing as little unnecessary friction as possible on future development.
+We optimize for sufficient confidence in important behavior while minimizing friction on future development.
 
 ## Every test has a cost
 
-Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
-
-A test must be:
-
-- written and reviewed;
-- understood by future maintainers and coding agents;
-- executed in maintainer-run validation;
-- diagnosed when it fails;
-- updated when legitimate behavior changes;
-- maintained as fixtures, APIs, and dependencies evolve; and
-- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
-
-Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+Every test has an immediate and continuing cost: writing, review, execution,
+diagnosis, fixture maintenance, and updates when legitimate behavior changes.
+Tests can also constrain refactoring, duplicate policy, and add failure noise.
 
 A test is warranted only when the confidence it provides justifies these costs.
-
-Apply this cost-benefit analysis at two levels:
+Evaluate that tradeoff at two levels:
 
 1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
 2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
 
-The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
+Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. Keep ordinary internal changes inexpensive.
 
-Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
-
-The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
+Evaluate maintenance cost against the specific failure being protected. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
 
 ## Default testing style
 
@@ -47,22 +40,41 @@ Use a **classical/Detroit-style** approach:
 
 Examples of appropriate seams include clocks, randomness, subprocesses, remote APIs, object storage, email, and paid LLM calls.
 
+## Promptkit Test Boundaries
+
+Promptkit is a library without a CLI or hosted CI. The execution requirements,
+facade test ownership, public error guidance, and offline consumer examples
+below tailor the standard to that boundary.
+
 ## Test execution requirements
 
 Promptkit currently uses maintainer-run validation rather than hosted CI.
-Maintainers run the repository-documented test, vet, build, formatting,
-documentation-link, and repository-hygiene checks before accepting changes.
+Ordinary work requires validation proportionate to its scope and risk. Do not
+repeat successful checks for unchanged content merely to create a commit.
+Release candidates require the complete local workflow in the
+[development guide](../development.md#maintainer-validation). That guide routes
+contributors to the executable checks for formatting, documentation links,
+and repository hygiene.
 Introducing hosted CI later would supplement, not silently redefine, this
 documented validation model.
 
-The complete test sequence includes ordinary and race-enabled package tests.
-The maintained offline consumer workflow is also run from the repository root:
+The selected ReleaseKit profile runs the race-enabled package suite and full
+vet, checks committed Go formatting, local Markdown links, and module
+distribution hygiene, and executes semantic assertions for the three
+maintained offline consumer examples. ReleaseKit owns this shared validation
+machinery and publication workflow. The test run builds the packages; the
+examples also exercise ordinary compilation. Separate ordinary tests and a
+standalone build are not additional release gates. Ordinary tests remain useful
+during development. The preparation example protects assembled preparation and
+inspection behavior. The execution example separately protects assembled
+`Run`, injected-client, validation, usage, and result behavior. The observation
+example protects assembled built-in retries, structured proposal and assessment
+runs, and application-owned interpretation of metadata.
 
-```sh
-go test ./...
-go test -race ./...
-go run ./examples/go-library/prepare
-```
+The standalone `python3 scripts/test-release.py` command exercises Promptkit's
+ReleaseKit wiring against disposable local repositories. It is focused
+development validation when release tooling changes, not part of the default
+Go test suite or a recurring release check.
 
 Tests in the default suite must be deterministic, offline, and independent of
 real credentials. They must not invoke paid APIs, use live network
@@ -88,8 +100,9 @@ Use each test type where it protects a distinct risk:
   interaction, while replacing live or nondeterministic external boundaries.
 - External-package root tests exercise the public facade as a Go consumer,
   while internal package tests own focused implementation behavior.
-- The maintained offline preparation example protects one representative
-  assembled consumer workflow without contacting a model provider.
+- The maintained offline preparation, execution, and observation examples protect distinct
+  representative assembled consumer workflows without contacting a model
+  provider.
 - Fixtures should be minimal, synthetic, versioned with the behavior they
   exercise, and free of credentials or private data.
 - Golden files are appropriate only when the complete output is intentionally
@@ -99,7 +112,7 @@ Use each test type where it protects a distinct risk:
 
 ## What deserves tests
 
-Prioritize tests for:
+To the extent the following features are present in the application, prioritize tests for:
 
 1. Public and package-level contracts.
 2. Domain rules and important invariants.
@@ -107,17 +120,29 @@ Prioritize tests for:
 4. Failure handling, cancellation, retries, recovery, and partial success.
 5. Serialization, schemas, compatibility, and round trips.
 6. Previously observed or plausible regressions.
-7. Representative integration and consumer workflows.
+7. Representative integration and end-to-end workflows.
 
-A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
+A package-level contract comprises behavior relied upon by another package or major collaborator.
 
-For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+Choose tests for requested behavior and established contracts. Tests must not
+introduce new restrictions, guarantees, or threat scenarios merely because they
+could be asserted. Follow the [agent policy](agents.md) when deciding whether a
+behavior belongs to the contract; current implementation behavior alone is not
+sufficient authority.
+
+For established behavior involving **data integrity, destructive operations,
+compatibility, security, concurrency, idempotency, or recovery**, favor durable
+tests for realistic failures whose consequences warrant protection. Identify
+the failure and the existing contract it would violate. These categories call
+for attention, not an automatic test inventory or permission to strengthen the
+application's guarantees. Account for protection already provided at other layers.
 
 Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
 
 ## Choose the right test boundary
 
-Test through the narrowest stable boundary that expresses the behavior clearly.
+Choose the narrowest stable boundary that expresses the behavior clearly and
+provides durable confidence with minimal incidental coupling.
 
 This is often the package API, but it may instead be:
 
@@ -125,30 +150,15 @@ This is often the package API, but it may instead be:
 - a package-level operation when several internal collaborators jointly produce the behavior; or
 - a larger integration boundary when correctness emerges from interaction with a real dependency.
 
-Do not force all behavior through oversized consumer-workflow tests. Do not
-test every private helper merely because it exists. Choose the boundary that
-gives durable confidence with the least incidental coupling.
-
 ## Test behavior, not implementation
 
-A test should protect a decision, contract, or invariant—not memorialize the current implementation.
+A test should protect a selected decision, contract, or invariant.
 
-Before adding or retaining a test, ask:
+A test is suspect when it primarily detects changes to private constants or
+helpers, equivalent operation ordering, incidental formatting, equivalent
+algorithms, or internal structure that preserves behavior.
 
-> What realistic defect would this test catch?
-
-A test is suspect when its main purpose is to detect that someone:
-
-- changed an internal constant;
-- renamed or split a private helper;
-- reordered equivalent internal operations;
-- changed incidental formatting;
-- replaced one correct algorithm with another; or
-- refactored internal object structure without changing behavior.
-
-Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
-
-A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
+Behavioral tests should normally remain stable through internal refactoring that preserves the contract.
 
 ## Expected effects of different changes
 
@@ -165,13 +175,13 @@ A test failing is not the same as a test needing to be edited. Many tests may co
 
 ## Separate mechanism from policy
 
-Configurable thresholds and defaults must not be duplicated throughout the test suite.
+Derive expectations from test-controlled configuration and relationships.
 
-The following fragments are illustrative rather than standalone Go programs.
-Do not encode an internal concurrency limit indirectly:
+For this example, assume the selected contract rejects a request to start more
+processes than a configured limit. Do not encode its production default indirectly:
 
 ```go
-// Production policy:
+// Production default:
 const maxConcurrency = 4
 
 // Brittle test:
@@ -189,37 +199,40 @@ require.NoError(t, runner.Start(limit))
 require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
 ```
 
-The test should prove:
+For this selected contract, the test should prove:
 
 - the configured limit is accepted; and
 - one beyond the configured limit is rejected.
 
+Rejection is not a general requirement for concurrency control. If the selected
+contract queues excess work or limits active workers while accepting larger
+batches, test that behavior instead. Do not add rejection or a configurable
+limit solely to reproduce this example.
+
 The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
 
-Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
+Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior.
 
-For concurrency limits, test both kinds of behavior when relevant:
+For established concurrency limits, distinguish these behaviors where the contract requires them:
 
-1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+1. **Configuration handling:** values are accepted or rejected according to the selected contract.
 2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
 
-Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
+Use a test-controlled limit and measure the behavior relative to that limit.
 
-## Avoid semantic duplication across layers
+## Assign each behavior a test owner
 
-Each behavior should have a clear test owner.
+Where these layers exist, assign ownership as follows:
 
 - Parser tests own parsing cases.
 - Validator tests own validation rules.
 - Domain tests own transformations and invariants.
-- Boundary tests own external integration behavior.
+- Adapter tests own external integration behavior.
 - Orchestrator tests own coordination and failure propagation.
-- Consumer-workflow tests prove that representative assembled library use
-  works.
+- External-package root tests own the public Go facade contract.
+- Consumer-workflow tests prove that representative assembled library use works.
 
-Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
-
-Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
+Give each layer its distinct responsibility and keep expectation updates localized to the contract being changed.
 
 ## Use test doubles deliberately
 
@@ -240,111 +253,78 @@ Mocks are appropriate when the contract includes facts such as:
 - an expensive API is called no more than once; or
 - a security audit event is emitted.
 
-Do not use mocks merely to isolate every object or reproduce the implementation's call graph.
-
 ## Go-specific guidance
 
-Use:
+Select techniques by behavior and risk:
 
 - table-driven tests for meaningful behavioral categories and boundaries;
 - `t.TempDir()` for real filesystem behavior;
 - `httptest.Server` for realistic HTTP interactions;
-- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- fuzz tests when exploring broad input spaces, such as parsers or path handling, adds useful protection beyond representative cases;
 - golden files only when the complete output is intentionally stable;
 - integration tests where correctness depends on component interaction; and
-- a small number of representative consumer-workflow tests.
+- a small number of representative end-to-end tests when assembled workflows expose meaningful risks not adequately covered at narrower boundaries.
 
-Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
+Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields. Use exact error-wording assertions only when the wording is contractual.
 
 At public API boundaries, prefer stable error identity, structured values, and
-the smallest semantic fragment needed to identify the failure. Do not snapshot
-complete diagnostic wording unless it is contractual.
+the smallest semantic fragment needed to identify the failure. Use complete
+diagnostic snapshots only when the wording is contractual.
 
-Golden-file updates must require an explicit local flag. Ordinary validation
-runs must never update golden files automatically, and maintainers must inspect
-the semantic diff before accepting an update.
+Golden-file updates must require an explicit local flag. Ordinary validation runs must never update golden files automatically, and reviewers must inspect the semantic diff before accepting an update.
 
-Keep tests readable and direct. Test helpers and fixture frameworks must earn their own maintenance cost; do not build elaborate test infrastructure for small or isolated needs.
+Keep tests readable and direct. Size test helpers and fixture frameworks to the maintenance benefit they provide.
 
 ## Coverage
 
-Coverage is a diagnostic, not a target.
+Use coverage to locate untested critical branches and unexpectedly weak packages. Evaluate test quality through meaningful behavioral protection and choose coverage according to risk.
 
-Use it to find untested critical branches and unexpectedly weak packages. Do not write low-value tests solely to increase a percentage, and do not infer test quality from coverage alone.
-
-Pure domain logic will often warrant higher coverage than facade wiring or
-external adapters. Uneven coverage is acceptable when it reflects risk.
-
-Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
+Pure domain logic will often warrant higher coverage than facade wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
 
 ## Regression tests
 
 A bug fix should normally include a regression test that fails before the fix and passes afterward.
 
-Retain the test when the defect could realistically recur and its consequences justify the ongoing cost. Prefer the narrowest durable test of the violated contract or invariant; do not preserve accidental implementation details from the original bug.
+Retain the test when the defect could realistically recur and its consequences justify the ongoing cost. Prefer the narrowest durable test of the violated contract or invariant.
 
-Not every historical bug requires a permanent test. If the underlying design has made recurrence impossible, the test has become redundant, or a stronger invariant test now subsumes it, remove or consolidate it.
+Remove or consolidate a test when the underlying design has made recurrence impossible, its protection is redundant, or a stronger invariant test subsumes it.
 
 ## Deleting or rewriting tests
 
-Tests are maintained code, not permanent historical artifacts.
-
-Delete or rewrite a test when its maintenance cost exceeds the confidence it provides.
+Delete or rewrite tests when maintenance cost exceeds their continuing
+protective value.
 
 Strong candidates include tests that:
 
 - require updates after harmless internal changes;
-- directly assert private constants without protecting a real contract;
-- duplicate the same policy across several layers;
+- assert incidental private constants or snapshot large amounts of incidental output;
 - verify mock choreography rather than outcomes;
-- snapshot large amounts of incidental output;
-- test trivial private helpers already exercised through stable package behavior;
-- protect risks already covered more effectively elsewhere;
+- duplicate policy across layers, retest trivial helpers covered by package
+  behavior, or protect risks covered more effectively elsewhere;
 - are flaky, misleading, obsolete, or disproportionately expensive to diagnose; or
 - no longer correspond to a plausible failure mode.
 
-Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
-
-Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
+Several brittle tests may encode one genuine requirement. Consolidate them into one durable behavior-level or invariant test.
 
 ## Reviewing a proposed test
 
-Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
+Use the following questions when a proposed test needs closer evaluation. Significant test additions should be reviewable against them; written answers are optional for routine tests.
 
-1. What realistic defect would it catch?
-2. How likely is that defect?
-3. How consequential would it be?
-4. Is the behavior already protected elsewhere?
-5. At which layer should this behavior be owned?
-6. Does the test assert a durable contract or an incidental implementation detail?
-7. Could the implementation be refactored without changing the behavior and without editing this test?
-8. What should cause this test to fail?
-9. What legitimate changes should not cause this test to fail?
-10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
-11. Is there a smaller or more direct test that protects the same risk?
-
-Do not add the test when its expected lifetime cost exceeds its expected protective value.
-
-When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+1. Which requested behavior or established contract does it protect, and what realistic defect would violate it?
+2. How likely and consequential is that defect, and is it already protected elsewhere?
+3. What is the narrowest stable boundary that can test it clearly?
+4. Would a correct internal refactor or a change to an incidental default require test edits?
+5. Does its confidence justify its maintenance, execution, and diagnostic cost, or would a smaller test suffice?
 
 ## Definition of sufficient
 
 A test suite is sufficient when:
 
-- important contracts and invariants are protected;
-- meaningful boundaries and failure modes are exercised;
-- realistic and consequential regressions are credibly protected against silent recurrence;
-- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
-- important external boundaries have realistic integration coverage;
-- representative complete workflows are tested;
-- failures provide useful signal rather than redundant noise;
-- legitimate internal changes usually do not require test edits; and
+- important requested behavior and established contracts are protected against realistic, consequential failures;
+- boundaries and component interactions have coverage where they introduce meaningful risks not protected elsewhere;
+- failures provide useful, distinct signals;
+- tests usually remain stable through legitimate internal changes; and
 - additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
 
-Sufficiency is a risk judgment, not a coverage percentage or test count.
-Reassess it as the library, its consumers, and the consequences of failure
-evolve.
-
-The governing rule is:
-
-> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.
+Reassess sufficiency as the application, its users, and the consequences of
+failure evolve.

docs/release.md

@@ -1,278 +1,138 @@
 # Release Procedure
 
-## Release Model
+This runbook implements the [release policy](policy/release.md). Promptkit
+publishes source-only Go module tags from `main` to `origin`. A successful push
+of the selected annotated tag, after its source commit is available on `main`,
+completes the release. Do not wait for CI or Go module availability.
 
-Promptkit publishes a Go library through source commits and semantic Go module
-tags. It does not publish runnable binaries or binary packages and does not
-currently use hosted CI. The release maintainer performs and records the
-required validation.
+## Prepare The Candidate
 
-`v0.1.0` is the initial published release. Later releases use semantic
-`vMAJOR.MINOR.PATCH` tags. Before `v1`, minor releases may change the public
-API and patch releases preserve compatibility within their minor line. Every
-pre-`v1` release note must summarize compatibility, identify public API
-changes, and state any action required of consumers.
+Required tools are Git, Go with race-detector support, POSIX shell, and Python
+3. Work from the repository root with push access to `origin`.
 
-Promptkit releases are source-only. The annotated tag message is the release
-note; there is no separate hosted release or binary packaging step.
-
-## Establish The Candidate
-
-Choose a version that has not been published and export it as
-`RELEASE_VERSION`. Run every command in this procedure from the Promptkit
-repository root in the same POSIX shell. Do not reuse `v0.1.0` or another
-existing version.
-
-The following guard derives the release commit from `HEAD` and stops on a
-missing or malformed version, a checkout other than synchronized `main`,
-uncommitted changes, an active Go workspace, a module replacement, a vendor
-tree, or an existing local or remote tag:
+Use the latest Go installation available locally for the release tool, the
+candidate validation, and the maintained examples. Make it the active `go`
+command and record its version:
 
 ```sh
-set -eu
-
-: "${RELEASE_VERSION:?export an unpublished vMAJOR.MINOR.PATCH version}"
-if ! printf '%s\n' "$RELEASE_VERSION" |
-    grep -Eq '^v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$'
-then
-    printf '%s\n' "invalid release version: $RELEASE_VERSION" >&2
-    exit 1
-fi
-
-RELEASE_COMMIT=$(git rev-parse --verify 'HEAD^{commit}')
-export RELEASE_COMMIT
-
-check_release_candidate() {
-    test "$(git branch --show-current)" = main
-    test -z "$(git status --porcelain)"
-
-    gowork_value=$(go env GOWORK)
-    case "$gowork_value" in
-        ''|off) ;;
-        *)
-            printf '%s\n' "active Go workspace: $gowork_value" >&2
-            return 1
-            ;;
-    esac
-
-    test -z "$(git ls-files go.work go.work.sum)"
-    test ! -e vendor
-    if grep -Eq '^[[:space:]]*replace([[:space:]]|\()' go.mod
-    then
-        printf '%s\n' 'go.mod contains a replacement' >&2
-        return 1
-    fi
-
-    git fetch origin main --tags
-    test "$RELEASE_COMMIT" = \
-        "$(git rev-parse --verify 'refs/remotes/origin/main^{commit}')"
-
-    if git show-ref --verify --quiet "refs/tags/$RELEASE_VERSION"
-    then
-        printf '%s\n' "local tag already exists: $RELEASE_VERSION" >&2
-        return 1
-    fi
-    if test -n "$(
-        git ls-remote --tags origin \
-            "refs/tags/$RELEASE_VERSION" \
-            "refs/tags/$RELEASE_VERSION^{}"
-    )"
-    then
-        printf '%s\n' "remote tag already exists: $RELEASE_VERSION" >&2
-        return 1
-    fi
-}
-
-check_release_candidate
+export GOTOOLCHAIN=local
+go version
 ```
 
-Do not continue unless the guard completes successfully. In particular, push
-the intended commit through the normal `main` branch workflow before release;
-the tag procedure is not a substitute for publishing the source commit.
-
-## Validate The Candidate
-
-Confirm the module and root package metadata:
+For a minor release, advance Promptkit's consumer minimum to the initial stable
+release of that Go series before committing the candidate:
 
 ```sh
-go list -m -f '{{.Path}} {{.GoVersion}}'
-go list -f '{{.Name}} {{.ImportPath}}' .
+release_go_version=$(go env GOVERSION)
+release_go_series=${release_go_version#go}
+release_go_series=${release_go_series%.*}
+go mod edit "-go=${release_go_series}.0"
 ```
 
-The output must be:
+For example, Go 1.26.8 normally results in `go 1.26.0`. Keep a higher patch
+minimum if a dependency or identified correctness fix requires it. For a patch
+release, leave the consumer minimum unchanged; using a newer validation
+toolchain does not itself raise that minimum. If the candidate needs a higher
+minimum, prepare a minor release instead. See the
+[consumer-minimum policy](policy/release.md#consumer-minimum-and-release-toolchain).
 
-```text
-gitea.maximumdirect.net/eric/promptkit 1.25.5
-promptkit gitea.maximumdirect.net/eric/promptkit
-```
-
-Run the complete maintainer validation required by the
-[development guide](development.md):
+Prepare module caches before validation; this may update `go.sum` and belongs
+before the candidate commit:
 
 ```sh
-go test ./...
-go test -race ./...
-go vet ./...
-go build ./...
-go run ./examples/go-library/prepare
+GOTOOLCHAIN=local go mod download
 ```
 
-Check every tracked Go file. This command must produce no output:
+Choose an unused stable version and create `docs/releases/<version>.md` with a
+title containing `Promptkit` and that exact version, a short summary, and any
+relevant compatibility impact or upgrade instructions. No fixed section
+headings are required. Describe a raised Go minimum and its compatibility
+impact under the [versioning policy](policy/release.md#versioning-and-compatibility).
+End each new note with the licensing footer below, using the selected version:
+
+```markdown
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/vX.Y.Z/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/vX.Y.Z/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/vX.Y.Z/docs/oss/attribution.md)
+```
+
+Check the footer paths against the candidate; the future tag need not resolve
+remotely during preparation. Follow the [OSS policy](policy/oss.md) for these
+links. Promptkit publishes no binaries, hosted download page, or other
+asynchronous deliverables.
+
+Retire completed roadmaps while preserving deferred work and repairing incoming
+links. Review the intended changes through the ordinary development workflow,
+commit the candidate on `main`, and push `main` to `origin` before publication.
+Preparation may update module metadata; the release bootstrap and validation
+preserve the committed candidate.
+
+## Validate And Publish
+
+For an optional local diagnostic of a committed candidate, run:
 
 ```sh
-unformatted=$(
-    git ls-files '*.go' |
-        while IFS= read -r go_file
-        do
-            gofmt -l "$go_file"
-        done
-)
-test -z "$unformatted"
+./scripts/releasekit.sh check vMAJOR.MINOR.PATCH
 ```
 
-Follow every maintained Markdown link and confirm that its local or published
-target exists. Review the repository for generated binaries, test or coverage
-output, credentials, template residue, downloaded assets, and other files that
-do not belong in source control.
+This is not a required step before publication. The upstream
+[candidate-check contract](https://gitea.maximumdirect.net/eric/ReleaseKit/src/tag/v0.4.1/docs/consumers/check.md)
+describes its requirements.
 
-Recheck module and repository hygiene, whitespace, and the clean checkout:
+Run the single publication command with the chosen version:
 
 ```sh
-test -z "$(git ls-files go.work go.work.sum)"
-test ! -e vendor
-if grep -Eq '^[[:space:]]*replace([[:space:]]|\()' go.mod
-then
-    printf '%s\n' 'go.mod contains a replacement' >&2
-    exit 1
-fi
-git diff --check
-test -z "$(git status --porcelain)"
+./scripts/release.sh vMAJOR.MINOR.PATCH
 ```
 
-## Write The Release Note
+The familiar wrapper delegates to the native ReleaseKit executable selected in
+`go.mod`. The committed [profile](../releasekit.json) selects the race-enabled
+Go suite, full vet, committed Go formatting, local Markdown links, workspace
+and filesystem replacement inspection, and semantic checks for the three
+maintained offline consumer examples. ReleaseKit also verifies the committed
+candidate and publication state, then publishes only the selected tag. Its
+checks run once for this candidate; do not repeat them as a separate sign-off.
 
-Prepare a plain-text annotated-tag message outside the repository and export
-its path as `RELEASE_NOTES_FILE`. Use this form, replacing each summary with
-release-specific text; write `None.` when there are no public API changes or
-consumer actions:
+The native command returns status 0 for confirmed publication, 1 for a known
+validation or publication failure, 2 for invalid input or profile, and 3 when
+the push outcome remains unknown. Successful publication needs no remote
+read-back, CI wait, or installation check. Provider probes and inference calls
+are not part of validation.
 
-```text
-Promptkit vMAJOR.MINOR.PATCH
+## Catalog Profile Compatibility
 
-Validated commit: full commit ID
-Compatibility: compatibility summary
-Public API changes: changes or None.
-Consumer action: required action or None.
-```
+The [release policy](policy/release.md#catalog-profile-compatibility) owns the
+catalog compatibility boundary. This heading preserves older incoming links.
 
-After writing it, require all release-note fields, the selected version, and
-the validated commit to be present:
+## Failure Recovery
+
+For status 1, use the diagnostic to distinguish a validation failure from a
+known publication conflict. Correct and commit candidate defects, push `main`,
+and rerun publication with an unused version when appropriate. Status 2 means
+the invocation or profile must be corrected.
+
+For status 3, do not retry or remove the local tag until the remote outcome is
+resolved. ReleaseKit reports the attempted version, tag object, and recovery
+state, including the captured publication endpoint when available. Use that
+endpoint—not merely the current `origin` URL—to query the exact tag; follow the
+upstream
+[failure-recovery procedure](https://gitea.maximumdirect.net/eric/ReleaseKit/src/tag/v0.4.1/docs/release.md#resolve-failures):
 
 ```sh
-: "${RELEASE_NOTES_FILE:?export the path to the release-note file}"
-test -f "$RELEASE_NOTES_FILE"
-test -s "$RELEASE_NOTES_FILE"
-grep -F "Promptkit $RELEASE_VERSION" "$RELEASE_NOTES_FILE"
-grep -F "Validated commit: $RELEASE_COMMIT" "$RELEASE_NOTES_FILE"
-grep -F 'Compatibility:' "$RELEASE_NOTES_FILE"
-grep -F 'Public API changes:' "$RELEASE_NOTES_FILE"
-grep -F 'Consumer action:' "$RELEASE_NOTES_FILE"
+releasekit_endpoint='COPY_CAPTURED_ENDPOINT'
+releasekit_version='vX.Y.Z'
+test "$(git ls-remote --get-url -- "$releasekit_endpoint")" = "$releasekit_endpoint" &&
+  git ls-remote --refs -- "$releasekit_endpoint" "refs/tags/$releasekit_version"
 ```
 
-Inspect the complete message and confirm that it accurately records the
-compatibility impact, public API changes, and required consumer action.
-
-## Create And Inspect The Tag
-
-Run the candidate guard again immediately before tag creation. This ensures
-that validation or release-note preparation did not change the checkout and
-that the commit is still published and untagged:
-
-```sh
-check_release_candidate
-```
-
-Create the annotated tag from the prepared release note and bind it explicitly
-to the validated commit:
-
-```sh
-git tag --annotate "$RELEASE_VERSION" \
-    --file "$RELEASE_NOTES_FILE" \
-    "$RELEASE_COMMIT"
-```
-
-Inspect both the tag message and its source commit before publication:
-
-```sh
-test "$(git cat-file -t "refs/tags/$RELEASE_VERSION")" = tag
-git show --no-patch --decorate "refs/tags/$RELEASE_VERSION"
-test "$(
-    git rev-parse --verify "refs/tags/$RELEASE_VERSION^{commit}"
-)" = "$RELEASE_COMMIT"
-```
-
-If inspection finds an error, delete the unpublished local tag, correct the
-release note or candidate, and repeat the guards. Never move or recreate a tag
-that has been published.
-
-## Publish The Selected Tag
-
-Push only the selected tag ref. Do not use `git push --tags`:
-
-```sh
-git push origin \
-    "refs/tags/$RELEASE_VERSION:refs/tags/$RELEASE_VERSION"
-```
-
-## Verify Publication
-
-Compare the remote annotated-tag object with the local object, then compare the
-remote peeled source commit with the validated commit:
-
-```sh
-remote_tag=$(
-    git ls-remote --tags origin "refs/tags/$RELEASE_VERSION" |
-        awk 'NR == 1 { print $1 }'
-)
-remote_commit=$(
-    git ls-remote --tags origin "refs/tags/$RELEASE_VERSION^{}" |
-        awk 'NR == 1 { print $1 }'
-)
-test -n "$remote_tag"
-test "$remote_tag" = \
-    "$(git rev-parse --verify "refs/tags/$RELEASE_VERSION")"
-test "$remote_commit" = "$RELEASE_COMMIT"
-```
-
-Finally, resolve the version as an ordinary Go module in a temporary module
-outside this repository and without a workspace or replacement:
-
-```sh
-resolution_dir=$(mktemp -d)
-(
-    trap 'rm -rf "$resolution_dir"' 0 1 2 15
-    cd "$resolution_dir"
-    GOWORK=off go mod init example.com/promptkit-release-check
-    GOWORK=off go mod download \
-        "gitea.maximumdirect.net/eric/promptkit@$RELEASE_VERSION"
-    resolved_version=$(
-        GOWORK=off go list -m -f '{{.Version}}' \
-            "gitea.maximumdirect.net/eric/promptkit@$RELEASE_VERSION"
-    )
-    test "$resolved_version" = "$RELEASE_VERSION"
-)
-```
-
-Promptkit must publish and verify the required version before Scriptorium or
-another consumer publishes a release that depends on it. This ordering does
-not replace the consumer project's own release procedure. Released consumers
-must select the published Promptkit tag through ordinary module resolution,
-without a workspace, replacement, vendored Promptkit source, or unpublished
-revision.
+Compare the returned object ID with the reported tag object. An exact match
+confirms publication. A missing ref confirms absence only when the endpoint
+identity check and query both succeed. If either check fails, the outcome
+remains unknown; correct the transport configuration and resolve it before
+retrying. Only after authoritative absence is confirmed may an unpublished
+local tag be removed before correcting the candidate or retrying. Never move,
+overwrite, or delete a published tag.
 
 ## Policy Changes
 
-Document and approve a durable policy change before introducing hosted
-automation, binary artifacts, or different release governance. Update this
-procedure in the same change so maintainers do not rely on hidden release
-requirements.
+Update the policy and this procedure together when changing release governance.
+Consumers own their dependency-availability and release checks; their readiness
+does not delay completion of a Promptkit release.

docs/releases/v0.10.0.md

@@ -0,0 +1,44 @@
+# Promptkit v0.10.0
+
+This supplemental guide summarizes the `v0.10.0` changes and migration needs.
+The annotated tag remains the authoritative release record.
+
+## Transition Summary
+
+`v0.10.0` adopts independently versioned OpenRouter and Rakestrawhome catalog
+modules whose external profile data can change between catalog releases. It
+also introduces the maintained literal names
+`promptkit-openrouter-standard` and `promptkit-openrouter-auto`.
+
+The public package now delegates assembly to an internal engine while retaining
+the existing Go API. This refactor requires no consumer code changes.
+
+This release selects OpenRouter `v2.20260922.1` and Rakestrawhome
+`v2.20260922.0`; their exact inventories are linked from the
+[format reference](../formats.md#external-profile-inventories).
+
+The maintained names provide continuity only within the catalog versions
+selected and tested by a Promptkit release and when consumers do not override
+the name or its base. Their current selections may change. External catalog
+profile IDs can be removed, renamed, or modified, so an application that
+selects one directly may need to update its configuration when upgrading.
+
+## Consumer Migration
+
+Before upgrading, review every profile ID used by prompts, requests, and
+application-owned inheritance. Choose a maintained name, accept an external
+catalog profile's volatility, or define a complete application-owned profile
+when the application must control its target and settings. An alias that
+inherits from an external base is not a frozen copy.
+
+In particular, the selected OpenRouter catalog removes `deepseek-3-2`,
+`deepseek-4-flash`, and `deepseek-4-pro`. Consumers using those IDs must choose
+an available profile or supply their own complete definition; the new
+`deepseek-flash-latest` and `deepseek-pro-latest` profiles are not automatic
+aliases and need not preserve the previous settings.
+
+See the [release policy](../release.md#catalog-profile-compatibility) for the
+compatibility boundary, the [format reference](../formats.md#built-in-profile-catalog)
+for current Promptkit-owned definitions and selected external inventories, and
+the [consumer guide](../consumers/pkg-promptkit.md#choose-a-profile-ownership-model)
+for configuration choices.

docs/releases/v0.11.0.md

@@ -0,0 +1,76 @@
+# Promptkit v0.11.0
+
+This release adds reliable, observable model execution and changes Promptkit's
+license to LGPL-3.0-or-later.
+
+## Summary
+
+- Built-in model calls can recover from selected transient HTTP failures.
+- Optional execution observations report attempts, timing, outcomes, and known
+  token usage, including work completed before a run fails.
+- Public errors distinguish transport failures, malformed responses, and retry
+  exhaustion without parsing diagnostic text.
+
+## Compatibility
+
+The built-in client now defaults to two additional transient attempts. This can
+increase latency, request counts, and cost; retries do not guarantee exactly-once
+generation. Injected `LLMClient` implementations retain their own retry policy
+and are not automatically replayed. Existing generation timeouts apply per
+attempt, while the caller context governs the complete operation. See the
+[provider integration contract](../integrations/openai-compatible-chat.md).
+
+Public retry and observation controls are additive. Existing successful-result
+usage semantics remain unchanged, and operational failures still return no
+partial `RunResult`. Observations distinguish unknown usage from reported zero;
+they are not a complete billing ledger. See the
+[consumer guide](../consumers/pkg-promptkit.md) and
+[observation GoDoc](../../observations.go).
+
+Promptkit is now licensed under LGPL-3.0-or-later. Review the
+[license summary](../../README.md#license), [LGPL text](../../LICENSE), and
+[accompanying GPL text](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.0/COPYING) before adoption or distribution.
+Previously published releases and dependency licenses are unchanged.
+
+## Upgrade
+
+Update the dependency:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.11.0
+```
+
+To retain single-attempt behavior for built-in generation, set
+`Config.TransientRetries` to a pointer to zero. `RunRequest.TransientRetries`
+can override the engine choice per request. Review any existing application or
+HTTP-transport retry loops to avoid unintentionally multiplying attempts.
+Transient retries and structural repairs have separate budgets; provide a
+caller deadline when a total execution time limit is needed.
+
+Observation is opt-in through `RunRequest.Observer`, including for prepared
+execution. Callbacks are synchronous; shared observers must support concurrent
+runs. Injected clients can supply `GenerateResponse.UsagePresence` to explicitly
+report known zero counts. No observer or usage-presence changes are required
+for existing consumers. See the [consumer guide](../consumers/pkg-promptkit.md)
+for adoption and the [offline example](../../examples/go-library/observe/main.go)
+for a structured proposal-and-assessment workflow.
+
+## Changes
+
+- Centralized built-in retry coordination with cancellable backoff and provider
+  retry timing. Retries and structural repair share existing backend capacity;
+  backoff releases active permits without releasing logical run admission.
+- Added `ErrTransport`, `ErrMalformedResponse`, `ErrRetriesExhausted`, and
+  `RetryExhaustedError`, retaining applicable underlying error identities.
+- Added request-scoped lifecycle observations and presence-aware token accounting
+  for ordinary and prepared execution, without recording prompt/response bodies.
+- Isolated delegated operations' observation state and made injected-client
+  attempt classifications consistent with terminal classifications.
+- Added an offline observation example and regression coverage for retry,
+  repair, capacity, error, and observation interactions.
+- Adopted LGPL-3.0-or-later and streamlined the documented local release process.
+
+Backend catalog selections are unchanged from v0.10.0: OpenRouter
+`v2.20260922.1` and Rakestrawhome `v2.20260922.0`. See the
+[catalog compatibility policy](../policy/release.md#catalog-profile-compatibility)
+and [profile inventories](../formats.md#external-profile-inventories).

docs/releases/v0.11.1.md

@@ -0,0 +1,43 @@
+# Promptkit v0.11.1
+
+This patch updates the maintained catalogs to MIT-licensed releases and
+improves licensing documentation.
+
+## Summary
+
+Both OpenRouter and Rakestrawhome catalogs now use `v2.20260926.0`. These
+catalog releases change licensing and documentation only; their profiles,
+backend definitions, and Go code are unchanged.
+
+## Compatibility
+
+No public API, runtime behavior, or profile changes are introduced.
+Promptkit remains LGPL-3.0-or-later; the catalogs are now MIT-licensed.
+See the [project grant](../../COPYRIGHT),
+[component attribution](../oss/attribution.md), and
+[catalog compatibility policy](../policy/release.md#catalog-profile-compatibility).
+
+## Upgrade
+
+Update the module dependency:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.11.1
+```
+
+No code or configuration migration is required from v0.11.0.
+Consumers upgrading from earlier versions should also read the
+[v0.11.0 migration guide](v0.11.0.md).
+
+## Changes
+
+- Adopt the MIT-licensed catalog releases and update inventory links.
+- Add centralized attribution and dependency license texts.
+- Separate the project grant into COPYRIGHT and consolidate the LGPLv3 and
+  incorporated GPLv3 texts in LICENSE, replacing the separate COPYING file.
+- Align documentation ownership and future release navigation with the
+  [OSS policy](../policy/oss.md), without adding release gates.
+- Repair the historical v0.11.0 GPL-text link and retire the completed
+  licensing roadmap.
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.1/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.1/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.1/docs/oss/attribution.md)

docs/releases/v0.11.2.md

@@ -0,0 +1,38 @@
+# Promptkit v0.11.2
+
+This patch aligns repository policies with the current standard and clarifies
+licensing and attribution for the copied policy documents.
+
+## Summary
+
+The update adds policy provenance, selective contributor reading guidance, and
+consistent upstream policy language while preserving Promptkit-specific rules.
+
+## Compatibility
+
+No public API, runtime behavior, dependency, or catalog changes are introduced.
+Promptkit remains LGPL-3.0-or-later. The [project grant](../../COPYRIGHT)
+identifies the separate CC-BY-SA-4.0 scope for copied or adapted policy material;
+the [attribution notices](../oss/attribution.md) identify the affected documents.
+The existing release checks and successful-tag-push completion point are unchanged.
+
+## Upgrade
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.11.2
+```
+
+No code or configuration migration is required from v0.11.1.
+
+## Changes
+
+- Align adopted policies with Standard Repository Policies while retaining
+  library-specific requirements and deliberate local decisions.
+- Add per-document upstream source links and the
+  [policy-set provenance document](../policy/meta.md).
+- Add selective policy reading to the [contributor guide](../development.md)
+  and update agent orientation.
+- Standardize factual attribution entries and identify copied-policy licensing.
+- Retire the completed policy-alignment roadmap.
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.2/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.2/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.11.2/docs/oss/attribution.md)

docs/releases/v0.12.0.md

@@ -0,0 +1,74 @@
+# Promptkit v0.12.0
+
+This release replaces Promptkit's configuration loader with ConfigKit and
+migrates maintained catalogs to keyed `/v3` assets. It intentionally changes
+the public construction API and configuration format.
+
+## Summary
+
+Backends, profiles, and versioned prompts now share one layered configuration
+snapshot built when `NewEngine` is called. ConfigKit supplies composition,
+inheritance, presence, and source history; Promptkit validates the final
+effective definitions and captures prompt content and templates. Consumers can
+inspect the captured configuration explicitly without triggering a reload.
+
+## Compatibility
+
+This is an incompatible minor release under Promptkit's pre-1.0
+[versioning policy](../policy/release.md#versioning-and-compatibility). The
+former source options, `Config.PromptDir`, `Config.ProfileDir`, and the old
+profile/backend convenience values have been removed. Configuration documents
+now use literal keys beneath `backends`, `profiles`, and `prompts`; embedded
+`id`/`version` fields and `base_profile` are no longer accepted. Source groups
+compose strictly within a group and overlay in the supplied order. Explicit
+zero, false, empty, and null values can now differ from omission; maps overlay
+recursively, and an empty map does not clear an inherited map.
+
+All final definitions, including unused ones, are validated during engine
+construction. Prompt text and static templates are captured then; an existing
+engine does not automatically reload changed configuration or message files.
+Request artifacts, schemas, credentials, and provider operations retain their
+runtime boundaries. See the [format reference](../formats.md) for the current
+configuration contract and the [consumer migration guide](../consumers/pkg-promptkit.md#migrate-from-the-previous-configuration-api)
+for API and source-selection changes.
+
+Maintained OpenRouter and Rakestrawhome catalogs are selected as `/v3` modules,
+both at `v3.20260930.0`. Their asset format changed; consumers that read catalog
+assets directly must adopt the keyed format. The maintained Promptkit profile
+IDs remain available under the [catalog compatibility policy](../policy/release.md#catalog-profile-compatibility),
+which does not promise fixed profile data.
+
+## Upgrade
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.12.0
+```
+
+Replace removed source and backend registration options with
+`Config.Sources`, explicitly ordered ConfigKit underlays, and optionally a
+highest-priority main input. Use `DirectoryInputs` or `FilesystemInputs` for
+selected directories, `ProfileSource` and `BackendSource` for typed
+programmatic values, or `configkit.Mapping` for explicit null and combined
+documents. Move profile `base_profile` to `extends` and embedded identities to
+mapping keys. Review zero/false/null values and any assumptions that replacing
+a model or backend discards unrelated explicit settings. Rebuild the engine
+when definitions or referenced message files change. The
+[consumer guide](../consumers/pkg-promptkit.md#migrate-from-the-previous-configuration-api)
+provides the migration details and maintained examples.
+
+## Changes
+
+- Add ConfigKit v0.5.0 as the configuration boundary and consume released
+  OpenRouter and Rakestrawhome `/v3` catalog modules.
+- Add keyed backend, profile, and prompt definitions with final-value domain
+  validation, explicit presence/null behavior, and source-relative content
+  capture.
+- Add `ConfigurationSources`, directory discovery helpers, programmatic
+  profile/backend source builders, and read-only configuration inspection.
+- Preserve execution workflows while rendering captured templates, applying
+  backend defaults after inheritance, and honoring explicit suppression of
+  provider settings.
+- Remove superseded loader, catalog preflight, and inheritance machinery;
+  update examples, format guidance, and internal documentation.
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.12.0/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.12.0/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.12.0/docs/oss/attribution.md)

docs/releases/v0.13.0.md

@@ -0,0 +1,69 @@
+# Promptkit v0.13.0
+
+This release adds prompt requirement discovery and profile/prompt listing,
+reduces avoidable configuration and execution allocations, and standardizes
+maintainer publication through ReleaseKit. The minimum supported Go version
+increases to 1.26.0.
+
+## Discover Requirements And Available Definitions
+
+- `Engine.InspectRequirements` reports recognizable input and caller-variable
+  dependencies for a selected prompt and supplied caller context before input
+  artifacts exist. It classifies dependencies as required, conditional, or
+  optional and reports structured limitations when discovery is incomplete.
+  Discovery coverage is not proof that requirements are satisfied or execution
+  will succeed. See the [requirement inspection guide](../consumers/pkg-promptkit.md#inspect-requirements-before-preparation)
+  and [GoDoc](../../requirements.go).
+- `Engine.ListProfiles` returns effective profile IDs, and `Engine.ListPrompts`
+  returns exact ID/version pairs as `PromptIdentity` values. Lists come from the
+  engine's captured configuration, are lexically ordered and caller-owned, and
+  include consumer definitions and maintained profiles that remain effective.
+  Listing performs no provider or execution work. See the
+  [discovery guide](../consumers/pkg-promptkit.md#discover-available-profiles-and-prompts)
+  and [GoDoc](../../listing.go).
+
+## Efficiency And Tooling
+
+- Reduce response-sized copies in basic validation and rendered-message output,
+  and encode valid provider extra parameters only as part of the complete
+  request rather than serializing them twice.
+- Use bounded initial file-size reservations without limiting accepted content,
+  releasing excess backing storage for small or empty completed artifacts.
+- Reuse owned preparation values when creating prepared-execution snapshots,
+  while preserving independent inspection details and caller-owned data.
+- Use lightweight captured profile identities and prompt metadata for inspection
+  instead of copying execution targets and definitions unnecessarily.
+- Upgrade ConfigKit to v0.6.0 for more efficient loading, resolution, metadata,
+  and inspection processing. Configuration semantics remain unchanged; these
+  improvements primarily benefit construction and explicit operator inspection,
+  not upstream inference latency.
+- Adopt module-selected ReleaseKit v0.3.0 for maintainer validation and tag
+  publication. The [release procedure](../release.md) owns the current commands
+  and recovery workflow; successful tag publication remains the completion
+  point, without waiting for CI.
+
+These changes preserve existing execution and ownership contracts. Benefits
+depend on workload; this release introduces no fixed performance guarantee.
+
+## Compatibility And Upgrade
+
+Consumers upgrading from v0.12.0 need **Go 1.26.0 or newer**, rather than the
+previous 1.25.5 minimum. Update application build environments and any CI images
+still using Go 1.25 before upgrading:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.13.0
+```
+
+No application-code or configuration migration is required from v0.12.0 beyond
+the Go toolchain update. The new discovery methods are optional additions;
+existing construction, inspection, preparation, and execution workflows remain
+available. See [Go compatibility](../consumers/pkg-promptkit.md#go-compatibility)
+for the distinction between the consumer minimum and maintainer toolchain.
+
+Maintainers should follow the current release runbook rather than the retired
+`scripts/release-check.sh`. The publication entry point remains
+`scripts/release.sh`; ReleaseKit now owns its shared validation and publication
+mechanics.
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.13.0/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.13.0/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.13.0/docs/oss/attribution.md)

docs/releases/v0.14.0.md

@@ -0,0 +1,48 @@
+# Promptkit v0.14.0
+
+This release adds comparison of captured configuration inspections and updates
+ConfigKit and the maintainer release tool. The minimum supported Go version
+remains 1.26.0.
+
+## Compare Captured Configurations
+
+`ConfigurationInspection.Diff` compares the effective configuration captured by
+two engines. Consumers can use it to review a candidate configuration, inspect
+an overlay's effect, or present configuration changes to operators.
+
+Comparison delegates to ConfigKit and returns its difference records, including
+presence information, paired redaction, and protected-change indicators. It
+does not reload sources or contact providers. Source-history differences alone
+do not produce value differences.
+
+An empty diff is not proof of equivalent execution: referenced file contents,
+resolved credentials, later execution preparation, and request-specific
+overrides are outside the comparison. Records can contain ordinary application
+data; callers retain responsibility for their disclosure.
+
+See the [configuration comparison guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration)
+and [GoDoc](../../configuration_inspection.go) for usage and the exact contract.
+
+## Dependency And Tooling Updates
+
+- Upgrade ConfigKit from v0.6.0 to v0.7.1, providing the upstream comparison API.
+  v0.7.1 also updates ConfigKit's maintainer tooling without changing its runtime
+  behavior from v0.7.0.
+- Upgrade the module-selected ReleaseKit tool from v0.3.0 to v0.4.1. Promptkit
+  retains its existing root-module profile, publication entry point, and release
+  gates. See the [release procedure](../release.md) for maintainer commands.
+
+## Compatibility And Upgrade
+
+No application-code, configuration, or Go-toolchain migration is required from
+v0.13.0. Configuration comparison is an optional API addition; existing
+construction, inspection, preparation, and execution workflows remain available.
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.14.0
+```
+
+See [Go compatibility](../consumers/pkg-promptkit.md#go-compatibility) for the
+distinction between the consumer minimum and maintainer toolchain.
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.14.0/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.14.0/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.14.0/docs/oss/attribution.md)

docs/releases/v0.15.0.md

@@ -0,0 +1,34 @@
+# Promptkit v0.15.0
+
+This release adds error-only validation of a concrete prompt request without
+calling a model. The minimum supported Go version remains 1.26.0.
+
+## Validate A Concrete Request
+
+`Engine.ValidateRequest` accepts an ordinary `RunRequest` and returns nil when
+it passes preparation, or the existing preparation error. Consumers can use it
+for fail-fast checks, including downstream CLI validation commands, without
+receiving a prepared snapshot.
+
+The method shares `Prepare`'s request conversion and preparation path. Unlike
+requirements inspection, it checks the actual supplied values and bindings.
+Artifact and schema resource I/O still occurs; success does not certify provider
+acceptance or later execution success. A subsequent run prepares independently.
+Validation is optional, not a required step before preparation or execution.
+
+See the [request validation guide](../consumers/pkg-promptkit.md#validate-a-concrete-request)
+and [GoDoc](../../engine.go) for usage and the exact contract. Consumers that
+need execution tied to a checked snapshot can use the existing
+[prepared execution workflow](../consumers/pkg-promptkit.md#prepare-now-and-execute-the-same-snapshot-later).
+
+## Compatibility And Upgrade
+
+No application-code, configuration, or Go-toolchain migration is required from
+v0.14.0. This is an optional API addition; existing inspection, preparation, and
+execution workflows remain available.
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.15.0
+```
+
+[COPYRIGHT](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.15.0/COPYRIGHT) | [LICENSE](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.15.0/LICENSE) | [ATTRIBUTION](https://gitea.maximumdirect.net/eric/promptkit/src/tag/v0.15.0/docs/oss/attribution.md)

docs/releases/v0.2.0.md

@@ -90,7 +90,7 @@ engine, err := promptkit.NewEngine(
 ```
 
 See the
-[custom-backend consumer guide](../consumers/pkg-promptkit.md#register-a-custom-backend)
+[local-endpoint consumer guide](../consumers/pkg-promptkit.md#configure-a-local-openai-compatible-endpoint)
 for task-oriented usage. The
 [`Backend` and `WithBackend` GoDoc](../../backends.go) owns exact registration,
 validation, copying, defaulting, and uniqueness semantics. The

docs/releases/v0.3.0.md

@@ -0,0 +1,74 @@
+# Promptkit v0.3.0
+
+This supplemental changelog summarizes the consumer-facing changes from
+`v0.2.0` to `v0.3.0`. The annotated `v0.3.0` tag is the authoritative release
+record. Exact current contracts belong to the linked GoDoc and durable
+documentation.
+
+## Summary
+
+`v0.3.0` adds a concise way to register the common local OpenAI-compatible
+backend configuration:
+
+- `BackendLocal` provides the conventional, non-reserved backend ID `"local"`;
+  and
+- `LocalBackend` constructs an ordinary `Backend` from an endpoint and
+  concurrency limit.
+
+The helper is explicit and additive. It does not pre-register a backend, read
+environment variables, select a model, or replace the complete `Backend`
+configuration interface.
+
+## Compatibility
+
+Existing `v0.2.0` consumers require no migration. Endpoint-only profiles,
+complete custom `Backend` values, the built-in OpenRouter backend, and existing
+registrations using the literal ID `"local"` continue to work unchanged.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.3.0
+go mod tidy
+```
+
+Run the consuming project's ordinary tests and race-enabled tests after the
+upgrade.
+
+## Configure A Local Backend
+
+Register the convenience value through the existing `WithBackend` option and
+select it from one or more profiles:
+
+```go
+engine, err := promptkit.NewEngine(
+	promptkit.Config{PromptDir: "prompts"},
+	promptkit.WithBackend(
+		promptkit.LocalBackend("http://localhost:8000/v1", 2),
+	),
+	promptkit.WithProfiles(promptkit.Profile{
+		ID:        "local-summary",
+		BackendID: promptkit.BackendLocal,
+		Model:     "example-model",
+	}),
+)
+```
+
+Use an endpoint-only profile when shared backend identity and capacity policy
+are unnecessary. Continue to use a complete keyed `Backend` value for custom
+IDs, authentication, extra request parameters, explicit queue capacity, or
+multiple local endpoints.
+
+See the
+[local-endpoint consumer guide](../consumers/pkg-promptkit.md#configure-a-local-openai-compatible-endpoint)
+for task-oriented configuration choices. The
+[`BackendLocal`, `LocalBackend`, and `WithBackend` GoDoc](../../backends.go)
+owns their exact construction, registration, validation, and concurrency
+semantics.
+
+## Consumer Action
+
+None. Adopt the convenience constructor when it simplifies local endpoint
+configuration.

docs/releases/v0.4.0.md

@@ -0,0 +1,189 @@
+# Promptkit v0.4.0
+
+This supplemental changelog and adoption guide summarizes the consumer-facing
+changes from `v0.3.0` to `v0.4.0`. The annotated `v0.4.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.4.0` adds four complementary capabilities:
+
+- opaque prepared-execution handles for preparing once, inspecting safe
+  details, and executing the same frozen snapshot;
+- exact prompt-definition inspection without profile resolution or execution;
+- exact profile inspection without selecting a prompt or checking credential
+  availability; and
+- structured backend identity on engine admission-capacity rejection.
+
+These APIs let consumers perform more precise preflight work and retain useful
+operational context without reproducing Promptkit's internal resolution logic.
+
+## Compatibility
+
+The release is additive for `v0.3.0` consumers. Existing uses of `Prepare`,
+`Run`, backend registration, endpoint-only profiles, local-backend helpers,
+runtime overrides, public JSON values, and error sentinels continue to work
+without migration.
+
+Capacity rejection now returns a structured error while continuing to match
+`ErrCapacityExceeded` through `errors.Is`. Error-string wording and direct
+sentinel equality were not public contracts.
+
+The new inspection values, capacity error, and prepared-execution handle do not
+have stable JSON representations. `PreparedExecution.Details` returns the
+existing stable `PreparedRun` value.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.4.0
+go mod tidy
+```
+
+Run the consuming project's ordinary and race-enabled tests after upgrading.
+No source migration is required.
+
+## Prepare Once And Execute The Same Snapshot
+
+Consumers that need to persist preparation details before generation can now
+prepare an opaque, engine-bound execution:
+
+```go
+prepared, err := engine.PrepareExecution(ctx, request)
+if err != nil {
+	// Handle preparation failure.
+}
+defer prepared.Discard()
+
+details := prepared.Details()
+// Persist a consumer-selected, appropriately protected preparation record.
+
+result, err := engine.RunPrepared(ctx, prepared)
+```
+
+Preparation freezes the selected sources, rendered messages, effective
+settings, input content, structured-output metadata, and validation resources
+needed by execution. `Details` returns a fresh, caller-owned,
+credential-redacted `PreparedRun`.
+
+A handle belongs to its creating engine and permits one execution attempt.
+`RunPrepared` consumes that attempt on success and on operational failure.
+`Discard` is idempotent and releases an unclaimed handle's execution-only
+state. Consumers should discard handles they will not execute, particularly
+when a direct request API key may be retained privately until claim or
+discard.
+
+Prepared execution does not reserve backend admission during preparation.
+Credential availability and backend admission are checked when execution
+begins. The execution context is independent of the preparation context.
+
+See the
+[prepared-execution consumer guide](../consumers/pkg-promptkit.md#prepare-now-and-execute-the-same-snapshot-later),
+the [`PreparedExecution` GoDoc](../../prepared_execution.go), and the
+[`Engine.PrepareExecution` and `Engine.RunPrepared` GoDoc](../../engine.go)
+for the exact lifecycle, ownership, cancellation, capacity, timing, and
+failure contracts.
+
+## Inspect A Prompt
+
+`Engine.InspectPrompt` resolves one prompt ID and optional version through the
+engine's configured prompt source:
+
+```go
+inspection, err := engine.InspectPrompt(ctx, "report.summary", "")
+```
+
+The result includes prompt identity, the opaque prompt hash, declared default
+profile ID, declared input metadata, and normalized output contract. It
+structurally loads the selected definition and referenced message content but
+does not resolve a profile, load schemas or artifacts, render templates,
+reserve capacity, or contact a model.
+
+Use inspection for exact configuration checks and metadata discovery. Use
+`PrepareExecution` rather than relying on a prior inspection when later
+execution must freeze one exact source state, because filesystem-backed
+inspection is only a point-in-time lookup.
+
+See the
+[prompt-inspection consumer guide](../consumers/pkg-promptkit.md#inspect-a-prompt-before-preparation)
+and [`Engine.InspectPrompt` GoDoc](../../engine.go) for exact selection,
+ownership, and error behavior.
+
+## Inspect A Profile
+
+`Engine.InspectProfile` resolves one explicit profile independently of a
+prompt:
+
+```go
+inspection, err := engine.InspectProfile(ctx, "report-production")
+```
+
+The result includes the resolved effective execution target and whether a
+later request must provide a direct credential. Environment-variable names may
+be reported, but inspection does not read credential values or require the
+named variable to be populated.
+
+Inspection applies the engine's profile source precedence and resolves any
+selected backend. It does not load a prompt, render content, reserve capacity,
+or contact a model.
+
+See the
+[profile-inspection consumer guide](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work)
+and [`Engine.InspectProfile` GoDoc](../../engine.go) for the exact resolution,
+credential, ownership, and error contracts.
+
+## Identify Capacity-Rejected Backends
+
+Calls rejected at Promptkit's bounded engine admission boundary continue to
+match `ErrCapacityExceeded`. Consumers can additionally obtain the selected
+registered backend ID without parsing diagnostic text:
+
+```go
+result, err := engine.Run(ctx, request)
+if errors.Is(err, promptkit.ErrCapacityExceeded) {
+	var capacityErr *promptkit.CapacityError
+	if errors.As(err, &capacityErr) {
+		// Record capacityErr.BackendID using application-owned diagnostics.
+	}
+
+	// Apply application-owned overload or retry policy.
+}
+```
+
+The structured error applies to `Run` and `RunPrepared` admission rejection.
+It does not represent provider throttling, quota exhaustion, cancellation
+while waiting for generation capacity, or another model-client failure.
+Promptkit does not prescribe retry timing or transport status mapping.
+
+See the
+[error-handling consumer guide](../consumers/pkg-promptkit.md#handle-errors),
+the [`CapacityError` GoDoc](../../capacity_error.go), and the
+[`ErrCapacityExceeded` GoDoc](../../engine.go) for the canonical contracts.
+
+## Public API Additions
+
+The release adds:
+
+- `Engine.PrepareExecution`;
+- `Engine.RunPrepared`;
+- `PreparedExecution`, including `Details`, `Discard`, `String`, and
+  `GoString`;
+- `Engine.InspectPrompt`;
+- `PromptInspection`;
+- `PromptInputDefinition`;
+- `Engine.InspectProfile`;
+- `ProfileInspection`; and
+- `CapacityError`.
+
+No public API was removed.
+
+## Consumer Action
+
+None. Existing `v0.3.0` workflows may upgrade without adopting the new APIs.
+
+Consumers that adopt prepared execution should discard unused handles.
+Consumers that need backend-specific capacity diagnostics may add an
+`errors.As` check while retaining their existing `errors.Is` classification.

docs/releases/v0.5.0.md

@@ -0,0 +1,125 @@
+# Promptkit v0.5.0
+
+This supplemental changelog and migration guide summarizes the consumer-facing
+changes from `v0.4.0` to `v0.5.0`. The annotated `v0.5.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.5.0` makes provider requests less prescriptive and adds an application
+fallback layer for profile definitions:
+
+- unset optional provider controls are omitted from OpenAI-compatible request
+  bodies instead of being populated with framework values; and
+- `WithFallbackProfileFS` lets an application package profile defaults that
+  operators can override through the existing ordinary profile sources.
+
+These changes let compatible providers apply their own model defaults while
+giving applications stable embedded profile IDs without weakening operator
+configuration precedence.
+
+## Compatibility
+
+The release adds one public function and removes no public declaration.
+Existing source code should continue to compile.
+
+There is one intentional behavior change: when no profile or runtime override
+selects `top_p`, Promptkit no longer sends the former framework value of `1`.
+It omits `top_p` and lets the provider choose its behavior. Unset
+`temperature` and `max_tokens` are likewise omitted. Explicit nonzero profile
+values and runtime values—including explicit runtime zero values—retain their
+precedence and wire effect.
+
+Consumers that relied on Promptkit always sending `top_p: 1` should add that
+value to the relevant profile or runtime override before upgrading. Consumers
+that did not rely on the implicit sampling value require no migration.
+
+Application fallback profiles are opt-in. Engines that do not call
+`WithFallbackProfileFS` retain the previous profile-source behavior.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.5.0
+go mod tidy
+```
+
+Run the consuming project's ordinary and race-enabled tests after upgrading.
+If request payloads or model behavior are asserted in fixtures, review them for
+the optional-parameter omission described below.
+
+## Omitted Optional Provider Controls
+
+The built-in OpenAI-compatible client now includes `temperature`,
+`max_tokens`, and `top_p` only when a profile or runtime override selects the
+value. An explicit runtime zero remains present because runtime override
+pointers distinguish zero from an unspecified value.
+
+Promptkit's positive generation deadline remains a framework concern and is
+not a provider request-body default. Required request fields, session IDs,
+structured output, reasoning selection, credentials, and explicit extra
+parameters retain their existing behavior.
+
+See the [framework default and precedence reference](../formats.md#defaults-and-overrides),
+the [`ExecutionTargetOverride` GoDoc](../../types.go), and the
+[OpenAI-compatible request-body contract](../integrations/openai-compatible-chat.md#request-body)
+for current details.
+
+## Embedded Application Fallback Profiles
+
+Applications can package ordinary profile YAML in an `fs.FS` and register it
+as a fallback source:
+
+```go
+//go:embed profiles/*.yaml
+var applicationProfiles embed.FS
+
+engine, err := promptkit.NewEngine(promptkit.Config{
+	PromptDir:  "prompts",
+	ProfileDir: operatorProfileDir,
+},
+	promptkit.WithFallbackProfileFS(applicationProfiles, "profiles"),
+)
+```
+
+Leave `operatorProfileDir` empty when no operator source is configured. A
+configured ordinary source is authoritative: a matching definition overrides
+the application fallback, while a read or validation failure remains an error
+instead of silently reaching a lower layer.
+
+Profile definitions resolve in this order:
+
+1. in-memory profiles supplied with `WithProfiles`;
+2. the ordinary configured source selected by `WithProfileFile`,
+   `WithProfileFS`, or `Config.ProfileDir`;
+3. the application source supplied with `WithFallbackProfileFS`; and
+4. Promptkit's embedded built-in profiles.
+
+Only an absent profile ID falls through. Sources provide complete profiles and
+do not merge fields. Loading remains lazy, and the new source uses the existing
+strict profile YAML and credential rules.
+
+See the
+[embedded-default consumer guidance](../consumers/pkg-promptkit.md#supply-embedded-application-defaults),
+the [`WithFallbackProfileFS` GoDoc](../../engine.go), and the
+[profile source reference](../formats.md#source-and-profile-precedence) for
+current details.
+
+## Public API Changes
+
+The release adds:
+
+- `WithFallbackProfileFS`.
+
+No public declaration was removed or changed.
+
+## Consumer Action
+
+- Review any workflow that depended on Promptkit's implicit `top_p: 1` and
+  configure the value explicitly when required.
+- Optionally adopt `WithFallbackProfileFS` when an application should package
+  overridable profile defaults.
+- Run consumer tests after updating the module dependency.

docs/releases/v0.6.0.md

@@ -0,0 +1,131 @@
+# Promptkit v0.6.0
+
+This supplemental changelog and migration guide summarizes the consumer-facing
+changes from `v0.5.0` to `v0.6.0`. The annotated `v0.6.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.6.0` is a broad correctness, safety, efficiency, and maintainability
+release. It does not add or remove public declarations. The release:
+
+- centralizes shared execution-setting, output-contract, endpoint, and
+  JSON-compatible-value rules;
+- unifies prompt repository behavior and avoids unnecessary prompt and profile
+  decoding;
+- bounds consumer-controlled JSON trees and successful provider responses;
+- hardens prompt content paths, artifact files, provider URLs, JSON framing,
+  and error propagation;
+- reuses compiled schema plans and rendered artifact text within an operation;
+  and
+- improves cancellation behavior, prepared-value ownership, deterministic
+  transport testing, and maintainer validation.
+
+## Compatibility
+
+No public declaration was added, removed, or changed. Ordinary valid `v0.5.0`
+configurations and requests should continue to compile and behave as before.
+
+The release intentionally rejects or reports several inputs that were
+previously accepted, altered, or misclassified:
+
+- execution settings must be finite, within their documented ranges, and safe
+  to convert to Go durations;
+- output formats, validation modes, repair counts, and JSON Schema dependencies
+  are validated consistently;
+- file-backed prompt and profile identity comes from normalized YAML metadata,
+  not filenames;
+- prompt `content_file` values must be exact relative paths contained by their
+  configured source root;
+- built-in file artifacts must resolve to regular files;
+- selected provider endpoints must be absolute HTTP or HTTPS URLs without user
+  information, query strings, or fragments;
+- JSON documents and successful provider responses must contain exactly one
+  value, and successful provider bodies are limited to 16 MiB; and
+- excessively deep or expansive JSON-compatible values fail with ordinary
+  validation errors.
+
+These are compatibility corrections and safety boundaries rather than new
+consumer configuration requirements. Consumers relying on an invalid or
+ambiguous input should correct that input before upgrading.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.6.0
+go mod tidy
+```
+
+Run the consuming project's ordinary and race-enabled tests after upgrading.
+Applications with custom prompt/profile sources, local provider endpoints,
+unusual artifact paths, or assertions over provider error identities should
+pay particular attention to the compatibility notes below.
+
+## Source Loading And Identity
+
+Prompt definitions now share one source-neutral selection and normalization
+flow across operating-system and `fs.FS` sources. YAML `id` and `version`
+metadata are authoritative; filenames do not create a second identity system.
+Only selected content bodies are loaded, malformed unrelated definitions do
+not shadow valid exact matches, and per-file read failures are reported as
+prompt-load failures rather than false absence.
+
+File-backed profiles likewise use normalized YAML IDs, reuse their metadata
+read for selected strict decoding, and avoid fully decoding unrelated files.
+Selected malformed definitions remain authoritative and do not silently fall
+through to a lower-precedence source.
+
+Prompt `content_file` paths are opened exactly as declared after a separate
+blank check. They must remain relative to and contained by the configured
+prompt source root, including across operating-system symlinks.
+
+See the [framework source and identity reference](../formats.md) and
+[internal source overview](../internal/sources.md) for the current contracts.
+
+## Validation, Cancellation, And Efficiency
+
+JSON Schema documents preserve exact JSON-number representations. Schema
+resource URLs safely escape legal filesystem names, and each operation loads
+and compiles its schema graph once. `Run` and prepared execution reuse that
+operation-local plan; Promptkit does not introduce a cross-operation cache.
+
+Artifact reading, rendering, schema loading, compilation, and validation now
+check cancellation at the synchronous boundaries Promptkit controls. Rendering
+memoizes each artifact's text within one render operation, while plain JSON
+validation avoids materializing an unnecessary generic tree.
+
+The shared JSON-compatible-value owner now limits nesting and produced work so
+unsafe consumer-controlled structures return errors instead of risking
+unbounded recursion or allocation. See the
+[architecture policy](../policy/architecture.md) for invariant ownership and
+the [format reference](../formats.md) for validation behavior.
+
+## Provider Transport Hardening
+
+OpenAI-compatible endpoints are parsed and composed structurally, including
+nested base paths. Underlying transport cancellation and deadline errors remain
+discoverable with `errors.Is` through Promptkit's generation error category.
+
+Successful provider bodies are read with a fixed 16 MiB bound and must contain
+exactly one JSON response object followed only by whitespace. Oversized,
+truncated, malformed, or multiply framed responses fail without returning a
+partial result. See the
+[OpenAI-compatible integration contract](../integrations/openai-compatible-chat.md)
+for the canonical request, endpoint, error, and response behavior.
+
+## Public API Changes
+
+None.
+
+## Consumer Action
+
+- Correct any configuration or request that depends on the formerly permissive
+  cases described under Compatibility.
+- Confirm custom local provider endpoints are absolute HTTP or HTTPS base URLs
+  without credentials, queries, or fragments.
+- Confirm prompt content paths remain within their configured source root and
+  file artifacts resolve to regular files.
+- Run ordinary and race-enabled consumer tests after updating the dependency.

docs/releases/v0.7.0.md

@@ -0,0 +1,155 @@
+# Promptkit v0.7.0
+
+This supplemental changelog and migration guide summarizes the consumer-facing
+changes from `v0.6.0` to `v0.7.0`. The annotated `v0.7.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.7.0` expands provider integration and profile composition while making
+credential and generation-failure handling more flexible:
+
+- Promptkit now includes the `rakestrawhome` backend and its Gemma profile;
+- built-in generation failures expose bounded structured provider details;
+- an unavailable optional API-key environment source no longer prevents a
+  request from reaching an upstream that permits unauthenticated access; and
+- profiles can inherit from and selectively refine another profile.
+
+## Compatibility
+
+This release adds public declarations and fields but removes none. Existing
+keyed configuration literals and ordinary `errors.Is` handling continue to
+work.
+
+Adding `BaseProfileID` to `Profile` and `OpenAICompatibleProfileConfig` changes
+their struct shape. Consumers using positional composite literals for either
+type must convert them to keyed literals. Existing keyed literals require no
+change.
+
+The `rakestrawhome` backend ID is now built in and reserved. A consumer that
+previously registered that exact ID with `WithBackend` must remove its manual
+registration before upgrading. Other custom backend registrations are
+unchanged.
+
+When an optional backend, profile, or request `APIKeyEnv` is unset, empty, or
+whitespace-only, the built-in client now omits `Authorization` and sends the
+request. Previously this condition could fail before transport. Set
+`Profile.APIKeyRequired` when missing credentials must remain a local
+preflight error.
+
+Provider non-success responses continue to match `ErrLLMGenerate`. Their
+rendered wording is not a compatibility contract; consumers can now use
+`errors.As` with `*GenerationError` when structured status information is
+needed.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.7.0
+go mod tidy
+```
+
+Remove any manual `rakestrawhome` backend registration, convert positional
+profile literals to keyed literals, and run the consuming project's ordinary
+and race-enabled tests.
+
+## Rakestrawhome Built-In Backend And Profile
+
+Every engine now includes the reserved `rakestrawhome` backend, identified by
+`BackendRakestrawHome`. The built-in `rakestrawhome-gemma-4-31b` profile
+selects that backend. Consumers can use the maintained endpoint, credential,
+capacity, and model defaults without registering either definition themselves.
+
+See the [built-in backend and profile catalogs](../formats.md#built-in-backends)
+and the [consumer adoption example](../consumers/pkg-promptkit.md#use-the-rakestrawhome-built-in-profile)
+for the current contracts.
+
+## Structured Generation Errors
+
+Non-2xx responses from the built-in OpenAI-compatible client now return an
+immutable `*GenerationError`. Consumers can inspect the HTTP status and any
+safely extracted provider code, type, or message while retaining the ordinary
+generation-error category:
+
+```go
+var generationErr *promptkit.GenerationError
+if errors.As(err, &generationErr) {
+	status := generationErr.StatusCode()
+	_ = status
+}
+```
+
+Provider fields are bounded and normalized but remain untrusted and may
+contain sensitive request or schema details. Default and Go-syntax formatting
+omit those fields. Applications must apply their own disclosure policy before
+logging or presenting accessor values.
+
+See the [`GenerationError` GoDoc](../../generation_error.go), the
+[consumer error-handling guide](../consumers/pkg-promptkit.md#handle-errors),
+and the [OpenAI-compatible response contract](../integrations/openai-compatible-chat.md#response-handling).
+
+## Optional Credential Sources
+
+`APIKeyEnv` names an optional environment lookup source unless the selected
+profile explicitly sets `APIKeyRequired`. When neither a direct request key nor
+a usable environment value exists, the built-in client omits the bearer header
+and handles the upstream response normally. This supports local and other
+OpenAI-compatible providers that permit unauthenticated requests without
+hiding an authentication error returned by a provider that requires one.
+
+The [credential format reference](../formats.md#credentials), the
+[`Backend` GoDoc](../../backends.go), the
+[`ExecutionTargetOverride` GoDoc](../../types.go), and the
+[authentication integration contract](../integrations/openai-compatible-chat.md#authentication)
+define the current precedence and availability rules.
+
+## Profile Inheritance
+
+YAML profiles can name one parent with `base_profile`; in-memory profiles use
+`Profile.BaseProfileID`, and `OpenAICompatibleProfileConfig` forwards the same
+field. A profile can act as an application-owned alias of a built-in or refine
+selected inherited settings:
+
+```go
+promptkit.WithProfiles(promptkit.Profile{
+	ID:              "weather-light",
+	BaseProfileID:   "deepseek-4-flash",
+	ReasoningEffort: "high",
+})
+```
+
+Base lookup observes the existing source precedence. Chains are linear,
+cycle-safe, and resolved afresh for ordinary operations. Prepared execution
+freezes the fully resolved target. The selected leaf ID remains public while
+effective execution settings reflect the resolved chain.
+
+See the [profile inheritance format reference](../formats.md#profile-inheritance),
+the [consumer alias example](../consumers/pkg-promptkit.md#alias-a-built-in-profile),
+and the [`Profile` GoDoc](../../types.go) for exact merge and validation
+behavior.
+
+## Public API Changes
+
+The release adds:
+
+- `BackendRakestrawHome`;
+- `GenerationError`, including `StatusCode`, `ProviderCode`, `ProviderType`,
+  `ProviderMessage`, `Error`, `GoString`, and `Unwrap`;
+- `Profile.BaseProfileID`; and
+- `OpenAICompatibleProfileConfig.BaseProfileID`.
+
+No public declaration was removed.
+
+## Consumer Action
+
+- Remove a manual backend registration whose ID is exactly `rakestrawhome`.
+- Convert positional `Profile` or `OpenAICompatibleProfileConfig` literals to
+  keyed literals.
+- Set `Profile.APIKeyRequired` where a missing credential must fail locally
+  instead of reaching the provider unauthenticated.
+- Treat `GenerationError` provider fields as untrusted and potentially
+  sensitive when adopting the new accessors.
+- Run consumer ordinary and race-enabled tests after updating the module.

docs/releases/v0.8.0.md

@@ -0,0 +1,109 @@
+# Promptkit v0.8.0
+
+This supplemental changelog and migration guide summarizes the consumer-facing
+changes from `v0.7.0` to `v0.8.0`. The annotated `v0.8.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.8.0` activates Promptkit's bounded output-repair workflow:
+
+- failed nonempty-text, JSON, and JSON Schema validation can make a limited
+  number of corrective model calls;
+- corrective calls preserve the original rendered conversation, effective
+  target, session, structured-output contract, and backend capacity policy;
+- results report cumulative usage and the number of corrective calls actually
+  made; and
+- explicitly empty OpenAI-compatible response content now reaches output
+  validation instead of being classified as a malformed provider envelope.
+
+## Compatibility
+
+This release adds no public declarations or fields and removes none. Existing
+source code remains source-compatible.
+
+The behavior of the existing `OutputContract.RepairAttempts` field and prompt
+YAML `repair_attempts` field has changed. A positive value now authorizes real
+additional model calls after eligible validation failures; earlier releases
+accepted the field but the public engine remained single-pass. Consumers that
+set a positive value should expect additional latency, token usage, and
+provider cost when repair is needed.
+
+Repair budgets must now be between zero and three. A positive budget requires
+`basic`, `json`, or `json_schema` validation. Values above three and a positive
+budget paired with `none` are invalid contracts rather than ignored settings.
+
+An explicitly present empty or whitespace-only string returned by the built-in
+OpenAI-compatible client is now a completed generation candidate. `none`
+validation permits it, while `basic`, `json`, and `json_schema` classify it
+under their ordinary validation rules and may repair it when configured.
+Missing, `null`, or non-string content remains a malformed provider response.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.8.0
+go mod tidy
+```
+
+Review every prompt definition and request override that sets a positive repair
+budget. Use zero or omit the field to retain single-pass execution. Ensure each
+positive budget is no greater than three and uses an eligible validation mode,
+then run the consuming project's ordinary and race-enabled tests.
+
+## Bounded Output Repair
+
+`repair_attempts` counts corrective calls in addition to the initial model
+call. Promptkit validates each completed candidate, stops at the first valid
+one, and never exceeds the configured bound. If every candidate remains
+invalid, the run completes successfully with the final candidate and its
+failed validation result rather than returning an operational error.
+
+Each correction starts from the original rendered messages and includes only
+the latest invalid candidate and latest validation diagnostics. JSON Schema
+mode retains the provider-native structured-output request as its first line of
+defense. Promptkit performs only deterministic structural validation; a valid
+response is not necessarily factual or correct for an application's domain.
+
+Usage in the final result is cumulative across the initial response and every
+completed corrective response. `ValidationResult.RepairAttempts` reports the
+number of corrective calls actually made. Corrective generation failures use
+the same public generation-error categories and structured provider details as
+an initial generation failure.
+
+See the [output-contract format reference](../formats.md#output-contract), the
+[consumer repair example](../consumers/pkg-promptkit.md#repair-a-structured-result),
+and the [`OutputContract` and `ValidationResult` GoDoc](../../types.go) for the
+current contracts.
+
+## Explicit Empty Content
+
+The built-in OpenAI-compatible client now distinguishes an explicitly present
+empty string from a missing or malformed `content` field. This aligns built-in
+and injected clients by letting the selected output contract decide whether an
+empty candidate is acceptable, invalid, or eligible for repair.
+
+See the
+[OpenAI-compatible response contract](../integrations/openai-compatible-chat.md#response-handling)
+for the exact envelope behavior.
+
+## Public API Changes
+
+None. This release activates and tightens the documented behavior of existing
+fields.
+
+## Consumer Action
+
+- Remove or set `repair_attempts` to zero where execution must remain
+  single-pass.
+- Keep every positive repair budget at three or fewer and pair it with
+  `basic`, `json`, or `json_schema` validation.
+- Account for additional latency, usage, and provider cost when enabling
+  repair.
+- Continue checking the returned validation status because bounded repair can
+  exhaust without producing a valid candidate.
+- Review workflows that previously treated explicit empty provider content as
+  a generation error.

docs/releases/v0.9.0.md

@@ -0,0 +1,125 @@
+# Promptkit v0.9.0
+
+This supplemental changelog and migration guide summarizes the consumer-facing
+changes from `v0.8.0` to `v0.9.0`. The annotated `v0.9.0` tag is the
+authoritative release record. Exact current contracts belong to the linked
+GoDoc and durable documentation.
+
+## Summary
+
+`v0.9.0` adds stateless request-message composition and separates maintained
+provider data from Promptkit's core implementation:
+
+- callers can append already-rendered messages to a configured prompt for
+  application-owned conversations and semantic correction workflows;
+- the public package now publishes constants for the four supported text-chat
+  roles, and prompt definitions use the same normalized role vocabulary; and
+- the OpenRouter and Rakestrawhome backend/profile catalogs now come from two
+  independently versioned Go module dependencies.
+
+## Compatibility
+
+This release adds one field and four constants to the public API and removes no
+public declarations. Existing keyed `RunRequest` literals that omit
+`AppendedMessages` retain their behavior. Adding the field changes the struct
+shape, so consumers using positional `RunRequest` literals must convert them
+to keyed literals.
+
+Message roles in prompt definitions are now trimmed, lowercased, and required
+to be `developer`, `system`, `user`, or `assistant`. Definitions using another
+role that earlier releases accepted as an arbitrary nonblank string now fail
+prompt loading. In particular, the text-only message contract does not support
+`tool` or the deprecated `function` role. Otherwise valid roles with different
+case or surrounding whitespace are normalized rather than rejected.
+
+The catalog extraction preserves Promptkit's public API, built-in backend and
+profile IDs, configuration, precedence, credential handling, and capacity
+behavior. Consumers do not import or register either catalog themselves, and
+no configuration migration is required. Promptkit now selects two
+independently versioned data dependencies and guarantees only the catalog
+versions selected and tested by this Promptkit release.
+
+## Upgrade
+
+Update the module dependency with:
+
+```sh
+go get gitea.maximumdirect.net/eric/promptkit@v0.9.0
+go mod tidy
+```
+
+Convert any positional `RunRequest` literals to keyed literals. Review prompt
+definitions for unsupported roles, then run the consuming project's ordinary
+and race-enabled tests.
+
+## Appended Request Messages
+
+`RunRequest.AppendedMessages` accepts caller-owned `RenderedMessage` values
+that Promptkit validates, copies, and appends after every rendered
+prompt-definition message in caller order. Promptkit does not template this
+content, retain conversation state between calls, impose a retry policy, or
+apply a message-count, byte-size, token, or context-window limit. Upstream
+rejections continue through the ordinary generation-error boundary.
+
+This primitive supports application-owned conversation continuations and
+domain-aware correction loops while preserving Promptkit's existing
+preparation, hashing, prepared-execution, structural repair, backend-capacity,
+credential, and cancellation behavior. Appended content can include sensitive
+model output or application feedback; prepared values expose the complete
+effective messages by design, while default request formatting reports only
+the appended-message count.
+
+See the
+[consumer appended-message example](../consumers/pkg-promptkit.md#append-already-rendered-messages),
+the [`RunRequest`, `RenderedMessage`, and `CacheControl` GoDoc](../../types.go),
+and the [OpenAI-compatible request contract](../integrations/openai-compatible-chat.md#request-body)
+for current behavior.
+
+## Supported Message Roles
+
+The new `RoleDeveloper`, `RoleSystem`, `RoleUser`, and `RoleAssistant`
+constants identify the complete role vocabulary accepted by Promptkit's
+text-chat message model. The same validation and normalization now apply to
+prompt-definition messages and request-supplied appended messages. Promptkit
+does not translate between roles; provider- or model-specific rejection of an
+otherwise supported role remains an upstream generation error.
+
+See the [message format reference](../formats.md#messages-and-templates) for
+the canonical prompt-definition contract.
+
+## Independently Versioned Backend Catalogs
+
+Promptkit imports immutable catalog data for its maintained OpenRouter and
+Rakestrawhome backends and profiles. Engine construction validates and
+assembles both catalogs behind the existing built-in registry and profile
+precedence rules. Promptkit no longer keeps duplicate embedded profile assets
+or hard-coded definitions for those maintained backends.
+
+The module versions in Promptkit's `go.mod` identify the catalog releases
+tested with this release. The [built-in backend and profile format
+reference](../formats.md#built-in-backends) remains the canonical consumer
+contract, while the [internal source documentation](../internal/sources.md#profiles-and-built-ins)
+describes the dependency boundary.
+
+## Public API Changes
+
+The release adds:
+
+- `RunRequest.AppendedMessages`;
+- `RoleDeveloper`;
+- `RoleSystem`;
+- `RoleUser`; and
+- `RoleAssistant`.
+
+No public declaration was removed.
+
+## Consumer Action
+
+- Convert positional `RunRequest` literals to keyed literals.
+- Replace unsupported prompt-definition roles with an appropriate supported
+  role, or keep richer tool-call protocols in an application-owned client.
+- Treat appended messages and prepared effective messages according to the
+  application's sensitive-data policy.
+- Do not add direct catalog imports or registration calls; existing Promptkit
+  construction and configuration remain correct.
+- Run consumer ordinary and race-enabled tests after updating the module.

docs/roadmap/concurrency.md

@@ -1,236 +0,0 @@
-# Backend-Specific Concurrency Management
-
-**Status:** Complete.
-
-## Purpose
-
-This roadmap defines the scope and target end state for engine-local,
-backend-specific concurrency management. It records the intended capability,
-consumer value, and important policy choices.
-
-This document is planning material, not a description of current behavior.
-Current exported contracts remain owned by Go declarations and GoDoc, backend
-registration guidance by the
-[consumer guide](../consumers/pkg-promptkit.md#register-a-custom-backend), and
-implemented orchestration by the
-[internal runner document](../internal/runner.md).
-
-## Motivation
-
-Different model backends can sustain very different request loads. A local
-network endpoint may need a small concurrency limit, while OpenRouter can
-usually accept substantially more simultaneous work. Requiring every consumer
-to build its own semaphores and queues would duplicate routing knowledge,
-create inconsistent cancellation behavior, and make it easy for one caller to
-bypass the intended backend limit.
-
-Promptkit should own this coordination because it already resolves each run to
-an engine-scoped backend identity and owns every model-generation call made by
-the runner. Consumers should continue submitting ready-to-run requests through
-the synchronous API, including concurrently from multiple goroutines, without
-implementing their own backend scheduler.
-
-The buffered queue is a safety boundary, not an ordinary throughput
-restriction. Its primary purpose is to prevent a bug or unintended submission
-loop from creating an unbounded in-memory backlog.
-
-## Scope
-
-The feature will add optional concurrency policy to registered backends and
-coordinate `Run` calls against independent per-backend capacity pools.
-
-Each policy has two distinct controls:
-
-- an active-generation limit, which protects the backend from too many
-  simultaneous model requests; and
-- a bounded waiting capacity, which protects the process from admitting an
-  unbounded backlog.
-
-Concurrency policy belongs to a backend registration. It is not a profile
-model parameter and cannot be overridden per run. Profiles select the policy
-through their backend ID, while a profile or request endpoint override remains
-in the selected backend's pool.
-
-`Prepare` does not call a model and will remain outside concurrency admission.
-
-## Defaults And Configuration
-
-The built-in OpenRouter backend will use:
-
-- an active-generation limit of 16; and
-- a waiting capacity of 1024.
-
-The waiting default is intentionally generous. Reaching it should indicate
-abnormal submission pressure rather than normal application behavior.
-
-Consumer-registered backends will remain unlimited unless the consumer
-configures an active-generation limit. When a consumer enables a limit and
-does not specify waiting capacity, the waiting capacity will default to 1024.
-Consumers may configure a different bounded capacity, including zero when
-they want no admitted backlog beyond the active-limit-sized run set.
-
-The public representation must distinguish an omitted waiting capacity from
-an explicit zero.
-
-Endpoint-only profiles have no backend registration from which to obtain
-policy and will remain unlimited. A future engine-wide or endpoint-keyed
-policy can be considered separately if consumers demonstrate that need.
-
-Invalid limits or capacities will fail engine construction as invalid
-configuration. Policy values will be copied into engine-owned immutable state
-along with the rest of the backend registration.
-
-## Admission And Execution Behavior
-
-`Run` remains a synchronous, wait-for-result operation. Concurrent callers may
-block inside `Run` while waiting for their selected backend, then receive the
-ordinary result or error from that invocation.
-
-For a configured pool, the active-generation limit plus the waiting capacity
-defines the maximum number of concurrent `Run` invocations that Promptkit will
-accept for that backend. A waiting capacity of zero therefore accepts no more
-runs than the active limit. Admission is immediate: a call either reserves one
-of those bounded slots or receives the capacity error. An accepted run may
-then wait internally for active-generation capacity.
-
-For a limited backend, Promptkit will bound the number of accepted runs before
-expensive artifact loading, prompt rendering, and large defensive copies where
-practical. Lightweight prompt, profile, and backend resolution may occur first
-when it is required to identify the correct capacity pool. This pre-admission
-resolution must not become a second execution-precedence path with behavior
-that can drift from `Prepare`.
-
-An accepted run retains its admission until it completes or fails. Every
-actual model-generation call for that run must separately observe the
-backend's active-generation limit. This includes:
-
-- the initial generation;
-- every output-repair generation; and
-- calls made through either the built-in or an injected model client.
-
-Preparation and output validation should not hold an active-generation permit.
-A repair remains part of its already-admitted run, but reacquires active
-generation capacity so repairs cannot exceed the backend limit. It must not be
-rejected merely because new runs filled the waiting queue after its initial
-generation.
-
-Within one backend pool, waiting generation calls should be served in FIFO
-order, subject to canceled calls being removed. Different backend pools make
-progress independently; a saturated local backend must not consume
-OpenRouter's active or waiting capacity.
-
-The feature will not promise ordering across backend pools or completion order
-among admitted runs.
-
-## Capacity Failure And Cancellation
-
-When a backend's bounded waiting capacity is full, a new `Run` call will fail
-promptly rather than waiting outside the bounded admission system. The public
-API will expose a recognizable capacity-exhaustion error identity distinct
-from invalid configuration, invalid requests, and model-client failures.
-Rejected calls return no partial result and do not invoke the model client.
-
-Waiting within the admitted backlog or for active-generation capacity must
-honor the caller's context. Cancellation or deadline expiry will:
-
-- stop waiting promptly;
-- release any admission or generation capacity held by that invocation;
-- preserve the applicable context error identity; and
-- avoid invoking the model client if cancellation wins before generation
-  starts.
-
-Capacity must also be released after preparation, generation, validation,
-repair, or collaborator failure. One failed or canceled run must not reduce
-the backend's future usable capacity.
-
-Elapsed `Run` timing will include time spent waiting after the call is
-accepted. `PreparedRun` timing will continue to describe preparation rather
-than queue waiting.
-
-## Engine And Client Boundaries
-
-All pools and queued state belong to one `Engine`. Separate engines do not
-share capacity, even when they register the same backend ID or endpoint. The
-feature introduces no process-global scheduler.
-
-The engine will apply policy consistently to the built-in model client and an
-injected `LLMClient`. Consumers calling their own client outside Promptkit are
-outside this boundary. Injected clients remain responsible for their internal
-thread safety and cancellation behavior.
-
-Backend policy is keyed by the resolved backend ID rather than endpoint text.
-This preserves stable routing when a selected backend's endpoint is overridden
-and avoids accidentally combining unrelated registrations that happen to use
-the same URL.
-
-## Queue Lifetime And Observability
-
-Admission state is buffered, ephemeral, and in-process. It is not persisted
-and has no survival guarantee across engine disposal or process termination.
-Promptkit will not introduce background job ownership or require consumers to
-start or stop workers.
-
-The initial feature does not require public queue-depth metrics, callbacks, or
-inspection APIs. Capacity errors and ordinary call timing provide the
-consumer-visible behavior. Operational observability can be added later
-without coupling the scheduling mechanism to an application logging or
-metrics system.
-
-## Compatibility
-
-Consumer-registered backends and endpoint-only profiles remain unlimited
-unless concurrency is explicitly configured, preserving their existing
-behavior.
-
-The built-in OpenRouter backend will change from unlimited concurrency to a
-limit of 16 with a bounded waiting capacity of 1024. Ordinary synchronous
-calls remain unchanged, while unusually high concurrent use may now wait or
-return the capacity error. This behavioral change must be identified in the
-release notes for the version that publishes it.
-
-Adding backend policy fields and a public capacity error is otherwise
-additive. The change will use a pre-`v1` minor release under Promptkit's
-[release policy](../release.md#release-model).
-
-## Non-Goals
-
-This scope does not include:
-
-- asynchronous job handles, polling, or detached result delivery;
-- durable or cross-process queues;
-- persistence or recovery across engine or process shutdown;
-- priorities, scheduling weights, or consumer-defined fairness classes;
-- automatic retries, backoff, rate-limit interpretation, or provider quota
-  discovery;
-- token-per-minute or request-per-minute rate limiting;
-- dynamic reconfiguration after engine construction;
-- per-profile or per-run concurrency overrides;
-- endpoint-keyed pooling for profiles without a backend ID;
-- process-global coordination across engines;
-- application worker lifecycle, logging, tracing, or metrics policy; or
-- changes to prompt, profile, schema, or model-provider wire formats.
-
-## Target End State
-
-This roadmap reaches its target end state when:
-
-- each engine independently coordinates configured backend capacity;
-- the built-in OpenRouter backend allows 16 active generations and up to 1024
-  waiting runs;
-- consumer backends can opt into their own active and waiting limits while
-  remaining unlimited by default;
-- endpoint overrides retain the selected backend's capacity pool and
-  endpoint-only profiles remain unlimited;
-- synchronous `Run` callers wait for and receive their ordinary result;
-- admission is bounded before expensive preparation work where practical;
-- every initial and repair generation observes the backend's active limit
-  without serializing preparation or validation;
-- a full waiting queue returns a recognizable capacity error without invoking
-  the model client;
-- cancellation and all failure paths promptly release capacity and preserve
-  context error identity;
-- built-in and injected model clients receive the same scheduling behavior;
-- pools remain ephemeral, engine-scoped, and independent across backend IDs;
-  and
-- current-state GoDoc, consumer, internal, and release documentation describe
-  the implemented behavior once it lands.

docs/roadmap/deferred.md

@@ -0,0 +1,62 @@
+# Deferred Feature Ideas
+
+## Purpose
+
+This document catalogs feature ideas that remain potentially useful but have
+been deliberately postponed. These ideas are not awaiting ordinary selection
+from the [future feature catalog](future.md); each has a stated reason to wait
+and should be reconsidered only when its trigger becomes relevant.
+
+Deferred entries are not commitments, schedules, active implementation plans,
+or descriptions of current behavior. When an entry is reactivated, move it to
+`future.md` for evaluation or directly into a focused roadmap after its open
+design dependencies have been resolved.
+
+## Deferred Ideas
+
+### Semantic Execution-Target Fingerprints
+
+**Reason for deferral:** A stable digest requires a deliberate semantic-
+equality and versioning design. Notarius can safely use conservative source
+hashes and a Promptkit release marker today, while Weatherreporter does not
+currently reuse LLM-dependent checkpoints.
+
+Promptkit could expose an opaque equality value for a resolved profile and its
+effective generation target. This would let checkpointing consumers detect
+generation-affecting configuration changes without hashing YAML presentation
+or depending on Promptkit's built-in catalog layout.
+
+The digest should change with semantically relevant state such as the resolved
+model, endpoint, backend routing identity, request defaults, extra parameters,
+profile generation settings, and selected built-in profile semantics. It
+should exclude credential values, concurrency and queue policy, source paths,
+comments, formatting, and other representation-only changes. Whether a
+credential environment-variable name affects equality must be decided
+explicitly. The encoding should remain opaque and internally versioned so
+Promptkit can deliberately invalidate earlier digests when its resolution
+semantics change.
+
+Reconsider this idea when a downstream consumer needs Promptkit-owned
+checkpoint equality or when a broader semantic identity design is selected.
+
+### Eager Source Validation
+
+**Reason for deferral:** Exact prompt and profile inspection may already
+provide a sufficiently small validation surface alongside construction-time
+definition validation. Experience from downstream adoption should establish
+whether a separate operation would add enough value to justify its broader
+contract.
+
+Promptkit could provide an explicit offline operation that also checks schema
+resources without model generation. `NewEngine` already validates effective
+definitions and captures prompt content; schema resources remain
+request-dependent.
+
+Such an operation would need coherent handling for schema syntax and transitive
+references, context cancellation, and source-specific public errors. Credential
+declarations must remain separate from credential values; checking current
+environment availability, if supported at all, should be an explicit option
+and must not expose secrets.
+
+Reconsider this idea after downstream use of `InspectPrompt`,
+`InspectProfile`, and fixture-based preparation demonstrates a concrete gap.

docs/roadmap/future.md

@@ -12,6 +12,9 @@ consumer value, and important scope boundaries. Defer API design,
 implementation details, sequencing, and acceptance criteria until an idea is
 selected.
 
+Ideas that have been deliberately postponed rather than left available for
+ordinary selection belong in the [deferred catalog](deferred.md).
+
 ## Using This Catalog
 
 - Add an idea when its purpose and likely value can be stated clearly.
@@ -23,6 +26,8 @@ selected.
 - When an idea is selected, move its active planning to a focused roadmap or,
   when it requires a durable architectural decision, an ADR. Update
   current-state documentation only when implementation lands.
+- Move an idea to `deferred.md` when maintainers decide to retain it but wait
+  for a stated design dependency, demand signal, or reconsideration trigger.
 - Remove ideas that are no longer relevant. Retain a rejected idea only when
   its rationale is likely to prevent repeated reconsideration.
 
@@ -33,9 +38,8 @@ consumers.
 
 ## Ideas
 
-No ideas are currently cataloged. Backend-specific concurrency management has
-been selected for active planning in the
-[focused concurrency roadmap](concurrency.md).
+No ideas are currently awaiting selection. Active feature work belongs in its
+focused roadmap rather than this catalog.
 
 ## Entry Format
 

docs/roadmap/implementation.md

@@ -1,841 +0,0 @@
-# Backend-Specific Concurrency Management Implementation Plan
-
-**Status:** Complete.
-
-## Purpose
-
-This document is the decision-complete implementation plan for
-[backend-specific concurrency management](concurrency.md). It is written for a
-coding agent that will implement each stage in order.
-
-The feature roadmap owns the intended capability, consumer value, policy
-choices, compatibility decision, and target end state. This document owns the
-concrete API, internal representation, scheduling architecture, implementation
-sequence, test ownership, documentation updates, and completion gates.
-
-## Implementation Rules
-
-- Complete the stages in order. Keep the repository compiling and the focused
-  tests passing at every stage boundary.
-- Preserve unrelated working-tree changes. In particular, `concurrency.md` and
-  the removal of its source idea from `future.md` may already be uncommitted
-  when implementation begins; retain both.
-- Follow every policy under `docs/policy/`, the task-specific reading guide in
-  `docs/development.md`, and the target behavior in `concurrency.md`.
-- Keep the public API in the root `promptkit` package and implementation
-  details under `internal/`. Do not expose scheduler types or create another
-  public package.
-- Use only the Go standard library for scheduling. Do not add a queue,
-  semaphore, worker-pool, or metrics dependency.
-- Preserve synchronous, wait-for-result `Run`, unrestricted `Prepare`,
-  engine-local state, endpoint-only profiles, backend-selected profiles,
-  backend identity through endpoint overrides, and injected `LLMClient`
-  behavior.
-- Do not broaden the work into asynchronous jobs, durable queues, retries,
-  rate limiting, dynamic configuration, priorities, worker lifecycle,
-  endpoint-keyed pools, or public queue observability.
-- Keep all tests deterministic, bounded, offline, and race-safe. Coordinate
-  concurrent tests with channels and barriers rather than timing assumptions
-  or live providers.
-- Update exact GoDoc with each exported declaration change. Update durable
-  current-state documents only after the corresponding behavior is
-  implemented.
-- Test configurable mechanisms with small test-owned limits. Assert the exact
-  OpenRouter `16` and default queue `1024` values only at the registry contract
-  that owns those operational defaults.
-- Do not create a release, change a module version, or tag a commit. The final
-  implementation handoff must identify the built-in OpenRouter behavior change
-  for the next pre-`v1` minor release.
-
-## Fixed Design
-
-### Public Backend Configuration
-
-Append these fields to the existing root `Backend` type in `backends.go`:
-
-```go
-type Backend struct {
-	// Existing fields remain unchanged and in their current order.
-
-	ConcurrencyLimit int
-	QueueCapacity    *int
-}
-```
-
-Use these exact semantics:
-
-| Public values | Meaning |
-| --- | --- |
-| `ConcurrencyLimit == 0`, `QueueCapacity == nil` | Unlimited backend; preserve current behavior. |
-| `ConcurrencyLimit > 0`, `QueueCapacity == nil` | Limit active generations and use the default waiting capacity of 1024. |
-| `ConcurrencyLimit > 0`, `QueueCapacity != nil` | Limit active generations and use the pointed-to capacity exactly, including zero. |
-| `ConcurrencyLimit < 0` | Invalid engine configuration. |
-| `QueueCapacity != nil` and `*QueueCapacity < 0` | Invalid engine configuration. |
-| `ConcurrencyLimit == 0` and `QueueCapacity != nil` | Invalid engine configuration because a queue without an active limit has no defined consumer value. |
-
-`ConcurrencyLimit` counts simultaneous calls to the engine-owned internal
-model-client boundary for this backend. `QueueCapacity` controls additional
-accepted `Run` invocations beyond that limit. The maximum admitted runs for a
-limited backend is therefore:
-
-```text
-ConcurrencyLimit + effective QueueCapacity
-```
-
-Guard that addition against integer overflow during backend validation.
-Do not impose an arbitrary upper bound beyond non-negativity and overflow
-safety.
-
-The `QueueCapacity` pointer exists only to distinguish omission from explicit
-zero. `WithBackend` and `NewEngine` must not retain the caller's pointer.
-`Backend` continues to have no stable JSON representation, and consumers
-remain directed to keyed literals.
-
-Do not add concurrency fields to `Profile`, `ExecutionTarget`,
-`ExecutionTargetOverride`, `RunRequest`, prompt or profile files, or stable
-prepared/result JSON.
-
-### Built-In And Custom Defaults
-
-The backend registry owns these exact operational defaults:
-
-```go
-const (
-	openRouterConcurrencyLimit = 16
-	defaultQueueCapacity       = 1024
-)
-```
-
-The built-in `openrouter` definition has a normalized concurrency limit of 16
-and queue capacity of 1024.
-
-Consumer registrations remain unlimited when concurrency is omitted. For a
-consumer backend with a positive limit and omitted queue capacity, normalize
-the queue capacity to 1024. Preserve an explicitly configured zero.
-
-Consumers still cannot replace the reserved `openrouter` registration.
-Endpoint-only profiles have no backend policy and remain unlimited. A selected
-backend retains its pool when a profile or request overrides only its endpoint.
-
-### Internal Backend Representation
-
-Extend `internal/domain.Backend` with scalar policy values and explicit
-presence rather than retaining a pointer:
-
-```go
-type Backend struct {
-	// Existing fields...
-	ConcurrencyLimit int
-	QueueCapacity    int
-	QueueCapacitySet bool
-}
-
-type BackendCapacityPolicy struct {
-	ConcurrencyLimit int
-	QueueCapacity    int
-}
-```
-
-`WithBackend` converts the public pointer into `QueueCapacity` plus
-`QueueCapacitySet`. Registry normalization validates the combinations above,
-fills the default, and leaves every limited stored backend with
-`QueueCapacitySet == true`. Unlimited stored backends retain zero values and
-`QueueCapacitySet == false`.
-
-Add this internal registry method:
-
-```go
-func (r *Registry) CapacityPolicies() map[string]domain.BackendCapacityPolicy
-```
-
-It returns a newly allocated map containing only limited backends. Values are
-scalars, so callers cannot mutate registry state. The built-in OpenRouter
-policy is included. `GetBackend` continues returning a defensive backend copy,
-now including normalized scalar capacity metadata.
-
-Capacity policy is operational registry metadata. Do not merge it into an
-execution target or expose it to injected model clients.
-
-### Public Capacity Error
-
-Add this root sentinel beside the other run errors in `engine.go`:
-
-```go
-var ErrCapacityExceeded = errors.New("backend capacity exceeded")
-```
-
-Its GoDoc must state that it identifies a `Run` rejected because the selected
-backend has already admitted `ConcurrencyLimit + QueueCapacity` runs. It is
-not an invalid request, an LLM/provider rate-limit response, or an
-`ErrLLMGenerate` failure.
-
-The internal capacity component owns a corresponding internal
-`ErrCapacityExceeded`. Add its mapping in `publicErrorFor` before the broader
-generation and invalid-request cases. The public error must preserve the
-internal error through wrapping while matching `ErrCapacityExceeded` with
-`errors.Is`.
-
-A capacity rejection returns no partial result and must not invoke the
-artifact reader, renderer, schema loader, validator, or model client. Prompt,
-profile, and backend loading needed to select the pool may already have
-occurred.
-
-### Internal Capacity Component
-
-Add `internal/capacity` as the single owner of engine-local run admission and
-active-generation permits.
-
-Use these package-level boundaries:
-
-```go
-var ErrCapacityExceeded error
-
-type Manager struct {
-	// Private immutable pool map.
-}
-
-func NewManager(
-	policies map[string]domain.BackendCapacityPolicy,
-) (*Manager, error)
-
-func (m *Manager) Admit(
-	ctx context.Context,
-	backendID string,
-) (release func(), err error)
-
-func NewClient(m *Manager, next llm.Client) llm.Client
-```
-
-`NewManager` copies the supplied map and creates one independent pool per
-limited backend. Defensively reject blank IDs, non-positive concurrency
-limits, negative queue capacities, or total-capacity overflow even though the
-registry normally supplies normalized values. Construction creates no worker
-goroutines.
-
-An absent manager, blank backend ID, or ID absent from the policy map is
-unlimited:
-
-- `Admit` succeeds with a non-nil no-op release function; and
-- the client wrapper calls the next client directly.
-
-For a limited pool, `Admit` is immediate and context-aware:
-
-1. return `ctx.Err()` if the context is already done;
-2. under the pool lock, compare admitted runs with
-   `ConcurrencyLimit + QueueCapacity`;
-3. return an error matching internal `ErrCapacityExceeded` when full; or
-4. increment admitted runs and return an idempotent release function.
-
-The release function decrements admission exactly once, even if accidentally
-called more than once. It does not release an active-generation permit; those
-permits have their own lifetime.
-
-### FIFO Generation Permits
-
-`NewClient` returns an internal `llm.Client` wrapper around either the built-in
-client or the public-client adapter. It must preserve requests, successful
-responses, nil responses, and collaborator error identities exactly.
-`next` must be non-nil; `NewEngine` and internal runner construction maintain
-that invariant. A nil manager returns `next` unchanged.
-
-For a configured backend ID, the wrapper:
-
-1. acquires one active-generation permit from the matching pool;
-2. waits in FIFO order when the active count equals `ConcurrencyLimit`;
-3. removes a canceled waiter and returns `ctx.Err()` when cancellation wins
-   before the permit is granted;
-4. invokes the next client only after a permit is granted; and
-5. releases the permit with `defer` after every success, nil response,
-   collaborator error, panic unwinding, or context outcome.
-
-Implement FIFO and cancellation explicitly with a mutex and an ordered waiter
-list. A channel used only as a counting semaphore is insufficient because it
-does not define FIFO ordering or safe removal of canceled waiters.
-
-Permit grant and cancellation must have one lock-protected linearization
-point. If cancellation removes the waiter first, do not invoke the next
-client. If grant wins first, invoke the next client with the caller's context;
-the next client may then observe cancellation normally. Never lose or
-double-release a permit in this race.
-
-Releasing a permit transfers it to the oldest non-canceled waiter before
-making it generally available. Different backend pools never share admission
-or active counts.
-
-The active wrapper enforces its limit even if an internal caller invokes it
-without a run admission lease. Bounded backlog is guaranteed for ordinary
-engine `Run` calls by the runner admission path; no public API exposes the
-wrapped internal client directly.
-
-### Engine Assembly
-
-In `NewEngine`, after constructing the validated backend registry:
-
-1. obtain `backendRegistry.CapacityPolicies()`;
-2. construct one `capacity.Manager`;
-3. construct the selected base internal LLM client exactly as today;
-4. wrap that base client with `capacity.NewClient`; and
-5. pass both the wrapped client and manager-as-admitter to the runner.
-
-Every `NewEngine` call constructs a distinct manager. Do not cache managers,
-pools, or policies in package globals. The wrapper must be applied after a
-public injected client is adapted to `internal/llm.Client`, so built-in and
-injected clients receive identical scheduling behavior.
-
-If `NewManager` reports a defensive configuration error, make `NewEngine`
-return an error matching `ErrInvalidConfig`.
-
-`Prepare` does not use the manager. An injected client remains required to be
-safe for concurrent calls because different backend pools and unlimited
-backends may still invoke it concurrently.
-
-### Shared Two-Phase Preparation
-
-Refactor `internal/usecase.Runner` so `Prepare` and `Run` share one preparation
-pipeline with two private phases. Do not duplicate prompt/profile/backend
-selection or execution precedence.
-
-The first phase resolves only the state required before admission:
-
-1. validate `PromptID`;
-2. normalize the direct session ID;
-3. load the prompt definition;
-4. hash the original prompt definition at its existing error-order position;
-5. select and load the execution profile;
-6. resolve the selected backend;
-7. resolve and validate the effective execution target and credentials; and
-8. resolve the effective output contract without loading its schema.
-
-Return a private state value containing the loaded definition, normalized
-direct session, prompt-definition hash, selected profile ID, effective target,
-numeric-presence metadata, effective output contract, and preparation start
-time. Keep this value private to `internal/usecase`.
-
-The second phase consumes that state and performs:
-
-1. structured-output schema loading;
-2. artifact loading and input hashing;
-3. message and prompt-session rendering;
-4. direct-session application;
-5. rendered-prompt hashing; and
-6. `PreparedRun` construction and timing.
-
-Preserve every existing precedence rule, error identity, direct-session
-template bypass, hash input, selected identity, copy guarantee, and timing
-field. Do not reload the prompt, profile, or backend between phases.
-
-`Runner.Prepare` records its start time, runs both phases consecutively, and
-never calls admission. Its behavior and error ordering remain unchanged.
-
-`Runner.Run` records its existing run start time, runs the first preparation
-phase, and then calls:
-
-```go
-release, err := r.admitter.Admit(ctx, effectiveBackendID)
-```
-
-Use a narrow use-case-owned interface with the same signature:
-
-```go
-type RunAdmitter interface {
-	Admit(context.Context, string) (func(), error)
-}
-```
-
-A nil admitter means unlimited behavior for internal constructors and tests.
-On successful admission, immediately `defer release()` around the remainder of
-the run. Then run the second preparation phase, initial generation,
-validation, and all repair attempts.
-
-If admission returns internal `capacity.ErrCapacityExceeded`, add useful
-backend context without changing its identity. If it returns `ctx.Err()`,
-preserve that identity directly rather than recategorizing it as invalid
-request or generation failure.
-
-This refactor intentionally replaces the current literal `Run`-calls-`Prepare`
-implementation with shared private phases. Update current-state documentation
-to describe one shared pipeline rather than retaining an inaccurate call-graph
-claim.
-
-### Generation And Repair Lifetime
-
-The admission lease covers the entire accepted run:
-
-- second-phase preparation;
-- initial generation;
-- validation;
-- every repair; and
-- all failure and cancellation exits.
-
-Preparation and validation do not hold an active-generation permit. The
-wrapped client acquires a permit only around each actual `Generate` call.
-
-The runner's initial generation already carries the effective backend ID in
-`GenerateRequest.Target`. Preserve that value. `RepairRequest.Target` and the
-default repairer's generated request must continue carrying the same backend
-ID, allowing each repair to reacquire the same pool's active permit.
-
-When testing or constructing `NewRunnerWithRepairer`, pass the same wrapped
-client to both the runner and `NewDefaultOutputRepairer`. Do not add capacity
-state to `RepairRequest`, `ExecutionTarget`, or public generation values.
-
-A repair remains within its existing admission lease. It waits for a FIFO
-active permit but never performs a second bounded admission and therefore
-cannot fail merely because later runs filled the admission capacity.
-
-### Error And Cancellation Semantics
-
-The required public outcomes are:
-
-| Situation | Required error identity |
-| --- | --- |
-| Admission capacity is full | `ErrCapacityExceeded` only; not `ErrInvalidRequest` or `ErrLLMGenerate`. |
-| Context is done before admission succeeds | Preserve `ctx.Err()`; do not return capacity exhaustion. |
-| Context cancels while waiting for an active permit | Preserve `ctx.Err()` through the existing `ErrLLMGenerate` generation category. |
-| Wrapped client fails after permit acquisition | Preserve existing `ErrLLMGenerate` and collaborator identities. |
-| Preparation or validation fails after admission | Preserve its existing category and release admission. |
-
-Maintain the existing rule that `Run` returns no partial result on any
-operational error. Do not add queue status to errors or results.
-
-`RunResult.Duration` continues to start at runner entry and therefore includes
-pre-admission resolution, accepted preparation, and active-permit waiting.
-`PreparedRun.DurationMS` continues to cover only its shared preparation phases;
-it does not include later generation waiting. Capacity-rejected calls have no
-result or timing value.
-
-### Ownership And Concurrency Safety
-
-The registry, capacity policy map, pool map, and per-pool limits are immutable
-after engine construction. Only admission counts, active counts, and waiter
-lists are mutable and must be protected by the owning pool mutex.
-
-Do not retain public queue pointers, caller request values, contexts, or
-generation requests after their call completes. A canceled waiter must be
-unlinked so its context and request cannot remain reachable from the pool.
-
-Do not hold a pool mutex while:
-
-- loading or rendering prompts;
-- reading artifacts or schemas;
-- invoking a model client;
-- validating output;
-- closing a waiter notification channel if the implementation could re-enter
-  pool code; or
-- calling consumer code.
-
-No scheduler operation may spawn a goroutine whose lifetime outlasts the
-calling `Run`. The zero steady-state goroutine count is part of the
-in-process/no-worker-lifecycle design.
-
-## Test Ownership
-
-Use this ownership split and avoid repeating the full policy matrix at every
-layer:
-
-- `internal/backend/registry_test.go` owns normalization, validation, the exact
-  OpenRouter policy, the custom default queue, explicit zero, unlimited
-  omission, and policy-map copying.
-- `internal/capacity/manager_test.go` owns admission bounds, idempotent release,
-  FIFO active permits, cancellation races, capacity recovery, independent
-  pools, unlimited IDs, and observed peak concurrency.
-- `internal/capacity/client_test.go` owns wrapper request/response/error
-  transparency and the rule that cancellation before grant does not invoke the
-  next client. Combine these with manager tests if one coherent package test
-  expresses the behavior more clearly.
-- `internal/usecase/runner_test.go` owns two-phase preparation parity, pool
-  selection, admission before expensive work, admission release across run
-  exits, `Prepare` bypass, and repair reuse of the admitted backend.
-- Root external-package tests own public configuration conversion, assembled
-  engine-local behavior, endpoint-override routing, injected-client limiting,
-  and public capacity/context error identities.
-- Existing model-client HTTP tests remain unchanged because scheduling does
-  not alter the OpenAI-compatible wire contract.
-
-Concurrency tests must use test-owned limits such as one or two and
-channel-controlled blocking clients. Record observed active and peak counts
-under a mutex or atomics. Do not use `time.Sleep` to infer queue state.
-Package-internal tests may inspect a waiter list under its mutex through a
-small test helper when necessary to establish deterministic FIFO ordering; do
-not add production metrics or hooks solely for tests.
-
-Do not add separate tests for trivial scalar copies when registry or assembled
-behavior already protects them.
-
-## Stage 1 — Backend Policy And Public Configuration
-
-**Status:** Complete.
-
-### Goal
-
-Add the public and internal backend policy representation, normalize all
-configured states, and expose immutable normalized policies without changing
-runtime scheduling yet.
-
-### Work
-
-1. Add `ConcurrencyLimit` and `QueueCapacity` to `Backend` in `backends.go`
-   with exact GoDoc for unlimited, defaulted, explicit-zero, invalid, and
-   engine-scoped behavior.
-2. Convert the public queue pointer into scalar value plus presence in
-   `WithBackend`; do not retain the pointer.
-3. Add the internal backend policy fields and
-   `BackendCapacityPolicy` to `internal/domain/domain.go`.
-4. Add the two registry-owned constants and configure the built-in OpenRouter
-   definition with 16 and 1024.
-5. Extend `normalizeBackend` with the fixed validation, defaulting, explicit
-   zero, and overflow rules.
-6. Add `Registry.CapacityPolicies`, returning only limited policies in a fresh
-   map.
-7. Update existing backend composite literals and assertions only where the
-   new fields are relevant. Continue using keyed literals.
-
-### Tests
-
-1. Extend the exact built-in registry test with the OpenRouter limit and queue.
-2. Add one coherent table covering unlimited omission, default queue,
-   explicit-zero queue, negative values, queue-without-limit, and total
-   overflow.
-3. Extend the registry copy/normalization test to prove returned policy maps
-   cannot mutate registry state.
-4. Add root coverage only if needed to prove the public pointer/presence
-   conversion; do not reproduce registry validation cases at the facade.
-
-### Focused Validation
-
-Run:
-
-```sh
-gofmt -w backends.go internal/domain/domain.go \
-  internal/backend/registry.go internal/backend/registry_test.go
-go test . ./internal/backend
-go vet . ./internal/backend
-git diff --check
-```
-
-Include another touched Go test file in `gofmt` only if it actually changed.
-
-### Completion Gate
-
-This stage is complete when every public configuration state has one normalized
-internal meaning, OpenRouter exposes exactly 16/1024, custom backends remain
-unlimited by omission, and no runtime call is scheduled yet.
-
-## Stage 2 — Engine-Local Capacity Manager
-
-**Status:** Complete.
-
-### Goal
-
-Implement and prove the bounded admission mechanism and FIFO active-generation
-client wrapper independently of runner orchestration.
-
-### Work
-
-1. Add `internal/capacity/manager.go` with the manager, immutable policy copy,
-   per-backend pools, internal error, immediate admission, idempotent release,
-   and FIFO context-aware active permits.
-2. Add `internal/capacity/client.go` with the transparent `llm.Client` wrapper.
-3. Use mutex-protected waiter state and an ordered list; explicitly resolve
-   grant-versus-cancel races.
-4. Ensure unlimited and independent-pool fast paths avoid queue allocation.
-5. Do not start workers, timers, cleanup goroutines, or process-global state.
-
-### Tests
-
-1. Add a compact constructor-validation table for blank IDs, non-positive
-   limits, negative queues, and total-capacity overflow.
-2. With a small configured policy, prove that exactly
-   `limit + queueCapacity` admissions succeed, the next matches
-   `ErrCapacityExceeded`, and a release permits another admission.
-3. Prove release is idempotent.
-4. Drive more blocked client calls than the active limit and assert observed
-   peak concurrency never exceeds that limit.
-5. Prove FIFO order with deterministic queue-entry synchronization.
-6. Cancel the first and a middle waiter and prove they are removed, never call
-   the wrapped client, and do not block later waiters.
-7. Exercise the grant/cancel race repeatedly under `go test -race`, asserting
-   no permit leak or double invocation.
-8. Prove different backend IDs proceed independently and blank, unknown, or
-   nil-manager paths remain unlimited.
-9. Prove request values, successful and nil responses, and collaborator errors
-   pass through unchanged after permit acquisition.
-
-### Focused Validation
-
-Run:
-
-```sh
-gofmt -w internal/capacity/manager.go \
-  internal/capacity/manager_test.go \
-  internal/capacity/client.go \
-  internal/capacity/client_test.go
-go test ./internal/capacity
-go test -race ./internal/capacity
-go vet ./internal/capacity
-git diff --check
-```
-
-If tests are combined into one file, omit the nonexistent file from `gofmt`.
-
-### Completion Gate
-
-This stage is complete when the standalone component enforces relational
-admission and active limits, FIFO cancellation is race-safe, separate pools
-are independent, and the wrapper is transparent apart from waiting.
-
-## Stage 3 — Shared Preparation And Early Run Admission
-
-**Status:** Complete.
-
-### Goal
-
-Refactor runner preparation into one shared two-phase pipeline and place
-bounded admission after backend resolution but before schema, artifact, and
-rendering work.
-
-### Work
-
-1. Add the private pre-admission preparation state and split the existing
-   `Prepare` logic according to the fixed design.
-2. Make `Runner.Prepare` call both phases without an admitter.
-3. Add the `RunAdmitter` interface and runner field.
-4. Update `NewRunner` and `NewRunnerWithRepairer` to accept the optional
-   admitter; update internal call sites with nil until root assembly is wired.
-5. Change `Runner.Run` to use the first phase, admit by effective backend ID,
-   defer the returned release, and then use the second phase.
-6. Preserve all existing error precedence, target resolution, hashes,
-   metadata, session behavior, and timing.
-7. Return capacity and context errors with the fixed identities. Do not invoke
-   later collaborators after rejection.
-
-### Tests
-
-1. Keep the existing `Run`/`Prepare` parity coverage passing to prove the
-   shared phases do not drift.
-2. Add a fake admitter that records backend IDs and release calls.
-3. Prove a backend-selected run admits with the selected ID even when the
-   endpoint is overridden.
-4. Prove an endpoint-only run uses the unlimited/blank identity and that
-   `Prepare` never calls admission.
-5. Reject admission and assert schema, artifact, renderer, validator, repairer,
-   and LLM collaborators are not invoked.
-6. Prove admission is released after one successful run and representative
-   second-phase, generation, and validation errors. Prefer a small table around
-   the single `defer` invariant rather than duplicating every error test.
-7. Retain direct-session, backend precedence, credential, hashing, and repair
-   tests unchanged except for constructor arguments.
-
-### Focused Validation
-
-Run:
-
-```sh
-gofmt -w internal/usecase/runner.go \
-  internal/usecase/runner_test.go
-go test ./internal/usecase
-go test -race ./internal/usecase
-go vet ./internal/usecase
-git diff --check
-```
-
-### Completion Gate
-
-This stage is complete when `Prepare` remains unrestricted, `Run` admits after
-one canonical routing phase and before expensive completion work, every exit
-releases admission, and existing preparation semantics remain unchanged.
-
-## Stage 4 — Engine Assembly And Public Runtime Contract
-
-**Status:** Complete.
-
-### Goal
-
-Wire one manager into each engine, schedule built-in and injected clients,
-expose the capacity error, and prove assembled runtime behavior.
-
-### Work
-
-1. Add public `ErrCapacityExceeded` and its exact GoDoc in `engine.go`.
-2. Map internal capacity exhaustion in `errors.go`.
-3. Construct the manager from the registry policy snapshot in `NewEngine`.
-4. Wrap the selected internal client after built-in or injected-client
-   selection and pass the manager and wrapped client to the runner.
-5. Update `Engine`, `NewEngine`, `Run`, `WithLLMClient`, and `LLMClient` GoDoc
-   only where concurrency, capacity, or cancellation statements change.
-6. Ensure manager-construction errors match `ErrInvalidConfig`.
-7. For internal repair coverage, construct the default repairer with the same
-   wrapped client used by its runner and confirm repair target backend identity
-   remains intact.
-
-### Tests
-
-1. Add an external-package assembled test with a small custom limit and a
-   blocking injected client; assert peak generation equals or remains below
-   the configured limit.
-2. With queue capacity zero, block one accepted run before generation and
-   assert the next matching-backend run returns `ErrCapacityExceeded`, does not
-   match `ErrInvalidRequest` or `ErrLLMGenerate`, returns no result, and never
-   reaches expensive collaborators or the client.
-3. In the same or another focused workflow, prove an endpoint override remains
-   in the selected backend's pool.
-4. Prove two engines with the same backend ID have independent capacity.
-5. Prove an unlimited custom backend and an endpoint-only profile preserve
-   concurrent behavior.
-6. Cancel a call waiting for an active permit; assert it matches both
-   `context.Canceled` and `ErrLLMGenerate`, never invokes the injected client,
-   and leaves capacity reusable.
-7. Add one internal repair workflow with concurrent runs or controlled permits
-   showing initial and repair generations never exceed the same backend limit
-   and repairs do not perform a second admission.
-8. Extend the public error sentinel contract test with
-   `ErrCapacityExceeded`.
-
-Avoid a second HTTP-level concurrency suite: the capacity client tests and one
-assembled injected-client workflow already protect the shared wrapper used by
-the built-in client.
-
-### Focused Validation
-
-Run:
-
-```sh
-gofmt -w engine.go errors.go backends.go \
-  internal/usecase/runner.go internal/usecase/runner_test.go \
-  engine_test.go public_contract_test.go
-go test . ./internal/backend ./internal/capacity ./internal/usecase
-go test -race . ./internal/capacity ./internal/usecase
-go vet . ./internal/backend ./internal/capacity ./internal/usecase
-git diff --check
-```
-
-Add any newly created capacity files to `gofmt` when they changed in this
-stage.
-
-### Completion Gate
-
-This stage is complete when every engine has independent pools, limited runs
-are bounded and FIFO at generation, endpoint routing is correct, capacity and
-context errors are stable, repairs reuse admission, and both client kinds pass
-through the same wrapper.
-
-## Stage 5 — Durable Documentation And Final Validation
-
-**Status:** Complete.
-
-### Goal
-
-Move implemented contracts into their durable owners, record compatibility
-impact, and validate the complete repository.
-
-### Work
-
-1. Review every changed exported declaration. Ensure GoDoc is the canonical
-   owner of exact field types, nil/zero semantics, defaulting, error identity,
-   engine scope, concurrency safety, cancellation, and source compatibility.
-2. Update `doc.go` so its concurrency summary acknowledges backend scheduling
-   while continuing to require injected collaborators to be concurrency-safe.
-3. Update `docs/consumers/pkg-promptkit.md` with task-oriented examples for:
-   - a limited local backend;
-   - omitted queue capacity selecting 1024;
-   - explicit zero queue capacity; and
-   - handling `ErrCapacityExceeded`.
-   Keep exact field semantics in GoDoc rather than duplicating a full table.
-4. Add `docs/internal/capacity.md` as the durable owner of pool lifecycle,
-   admission, FIFO active permits, cancellation, client wrapping, and test
-   ownership.
-5. Add `internal/capacity` to `docs/internal/overview.md`.
-6. Update `docs/policy/architecture.md` to include the implemented component
-   and root assembly dependency without turning policy into an API reference.
-7. Update `docs/internal/runner.md` to describe the shared two-phase
-   preparation pipeline, early bounded admission, lease lifetime, generation
-   permits, repairs, capacity failures, and cancellation.
-8. Review `docs/formats.md`; add only a concise link or clarification if needed
-   to explain that endpoint overrides preserve backend capacity identity.
-   Do not add concurrency fields to YAML.
-9. Do not change the OpenAI-compatible integration contract or
-   `docs/internal/llm.md` unless implementation changes their current
-   statements; scheduling is outside the provider wire contract and concrete
-   model-client implementation.
-10. Record in the implementation handoff that built-in OpenRouter now limits
-    active generations to 16 with queue capacity 1024 and that the release
-    must be a pre-`v1` minor release. Do not edit the release procedure or
-    create a tag.
-11. After every check passes, set `concurrency.md`, this implementation plan,
-    and each stage status to `Complete`. Do not remove the roadmaps in the
-    implementation change; lifecycle retirement follows review.
-
-### Full Validation
-
-Run the complete sequence from `docs/development.md`:
-
-```sh
-go test ./...
-go test -race ./...
-go vet ./...
-go build ./...
-go run ./examples/go-library/prepare
-gofmt -l $(git ls-files '*.go')
-git diff --check
-```
-
-The formatting command must produce no paths. Follow every added or changed
-Markdown link and confirm its target and heading exist.
-
-Also inspect:
-
-```sh
-git status --short
-git diff --stat
-git diff
-```
-
-Confirm that:
-
-- only intended backend, capacity, runner, facade, test, documentation, and
-  roadmap files changed;
-- no `go.work`, `go.work.sum`, local module replacement, credential,
-  generated binary, coverage output, or unrelated change was introduced;
-- the built-in OpenRouter policy is exactly 16/1024;
-- custom and endpoint-only backends remain unlimited by omission;
-- explicit queue zero is distinguishable from omission;
-- no capacity value enters execution targets, generated requests, stable JSON,
-  prompt/profile YAML, or provider payloads;
-- every engine owns distinct pools with no package-global mutable state;
-- every initial and repair generation uses the active permit wrapper;
-- capacity and waiter state is released on success, error, panic unwinding,
-  and cancellation;
-- concurrency tests use deterministic coordination rather than sleeps;
-- current-state documentation describes implemented behavior rather than
-  referring readers to the roadmaps; and
-- the feature and implementation roadmaps contain no unresolved work marked
-  complete.
-
-### Completion Gate
-
-The implementation is complete only when every target-end-state item in
-`concurrency.md` is implemented, race-enabled tests demonstrate the configured
-limits and cancellation safety, durable contracts no longer depend on roadmap
-prose, and the OpenRouter compatibility change is clearly reported for the
-next minor release.
-
-## Implementation Handoff
-
-Backend-specific capacity management is implemented and has passed the complete
-repository validation sequence. The built-in OpenRouter backend now permits 16
-active generations and a waiting capacity of 1024. Custom backends remain
-unlimited when their limit is omitted, and endpoint-only profiles remain
-unlimited.
-
-Publishing this behavior requires a pre-`v1` minor release. Its release notes
-must identify that unusually high concurrent OpenRouter use can now wait or
-return `ErrCapacityExceeded`. This implementation does not change a module
-version or create a tag.
-
-## Open Questions
-
-None. The feature roadmap and this plan fix the public representation,
-registry defaults, admission bound, FIFO generation behavior, early-routing
-refactor, cancellation races, error identities, engine and repair lifetimes,
-test ownership, compatibility treatment, and non-goals required for
-implementation.

engine.go

@@ -6,27 +6,17 @@ import (
 	"fmt"
 	"io/fs"
 	"net/http"
-	"os"
-	"path/filepath"
-	"strings"
 	"time"
 
-	artifactadapter "gitea.maximumdirect.net/eric/promptkit/internal/artifact"
-	"gitea.maximumdirect.net/eric/promptkit/internal/backend"
-	"gitea.maximumdirect.net/eric/promptkit/internal/capacity"
-	"gitea.maximumdirect.net/eric/promptkit/internal/defaults"
+	"gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
-	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
-	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
-	"gitea.maximumdirect.net/eric/promptkit/internal/profile/builtin"
-	"gitea.maximumdirect.net/eric/promptkit/internal/prompt"
-	"gitea.maximumdirect.net/eric/promptkit/internal/promptdef"
+	internalengine "gitea.maximumdirect.net/eric/promptkit/internal/engine"
+	"gitea.maximumdirect.net/eric/promptkit/internal/observation"
 	"gitea.maximumdirect.net/eric/promptkit/internal/usecase"
-	"gitea.maximumdirect.net/eric/promptkit/internal/validate"
 )
 
 // ErrInvalidConfig identifies invalid engine construction, including missing
-// required configuration, invalid options or backend registrations, and a nil
+// effective configuration, invalid options, and a nil
 // Engine receiver.
 var ErrInvalidConfig = errors.New("invalid engine configuration")
 
@@ -45,17 +35,17 @@ var (
 	// nor the selected prompt's default profile is present. Such an error also
 	// matches ErrInvalidRequest.
 	ErrProfileRequired = errors.New("profile selection is required")
-	// ErrPromptLoad identifies a failure to read, decode, validate, select, or
-	// hash a prompt definition, except for the not-found case represented by
-	// ErrPromptNotFound.
+	// ErrPromptLoad identifies a prompt selection failure other than the
+	// not-found case represented by ErrPromptNotFound. Invalid definitions
+	// instead fail construction with ErrInvalidConfig.
 	ErrPromptLoad = errors.New("failed to load prompt definition")
-	// ErrProfileLoad identifies a failure to read, decode, validate, or select
-	// an execution profile or resolve its backend, except for the profile
-	// not-found case represented by ErrProfileNotFound.
+	// ErrProfileLoad identifies a profile selection or resolution failure other
+	// than the not-found case represented by ErrProfileNotFound. Invalid
+	// definitions instead fail construction with ErrInvalidConfig.
 	ErrProfileLoad = errors.New("failed to load execution profile")
-	// ErrAPIKeyEnvMissing identifies an APIKeyEnv whose environment variable is
-	// unset or empty when no direct RunRequest.APIKey takes precedence. Such an
-	// error also matches ErrInvalidRequest.
+	// ErrAPIKeyEnvMissing identifies an explicitly required APIKeyEnv whose
+	// environment variable is unset or empty after direct RunRequest.APIKey
+	// precedence is applied. Such an error also matches ErrInvalidRequest.
 	ErrAPIKeyEnvMissing = errors.New("api_key_env points to an unset environment variable")
 	// ErrArtifactLoad identifies a failure to resolve an input artifact. Errors
 	// returned by an injected ArtifactReader remain available through errors.Is.
@@ -63,41 +53,64 @@ var (
 	// ErrPromptRender identifies a failure to render prompt messages or the
 	// session ID from the resolved inputs and variables.
 	ErrPromptRender = errors.New("failed to render prompt")
-	// ErrCapacityExceeded identifies a Run rejected because the selected backend
-	// already admitted ConcurrencyLimit + QueueCapacity calls. It is not an
-	// invalid request, an LLM or provider rate-limit response, or ErrLLMGenerate.
+	// ErrCapacityExceeded identifies a Run or RunPrepared rejected because the
+	// selected backend already admitted ConcurrencyLimit + QueueCapacity calls.
+	// A [CapacityError] reports the selected backend ID. It is not an invalid
+	// request, an LLM or provider rate-limit response, or ErrLLMGenerate.
 	ErrCapacityExceeded = errors.New("backend capacity exceeded")
 	// ErrLLMGenerate identifies a model-client failure or a nil successful
-	// response. Errors returned by an injected LLMClient remain available
-	// through errors.Is.
+	// response. A built-in OpenAI-compatible non-2xx response is available as a
+	// [GenerationError]. Errors returned by an injected LLMClient remain
+	// available through errors.Is.
 	ErrLLMGenerate = errors.New("failed to generate output")
+	// ErrTransport identifies a failed HTTP request by the built-in model
+	// client. Generation errors in this category also match ErrLLMGenerate and
+	// retain the transport cause for errors.Is and errors.As. It does not imply
+	// that the provider received no work.
+	ErrTransport = errors.New("model transport failed")
+	// ErrMalformedResponse identifies an invalid successful model response or
+	// a nil successful response from an injected client. Generation errors in
+	// this category also match ErrLLMGenerate.
+	ErrMalformedResponse = errors.New("malformed model response")
+	// ErrRetriesExhausted identifies a built-in generation sequence whose
+	// positive transient retry budget ended with another eligible HTTP status.
+	// RetryExhaustedError retains the final cause and actual attempt count.
+	ErrRetriesExhausted = errors.New("transient model retries exhausted")
 	// ErrValidation identifies an operational failure to load or compile a
 	// schema or validate output. A completed validation whose Status is
 	// ValidationFailed is returned in RunResult without this error.
 	ErrValidation = errors.New("failed to validate output")
 )
 
-// Engine prepares and runs Promptkit prompt requests.
+// Engine inspects prompts and profiles, enumerates prompt requirements, and
+// prepares and runs Promptkit prompt requests.
 //
-// An Engine is safe for concurrent calls to [Engine.Prepare] and [Engine.Run].
-// Each Engine owns independent backend-capacity pools that coordinate Run
-// admission and model generation. Injected collaborators may still be invoked
-// concurrently across different backend pools or for unlimited backends.
+// An Engine is safe for concurrent calls to [Engine.InspectConfiguration],
+// [Engine.InspectPrompt],
+// [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.Prepare], [Engine.PrepareExecution],
+// [Engine.Run], and [Engine.RunPrepared]. Each Engine owns independent
+// backend-capacity pools that coordinate Run and RunPrepared admission and
+// model generation. Injected collaborators may still be invoked concurrently
+// across different backend pools or for unlimited backends.
 type Engine struct {
-	runner *usecase.Runner
+	internal *internalengine.Engine
 }
 
-// Config selects the directory-backed sources and built-in model-client
-// transport used by [NewEngine]. Config has no stable JSON representation.
+// ConfigurationSources selects ordered ConfigKit underlays and an optional
+// highest-priority main input. Each underlay's inputs strictly compose within
+// their group; later underlays overlay earlier ones. A nil Main uses
+// ConfigKit's empty source. A non-nil Main is passed through unchanged and
+// may fail construction if invalid.
+type ConfigurationSources struct {
+	Underlays []configkit.Underlay
+	Main      *configkit.Input
+}
+
+// Config selects configuration sources and built-in model-client transport
+// used by [NewEngine]. Config has no stable JSON representation.
 type Config struct {
-	// PromptDir is the directory searched recursively for prompt definitions.
-	// It is required unless a WithPromptFS or WithPromptFile option supplies the
-	// prompt source.
-	PromptDir string
-	// ProfileDir is an optional directory whose profiles take precedence over
-	// embedded built-in profiles. An empty value selects only built-ins unless
-	// profile options are also supplied.
-	ProfileDir string
+	// Sources combines maintained definitions with caller-selected inputs.
+	Sources ConfigurationSources
 	// SchemaDir is the root for JSON Schema files. An empty value uses the
 	// current directory. WithSchemaFS or WithSchemaFile replaces this source.
 	SchemaDir string
@@ -110,17 +123,19 @@ type Config struct {
 	// Timeout or the 10-minute default. The supplied client is not mutated. This
 	// field is ignored when WithLLMClient is used.
 	HTTPClient *http.Client
+	// TransientRetries selects additional built-in HTTP attempts after eligible
+	// provider statuses. Nil defaults to two; zero disables retries. Negative
+	// values are invalid. Injected LLMClient implementations are never retried
+	// automatically.
+	TransientRetries *int
 }
 
 // Option customizes engine construction.
 //
 // NewEngine applies options in argument order and ignores nil options. Within
-// each prompt-source, profile-source, in-memory-profile, schema-source,
-// model-client, and artifact-reader category, the last non-nil valid option
-// replaces earlier options in that category. WithBackend is the additive
-// exception: unique registrations accumulate, and a repeated backend ID is an
-// error rather than a replacement. An invalid option fails construction even
-// if a later option would replace it.
+// each schema-source, model-client, and artifact-reader category, the last
+// non-nil valid option replaces earlier options. An invalid option fails
+// construction even if a later option would replace it.
 type Option interface {
 	apply(*engineOptions) error
 }
@@ -132,32 +147,23 @@ func (f optionFunc) apply(options *engineOptions) error {
 }
 
 type engineOptions struct {
-	llmClient       llm.Client
-	artifactReader  artifactadapter.Reader
-	promptDefs      promptdef.Repository
-	profiles        profile.Repository
-	memoryProfiles  profile.Repository
-	backends        []domain.Backend
-	validator       validate.Validator
-	promptSource    bool
-	profileSource   bool
-	memorySource    bool
-	validatorSource bool
-	artifactSource  bool
+	internal internalengine.Options
 }
 
-// WithLLMClient replaces the built-in model client used by [Engine.Run].
+// WithLLMClient replaces the built-in model client used by [Engine.Run] and
+// [Engine.RunPrepared].
 //
 // A nil client makes NewEngine fail with ErrInvalidConfig. The Engine schedules
 // Generate calls according to the selected backend's capacity policy, but the
 // client may still be called concurrently across different backend pools or for
-// unlimited backends. The client is not used by [Engine.Prepare].
+// unlimited backends. The client is not used by [Engine.Prepare] or
+// [Engine.PrepareExecution].
 func WithLLMClient(client LLMClient) Option {
 	return optionFunc(func(options *engineOptions) error {
 		if client == nil {
 			return ErrInvalidConfig
 		}
-		options.llmClient = publicLLMClientAdapter{client: client}
+		options.internal.SetLLMClient(publicLLMClientAdapter{client: client})
 		return nil
 	})
 }
@@ -172,106 +178,7 @@ func WithArtifactReader(reader ArtifactReader) Option {
 		if reader == nil {
 			return ErrInvalidConfig
 		}
-		options.artifactReader = publicArtifactReaderAdapter{reader: reader}
-		options.artifactSource = true
-		return nil
-	})
-}
-
-// WithPromptFS loads prompt definitions from fsys under root.
-//
-// The source uses the same strict prompt YAML rules as configured prompt
-// directories, and prompt content_file paths resolve within this source.
-// fsys must be non-nil and root must be non-empty; otherwise NewEngine fails
-// with ErrInvalidConfig. This option replaces Config.PromptDir and earlier
-// prompt-source options.
-func WithPromptFS(fsys fs.FS, root string) Option {
-	return optionFunc(func(options *engineOptions) error {
-		if fsys == nil {
-			return ErrInvalidConfig
-		}
-		if strings.TrimSpace(root) == "" {
-			return ErrInvalidConfig
-		}
-		options.promptDefs = promptdef.NewFSRepository(fsys, root)
-		options.promptSource = true
-		return nil
-	})
-}
-
-// WithPromptFile loads prompt definitions from the single prompt file at path.
-//
-// Relative prompt content_file paths resolve from the file's directory. path
-// must name an existing non-directory file when NewEngine applies the option.
-// This option replaces Config.PromptDir and earlier prompt-source options.
-func WithPromptFile(path string) Option {
-	return optionFunc(func(options *engineOptions) error {
-		fsys, root, err := fileSource(path)
-		if err != nil {
-			return err
-		}
-		options.promptDefs = promptdef.NewFSRepository(fsys, root)
-		options.promptSource = true
-		return nil
-	})
-}
-
-// WithProfileFS loads execution profiles from fsys under root.
-//
-// Profiles from this source overlay built-in profiles. Profile YAML must use
-// api_key_env for environment-based credentials; raw API keys are rejected.
-// fsys must be non-nil and root must be non-empty; otherwise NewEngine fails
-// with ErrInvalidConfig. This option replaces Config.ProfileDir and earlier
-// file or FS profile-source options, but remains below WithProfiles in
-// precedence.
-func WithProfileFS(fsys fs.FS, root string) Option {
-	return optionFunc(func(options *engineOptions) error {
-		if fsys == nil {
-			return ErrInvalidConfig
-		}
-		if strings.TrimSpace(root) == "" {
-			return ErrInvalidConfig
-		}
-		options.profiles = profile.NewFSRepository(fsys, root)
-		options.profileSource = true
-		return nil
-	})
-}
-
-// WithProfileFile loads execution profiles from the single profile file at path.
-//
-// The profile overlays built-in profiles. Profile YAML must use api_key_env for
-// environment-based credentials; raw API keys are rejected. path must name an
-// existing non-directory file when NewEngine applies the option. This option
-// replaces Config.ProfileDir and earlier file or FS profile-source options,
-// but remains below WithProfiles in precedence.
-func WithProfileFile(path string) Option {
-	return optionFunc(func(options *engineOptions) error {
-		fsys, root, err := fileSource(path)
-		if err != nil {
-			return err
-		}
-		options.profiles = profile.NewFSRepository(fsys, root)
-		options.profileSource = true
-		return nil
-	})
-}
-
-// WithProfiles configures in-memory profiles that take precedence over
-// configured profile files and built-in profiles.
-//
-// NewEngine validates and copies every profile. IDs must be unique within one
-// call. An invalid profile, duplicate ID, or unsupported ExtraParams value
-// makes construction fail with ErrInvalidConfig. Repeating WithProfiles
-// replaces the complete earlier in-memory set rather than merging it.
-func WithProfiles(profiles ...Profile) Option {
-	return optionFunc(func(options *engineOptions) error {
-		repo, err := newMemoryProfileRepository(profiles)
-		if err != nil {
-			return err
-		}
-		options.memoryProfiles = repo
-		options.memorySource = true
+		options.internal.SetArtifactReader(publicArtifactReaderAdapter{reader: reader})
 		return nil
 	})
 }
@@ -284,15 +191,7 @@ func WithProfiles(profiles ...Profile) Option {
 // replaces Config.SchemaDir and earlier schema-source options.
 func WithSchemaFS(fsys fs.FS, root string) Option {
 	return optionFunc(func(options *engineOptions) error {
-		if fsys == nil {
-			return ErrInvalidConfig
-		}
-		if strings.TrimSpace(root) == "" {
-			return ErrInvalidConfig
-		}
-		options.validator = validate.NewFSValidator(fsys, root)
-		options.validatorSource = true
-		return nil
+		return options.internal.SetSchemaFS(fsys, root)
 	})
 }
 
@@ -303,28 +202,24 @@ func WithSchemaFS(fsys fs.FS, root string) Option {
 // replaces Config.SchemaDir and earlier schema-source options.
 func WithSchemaFile(path string) Option {
 	return optionFunc(func(options *engineOptions) error {
-		fsys, root, err := fileSource(path)
-		if err != nil {
-			return err
-		}
-		options.validator = validate.NewFSValidator(fsys, root)
-		options.validatorSource = true
-		return nil
+		return options.internal.SetSchemaFile(path)
 	})
 }
 
 // NewEngine constructs an Engine from configuration and options.
 //
-// Options are applied in order according to [Option]. PromptDir is required
-// unless a prompt-source option is present. Construction validates option
-// arguments, in-memory profiles, and backend registrations but defers reading
-// and validating prompt, file-backed profile, and schema contents until Prepare
-// or Run needs them.
+// ConfigKit composes maintained catalog inputs, PromptKit-owned definitions,
+// caller underlays, and the optional main input in that order. Construction
+// validates the final effective backend, profile, and prompt definitions and
+// captures referenced prompt content. A missing prompt collection is allowed.
+// Schema documents and request artifacts remain runtime resources.
 //
 // NewEngine returns an error matching ErrInvalidConfig for invalid
-// configuration, options, or backend-capacity policies. Each constructed
+// configuration or options, preserving a ConfigKit LoadError when loading
+// fails. Each constructed
 // Engine has independent backend-capacity pools. Construction does not perform
-// model requests or require credentials.
+// model requests or require credentials. A negative Config.TransientRetries
+// is invalid.
 func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 	var options engineOptions
 	for _, opt := range opts {
@@ -332,98 +227,108 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 			continue
 		}
 		if err := opt.apply(&options); err != nil {
-			return nil, fmt.Errorf("%w: %v", ErrInvalidConfig, err)
+			return nil, fmt.Errorf("%w: %w", ErrInvalidConfig, err)
 		}
 	}
 
-	promptDefs := options.promptDefs
-	if !options.promptSource {
-		if strings.TrimSpace(cfg.PromptDir) == "" {
-			return nil, fmt.Errorf("%w: prompt directory is required", ErrInvalidConfig)
-		}
-		promptDefs = promptdef.NewFilesystemRepository(cfg.PromptDir)
-	}
-
-	profiles := builtin.NewRepositoryWithDirectory(cfg.ProfileDir)
-	if options.profileSource {
-		profiles = builtin.NewRepositoryWithPrimary(options.profiles)
-	}
-	if options.memorySource {
-		profiles = profile.NewOverlayRepository(options.memoryProfiles, profiles)
-	}
-
-	backendRegistry, err := backend.NewRegistry(options.backends)
+	assembled, err := internalengine.New(
+		internalengine.Config{Sources: internalengine.ConfigurationSources{Underlays: cfg.Sources.Underlays, Main: cfg.Sources.Main}, SchemaDir: cfg.SchemaDir, Timeout: cfg.Timeout, HTTPClient: cfg.HTTPClient, TransientRetries: copyIntPtr(cfg.TransientRetries)},
+		options.internal,
+	)
 	if err != nil {
-		return nil, fmt.Errorf("%w: failed to construct backend registry: %v", ErrInvalidConfig, err)
+		return nil, fmt.Errorf("%w: %w", ErrInvalidConfig, err)
 	}
-
-	capacityManager, err := capacity.NewManager(backendRegistry.CapacityPolicies())
-	if err != nil {
-		return nil, fmt.Errorf("%w: failed to construct backend capacity manager: %v", ErrInvalidConfig, err)
-	}
-
-	validator := options.validator
-	if !options.validatorSource {
-		schemaDir := cfg.SchemaDir
-		if strings.TrimSpace(schemaDir) == "" {
-			schemaDir = defaults.SchemaDirDefault
-		}
-		validator = validate.NewStandardValidator(schemaDir)
-	}
-
-	llmClient := options.llmClient
-	if llmClient == nil {
-		var err error
-		llmClient, err = llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
-			Timeout:    cfg.Timeout,
-			HTTPClient: cfg.HTTPClient,
-		})
-		if err != nil {
-			return nil, fmt.Errorf("%w: %v", ErrInvalidConfig, err)
-		}
-	}
-	llmClient = capacity.NewClient(capacityManager, llmClient)
-
-	artifacts := options.artifactReader
-	if !options.artifactSource {
-		artifacts = artifactadapter.NewCompositeReader()
-	}
-
-	return &Engine{
-		runner: usecase.NewRunner(
-			promptDefs,
-			profiles,
-			backendRegistry,
-			artifacts,
-			prompt.NewGoRenderer(),
-			llmClient,
-			validator,
-			capacityManager,
-		),
-	}, nil
+	return &Engine{internal: assembled}, nil
 }
 
-func fileSource(name string) (fs.FS, string, error) {
-	cleanName := strings.TrimSpace(name)
-	if cleanName == "" {
-		return nil, "", ErrInvalidConfig
+// InspectPrompt resolves one explicit prompt definition without selecting a
+// profile or starting execution work.
+//
+// InspectPrompt requires a nonblank promptID. It passes nonblank promptID and
+// promptVersion values unchanged to the engine's ordinary, case-sensitive
+// prompt selection. An empty version succeeds only when that ID has one
+// version; a nonempty version selects one exact ID/version pair from the
+// captured configuration snapshot.
+//
+// A successful result proves that the selected definition and any referenced
+// message content files were structurally loaded. Inputs are returned in
+// definition order. DefaultProfileID is declared metadata only and is not
+// resolved. OutputContract is the normalized declared contract, with a JSON
+// Schema path when declared but without loading or compiling that schema.
+// PromptHash is the same opaque equality value as PreparedRun.PromptHash for
+// the selected captured definition; its spelling, length,
+// encoding, algorithm, and security properties are not contracts.
+//
+// This method does not return prompt bodies, templates, source paths, schemas,
+// rendered messages, or execution settings. It does not resolve a profile or
+// credential, read artifacts or schemas, render, validate, admit capacity,
+// contact a provider, or generate model output. The returned PromptInspection
+// and its input slice are caller-owned. Inspection does not reload configuration
+// or content files; construct another Engine to observe source changes.
+//
+// A nil Engine returns an error matching ErrInvalidConfig. A blank prompt ID
+// matches ErrInvalidRequest. An absent exact ID or version matches
+// ErrPromptNotFound and not ErrPromptLoad. An ambiguous omitted version matches
+// ErrPromptLoad. Invalid definitions and resource failures reject construction
+// instead. Cancellation during lookup preserves the context error. InspectPrompt
+// returns no partial result on error.
+func (e *Engine) InspectPrompt(
+	ctx context.Context,
+	promptID string,
+	promptVersion string,
+) (*PromptInspection, error) {
+	if e == nil || e.internal == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
 	}
-	dir := filepath.Dir(cleanName)
-	base := filepath.Base(cleanName)
-	if base == "." || base == string(filepath.Separator) || strings.TrimSpace(base) == "" {
-		return nil, "", ErrInvalidConfig
-	}
-	info, err := os.Stat(cleanName)
+
+	inspection, err := e.internal.InspectPrompt(ctx, promptID, promptVersion)
 	if err != nil {
-		return nil, "", fmt.Errorf("%w: failed to access source file %q: %v", ErrInvalidConfig, cleanName, err)
+		return nil, mapPublicError(err)
 	}
-	if info.IsDir() {
-		return nil, "", fmt.Errorf("%w: source path %q must be a file", ErrInvalidConfig, cleanName)
+	return fromDomainPromptInspection(inspection), nil
+}
+
+// InspectProfile resolves one explicit profile without selecting a prompt or
+// starting execution work.
+//
+// InspectProfile looks up a nonblank profileID exactly and case-sensitively
+// in the captured effective configuration. It applies the framework timeout
+// baseline, selected
+// backend, and then selected profile to EffectiveModelParams without a request
+// override. BackendID is empty for an endpoint-only profile.
+//
+// APIKeyEnv in the returned target is an environment-variable name, never its
+// value. APIKeyRequired instead reports a direct credential requirement and is
+// mutually exclusive with a nonblank APIKeyEnv. InspectProfile neither derives
+// an ID from a prompt default_profile nor checks credential availability, so an
+// absent or blank named environment variable is not an error.
+//
+// The returned ProfileInspection and all nested mutable values are
+// caller-owned. Inspection does not reload configuration; construct another
+// Engine to observe source changes. This method does not load
+// a prompt, render, read artifacts or schemas, admit backend capacity, contact
+// a provider, or generate model output.
+//
+// A nil Engine returns an error matching ErrInvalidConfig. A blank profile ID
+// matches ErrInvalidRequest. An absent exact ID matches ErrProfileNotFound and
+// not ErrProfileLoad. Invalid definitions or unknown backends reject
+// construction instead. Cancellation preserves the context error.
+// InspectProfile returns no partial result on error.
+func (e *Engine) InspectProfile(ctx context.Context, profileID string) (*ProfileInspection, error) {
+	if e == nil || e.internal == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
 	}
-	return os.DirFS(dir), filepath.ToSlash(base), nil
+
+	inspection, err := e.internal.InspectProfile(ctx, profileID)
+	if err != nil {
+		return nil, mapPublicError(err)
+	}
+	return fromDomainProfileInspection(inspection), nil
 }
 
 // Prepare resolves and renders a prompt request without calling an LLM.
+// It appends any validated RunRequest.AppendedMessages after rendered
+// definition messages in the returned caller-owned snapshot.
 //
 // Prepare selects the prompt and profile, resolves any selected backend and
 // effective execution settings, resolves the output contract, loads and hashes
@@ -439,44 +344,36 @@ func fileSource(name string) (fs.FS, string, error) {
 // applicable. Cancellation is passed to the active collaborator and is
 // reported in the applicable operation category; no general errors.Is
 // relationship to ctx.Err is promised. Prepare returns no partial result on
-// error.
+// error. It validates RunRequest.TransientRetries but makes no model attempts.
 func (e *Engine) Prepare(ctx context.Context, req RunRequest) (*PreparedRun, error) {
-	if e == nil || e.runner == nil {
-		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
-	}
-
-	domainReq, err := toDomainRunRequest(req)
+	prepared, err := e.prepareRequest(ctx, req)
 	if err != nil {
-		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
-	}
-
-	prepared, err := e.runner.Prepare(ctx, domainReq)
-	if err != nil {
-		return nil, mapPublicError(err)
+		return nil, err
 	}
 	return fromDomainPreparedRun(prepared), nil
 }
 
-// Run prepares a request, invokes the configured LLMClient, and validates the
-// generated output.
+// ValidateRequest reports whether req passes the same preparation path used by
+// [Engine.Prepare] at the time of this call. A nil error means this concrete
+// request passed preparation. It checks only this invocation and returns errors
+// as encountered rather than aggregating them. All supplied inputs follow the
+// same loading behavior as [Engine.Prepare], including inputs unused by the
+// selected template branch. The method does not call an LLM or retain a
+// prepared snapshot or executable handle, but preparation may still perform
+// resource I/O and incur its associated time and memory costs.
 //
-// A content-validation failure is a successful run whose
-// RunResult.Validation has Status ValidationFailed. An inability to perform
-// validation returns an error matching ErrValidation and no partial result.
-// The public Engine does not perform output repair, so validation is
-// single-pass even when OutputContract.RepairAttempts is positive.
-//
-// Run can return every error category documented by [Engine.Prepare], plus
-// ErrCapacityExceeded and ErrLLMGenerate. ErrCapacityExceeded identifies
-// rejection before artifacts, schemas, rendering, or model generation because
-// the selected backend's admission capacity is full; it does not match
-// ErrInvalidRequest or ErrLLMGenerate. Errors from injected clients remain
-// available through errors.Is. Cancellation while waiting for model-generation
-// capacity matches both ErrLLMGenerate and the context error. Cancellation
-// otherwise follows the active collaborator's documented behavior. A nil
-// Engine returns ErrInvalidConfig. Run returns no partial result on error.
-func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
-	if e == nil || e.runner == nil {
+// Its errors and context handling follow [Engine.Prepare], including an error
+// matching ErrInvalidConfig when the Engine is nil. Success does not guarantee
+// provider acceptance or later execution success. Use [Engine.PrepareExecution]
+// when execution must use the snapshot produced by preparation.
+func (e *Engine) ValidateRequest(ctx context.Context, req RunRequest) error {
+	_, err := e.prepareRequest(ctx, req)
+	return err
+}
+
+func (e *Engine) prepareRequest(ctx context.Context, req RunRequest) (*domain.PreparedRun, error) {
+	ctx = observation.WithoutExecutionState(ctx)
+	if e == nil || e.internal == nil {
 		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
 	}
 
@@ -485,9 +382,186 @@ func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
 		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
 	}
 
-	result, err := e.runner.Run(ctx, domainReq)
+	prepared, err := e.internal.Prepare(ctx, domainReq)
+	if err != nil {
+		return nil, mapPublicError(err)
+	}
+	return prepared, nil
+}
+
+// PrepareExecution completely prepares a prompt request without calling the
+// configured LLMClient or reserving backend admission capacity. Validated
+// RunRequest.AppendedMessages are included in the frozen effective messages.
+//
+// The returned opaque handle is bound to this Engine and permits one
+// [Engine.RunPrepared] invocation. Preparation freezes the selected sources,
+// rendered messages, effective settings, inputs, provider structured-output
+// metadata, and validation resources needed by that invocation. The handle
+// retains a direct RunRequest.APIKey only in private execution state;
+// [PreparedExecution.Details] is credential-redacted.
+//
+// The context governs preparation only. Cancellation after this method
+// returns does not invalidate the handle or propagate to RunPrepared.
+// PrepareExecution returns the same error categories as [Engine.Prepare] and
+// returns no handle on error. A nil Engine returns an error matching
+// ErrInvalidConfig. The effective transient retry budget is frozen privately
+// for RunPrepared and is absent from Details. A supplied Observer is retained
+// privately for RunPrepared; preparation itself emits no events.
+func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*PreparedExecution, error) {
+	ctx = observation.WithoutExecutionState(ctx)
+	if e == nil || e.internal == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+
+	domainReq, err := toDomainRunRequest(req)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
+	}
+
+	prepared, err := e.internal.PrepareExecution(ctx, domainReq)
+	if err != nil {
+		return nil, mapPublicError(err)
+	}
+	return &PreparedExecution{internal: prepared}, nil
+}
+
+// Run prepares a request, invokes the configured LLMClient, and validates the
+// generated output. Each call uses the Engine's captured configuration and
+// composes a fresh effective prompt with any validated
+// RunRequest.AppendedMessages.
+//
+// A content-validation failure is a successful run whose
+// RunResult.Validation has Status ValidationFailed. When its output contract
+// has a positive repair budget, a failed eligible validation can make bounded
+// additional model calls and stops at the first valid candidate. Exhaustion
+// returns the final failed validation result with cumulative usage and actual
+// repair attempts. An inability to generate or validate returns an error and
+// no partial result.
+//
+// The built-in client retries eligible provider statuses using the effective
+// Config or request budget. With R additional transient attempts and P repair
+// generations, one run can dispatch at most (R+1)×(P+1) Promptkit-controlled
+// model attempts. Each dispatched attempt receives its own generation and
+// HTTP-client timeout; the caller context governs the full operation.
+// Injected LLMClient implementations are not retried automatically.
+// With Observer, Run emits started before facade request conversion or a nil
+// engine check, and finished on every normal return after public error mapping
+// and resource release. The event RunID matches a successful result. Without
+// Observer, result timing retains its existing convention; observed wall
+// duration can be slightly longer because it spans the facade boundary.
+//
+// Run can return every error category documented by [Engine.Prepare], plus
+// ErrCapacityExceeded and ErrLLMGenerate. Built-in HTTP execution failures
+// additionally match ErrTransport; malformed successful responses and nil
+// successful injected responses additionally match ErrMalformedResponse.
+// An engine admission rejection is
+// discoverable as [CapacityError] and still matches ErrCapacityExceeded. It
+// occurs before artifacts, schemas, rendering, or model generation because the
+// selected backend's admission capacity is full; it does not match
+// ErrInvalidRequest or ErrLLMGenerate. A built-in OpenAI-compatible non-2xx
+// response is discoverable as [GenerationError]. Eligible-status exhaustion
+// with a positive budget is discoverable as [RetryExhaustedError]. Injected
+// errors remain available through errors.Is. Cancellation while waiting for
+// model-generation capacity matches both ErrLLMGenerate and the context error.
+// Cancellation otherwise follows the active collaborator's documented
+// behavior. A nil Engine returns ErrInvalidConfig. Run returns no partial
+// result on error.
+func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
+	ctx = observation.WithoutExecutionState(ctx)
+	if req.Observer == nil {
+		return e.run(ctx, req)
+	}
+	recorder, idErr := observation.New(func(event domain.ExecutionEvent) { req.Observer(fromDomainExecutionEvent(event)) })
+	recorder.Start()
+	if idErr != nil {
+		err := fmt.Errorf("failed to create run id: %w", idErr)
+		terminal := observedFinish(nil, err)
+		recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure})
+		return nil, err
+	}
+	result, err := e.run(observation.WithRecorder(ctx, recorder), req)
+	terminal := observedFinish(result, err)
+	recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure,
+		StatusCode: terminal.StatusCode, RetriesExhausted: terminal.RetriesExhausted})
+	return result, err
+}
+
+func (e *Engine) run(ctx context.Context, req RunRequest) (*RunResult, error) {
+	if e == nil || e.internal == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+
+	domainReq, err := toDomainRunRequest(req)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
+	}
+
+	result, err := e.internal.Run(ctx, domainReq)
 	if err != nil {
 		return nil, mapPublicError(err)
 	}
 	return fromDomainRunResult(result), nil
 }
+
+// RunPrepared atomically claims and executes a handle created by
+// [Engine.PrepareExecution].
+//
+// A valid owning-Engine invocation consumes the handle's one attempt before
+// credential revalidation, backend admission, generation, or validation.
+// Cancellation, capacity rejection, generation failure, operational
+// validation failure, and success all leave the handle unusable. A nil,
+// zero-value, foreign-Engine, discarded, claimed, or used handle returns an
+// error matching ErrInvalidRequest; a nil Engine returns ErrInvalidConfig and
+// does not claim the handle.
+//
+// The supplied context governs this execution attempt independently of the
+// preparation context. It covers credential revalidation, admission,
+// generation, validation, and any bounded output repair. Result timing begins
+// after the claim and excludes preparation and consumer-held delay.
+//
+// RunPrepared can return ErrInvalidRequest, ErrAPIKeyEnvMissing,
+// ErrCapacityExceeded, ErrLLMGenerate, or ErrValidation as applicable while
+// preserving documented collaborator and context identities. An engine
+// admission rejection is discoverable as [CapacityError] and still matches
+// ErrCapacityExceeded. A built-in OpenAI-compatible non-2xx response is
+// discoverable as [GenerationError]. Built-in HTTP execution failures match
+// ErrTransport; malformed successful responses and nil successful injected
+// responses match ErrMalformedResponse. Positive-budget eligible-status
+// exhaustion is discoverable as [RetryExhaustedError]. A completed
+// content-validation rejection, including repair exhaustion, is returned in
+// RunResult, not as an operational
+// error. RunPrepared uses the retry choice frozen during PrepareExecution. An
+// operational error returns no partial RunResult. A retained Observer begins
+// only after a successful atomic claim, with identities from the frozen
+// snapshot, and receives a terminal event on normal return after public error
+// mapping. Invalid, foreign, discarded, or reused handles emit no events.
+// Callbacks run outside the handle and capacity locks; panics propagate and
+// do not promise a terminal event.
+func (e *Engine) RunPrepared(ctx context.Context, prepared *PreparedExecution) (*RunResult, error) {
+	ctx = observation.WithoutExecutionState(ctx)
+	if e == nil || e.internal == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+
+	var internal *usecase.PreparedExecution
+	if prepared != nil {
+		internal = prepared.internal
+	}
+	var recorder *observation.Recorder
+	result, err := e.internal.RunPrepared(observation.WithClaimSink(ctx, &recorder), internal)
+	if err != nil {
+		mapped := mapPublicError(err)
+		if recorder != nil {
+			terminal := observedFinish(nil, mapped)
+			recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure,
+				StatusCode: terminal.StatusCode, RetriesExhausted: terminal.RetriesExhausted})
+		}
+		return nil, mapped
+	}
+	publicResult := fromDomainRunResult(result)
+	if recorder != nil {
+		terminal := observedFinish(publicResult, nil)
+		recorder.Finish(observation.FinishData{Outcome: terminal.Outcome})
+	}
+	return publicResult, nil
+}

errors.go

@@ -3,8 +3,11 @@ package promptkit
 import (
 	"errors"
 	"fmt"
+	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/capacity"
+	"gitea.maximumdirect.net/eric/promptkit/internal/generation"
+	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
 	"gitea.maximumdirect.net/eric/promptkit/internal/promptdef"
 	"gitea.maximumdirect.net/eric/promptkit/internal/usecase"
@@ -14,11 +17,47 @@ func mapPublicError(err error) error {
 	if err == nil {
 		return nil
 	}
+	var internalCapacityError *usecase.CapacityError
+	if errors.As(err, &internalCapacityError) && internalCapacityError != nil &&
+		strings.TrimSpace(internalCapacityError.BackendID) != "" {
+		return &CapacityError{BackendID: internalCapacityError.BackendID}
+	}
+	var internalExhausted *generation.ExhaustedError
+	if errors.As(err, &internalExhausted) && internalExhausted != nil {
+		mapped := newRetryExhaustedError(internalExhausted.Attempts(), mapPublicError(internalExhausted.Unwrap()))
+		if errors.Is(err, usecase.ErrValidation) && !errors.Is(mapped, ErrValidation) {
+			return fmt.Errorf("%w: %w", ErrValidation, mapped)
+		}
+		if publicErr := publicErrorFor(err); publicErr != nil && !errors.Is(mapped, publicErr) {
+			return fmt.Errorf("%w: %w", publicErr, mapped)
+		}
+		return mapped
+	}
 	publicErr := publicErrorFor(err)
+	var providerHTTPError *llm.ProviderHTTPError
+	if errors.As(err, &providerHTTPError) && providerHTTPError != nil {
+		generationErr := newGenerationError(
+			providerHTTPError.StatusCode(),
+			providerHTTPError.ProviderCode(),
+			providerHTTPError.ProviderType(),
+			providerHTTPError.ProviderMessage(),
+		)
+		if publicErr != nil && !errors.Is(publicErr, ErrLLMGenerate) {
+			return fmt.Errorf("%w: %w", publicErr, generationErr)
+		}
+		return generationErr
+	}
 	if publicErr == nil {
 		return err
 	}
-	return fmt.Errorf("%w: %w", publicErr, err)
+	switch {
+	case errors.Is(err, llm.ErrTransport):
+		return fmt.Errorf("%w: %w: %w", publicErr, ErrTransport, err)
+	case errors.Is(err, llm.ErrMalformedResponse):
+		return fmt.Errorf("%w: %w: %w", publicErr, ErrMalformedResponse, err)
+	default:
+		return fmt.Errorf("%w: %w", publicErr, err)
+	}
 }
 
 func publicErrorFor(err error) error {
@@ -33,10 +72,6 @@ func publicErrorFor(err error) error {
 		return ErrPromptLoad
 	case errors.Is(err, usecase.ErrProfileLoad):
 		return ErrProfileLoad
-	case errors.Is(err, promptdef.ErrInvalidYAML), errors.Is(err, promptdef.ErrInvalidPromptDefinition):
-		return ErrPromptLoad
-	case isProfileLoadCause(err):
-		return ErrProfileLoad
 	case errors.Is(err, usecase.ErrAPIKeyEnvMissing):
 		return errors.Join(ErrInvalidRequest, ErrAPIKeyEnvMissing)
 	case errors.Is(err, capacity.ErrCapacityExceeded):
@@ -55,9 +90,3 @@ func publicErrorFor(err error) error {
 		return nil
 	}
 }
-
-func isProfileLoadCause(err error) bool {
-	return errors.Is(err, profile.ErrInvalidYAML) ||
-		errors.Is(err, profile.ErrInvalidProfile) ||
-		errors.Is(err, profile.ErrRawAPIKeyNotAllowed)
-}

errors_internal_test.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"testing"
 
+	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 	"gitea.maximumdirect.net/eric/promptkit/internal/usecase"
 )
 
@@ -20,3 +21,62 @@ func TestMapPublicErrorPreservesGenerationCancellation(t *testing.T) {
 		t.Fatalf("mapped error=%v, want context.Canceled", err)
 	}
 }
+
+func TestMapPublicErrorTranslatesCapacityError(t *testing.T) {
+	internalErr := &usecase.CapacityError{BackendID: "limited"}
+
+	err := mapPublicError(internalErr)
+	var publicErr *CapacityError
+	if !errors.As(err, &publicErr) || publicErr == nil {
+		t.Fatalf("mapped error=%v, want public CapacityError", err)
+	}
+	if publicErr.BackendID != "limited" {
+		t.Fatalf("mapped backend ID=%q, want limited", publicErr.BackendID)
+	}
+	if !errors.Is(err, ErrCapacityExceeded) {
+		t.Fatalf("mapped error=%v, want ErrCapacityExceeded", err)
+	}
+	if errors.Is(err, ErrInvalidRequest) || errors.Is(err, ErrLLMGenerate) {
+		t.Fatalf("mapped capacity error has an unrelated category: %v", err)
+	}
+	var leakedInternalErr *usecase.CapacityError
+	if errors.As(err, &leakedInternalErr) {
+		t.Fatalf("mapped error exposes internal CapacityError: %v", err)
+	}
+
+	internalErr.BackendID = "changed"
+	if publicErr.BackendID != "limited" {
+		t.Fatalf("mapped backend ID changed with source error: %q", publicErr.BackendID)
+	}
+}
+
+func TestMapPublicErrorPreservesValidationAroundGenerationError(t *testing.T) {
+	internalErr := fmt.Errorf(
+		"%w: %w",
+		usecase.ErrValidation,
+		&llm.ProviderHTTPError{},
+	)
+
+	err := mapPublicError(internalErr)
+	if !errors.Is(err, ErrValidation) {
+		t.Fatalf("mapped error=%v, want ErrValidation", err)
+	}
+	if !errors.Is(err, ErrLLMGenerate) {
+		t.Fatalf("mapped error=%v, want ErrLLMGenerate", err)
+	}
+	var generationErr *GenerationError
+	if !errors.As(err, &generationErr) || generationErr == nil {
+		t.Fatalf("mapped error=%v, want GenerationError", err)
+	}
+	var leakedInternalErr *llm.ProviderHTTPError
+	if errors.As(err, &leakedInternalErr) {
+		t.Fatalf("mapped error exposes internal ProviderHTTPError: %v", err)
+	}
+}
+
+func TestMapPublicErrorDoesNotTreatRequestEncodingAsTransport(t *testing.T) {
+	err := mapPublicError(fmt.Errorf("%w: %w", usecase.ErrLLMGenerate, llm.ErrRequestFailed))
+	if !errors.Is(err, ErrLLMGenerate) || errors.Is(err, ErrTransport) {
+		t.Fatalf("request encoding category = %v", err)
+	}
+}

examples/go-library/observe/main.go

@@ -0,0 +1,168 @@
+package main
+
+import (
+	"context"
+	"embed"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+//go:embed prompts/*.yaml schemas/*.json
+var assets embed.FS
+
+type scriptedTransport struct {
+	proposalCalls int
+}
+
+func (t *scriptedTransport) RoundTrip(req *http.Request) (*http.Response, error) {
+	if err := req.Context().Err(); err != nil {
+		return nil, err
+	}
+	var request struct {
+		Model string `json:"model"`
+	}
+	if err := json.NewDecoder(req.Body).Decode(&request); err != nil {
+		return nil, err
+	}
+	switch request.Model {
+	case "proposal-model":
+		t.proposalCalls++
+		if t.proposalCalls == 1 {
+			return httpResponse(http.StatusServiceUnavailable, `{"error":{}}`), nil
+		}
+		return modelResponse(`{"title":"Queue cleanup","risk":"low"}`, 10, 8), nil
+	case "assessment-model":
+		return modelResponse(`{"recommendation":"accept","confidence":0.9}`, 7, 5), nil
+	default:
+		return nil, fmt.Errorf("unexpected model in offline transport")
+	}
+}
+
+func modelResponse(content string, promptTokens, completionTokens int) *http.Response {
+	body, _ := json.Marshal(map[string]any{
+		"choices": []any{map[string]any{"message": map[string]string{"content": content}}},
+		"usage":   map[string]int{"prompt_tokens": promptTokens, "completion_tokens": completionTokens},
+	})
+	return httpResponse(http.StatusOK, string(body))
+}
+
+func httpResponse(status int, body string) *http.Response {
+	return &http.Response{StatusCode: status, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(body))}
+}
+
+type eventLog struct{ events []promptkit.ExecutionEvent }
+
+func (l *eventLog) record(event promptkit.ExecutionEvent) { l.events = append(l.events, event) }
+
+func (l *eventLog) terminal() (promptkit.ExecutionEvent, error) {
+	if len(l.events) == 0 || l.events[len(l.events)-1].Kind != promptkit.ExecutionFinished {
+		return promptkit.ExecutionEvent{}, fmt.Errorf("execution did not finish")
+	}
+	return l.events[len(l.events)-1], nil
+}
+
+type summary struct {
+	ProposalAttempts           uint64 `json:"proposal_attempts"`
+	ProposalTransientRetries   uint64 `json:"proposal_transient_retries"`
+	ProposalUsageComplete      bool   `json:"proposal_usage_complete"`
+	AssessmentAttempts         uint64 `json:"assessment_attempts"`
+	AssessmentTransientRetries uint64 `json:"assessment_transient_retries"`
+	AssessmentOutcome          string `json:"assessment_outcome"`
+}
+
+func run() (summary, error) {
+	transport := &scriptedTransport{}
+	inputs, err := promptkit.FilesystemInputs("example prompts", assets, "prompts")
+	if err != nil {
+		return summary{}, err
+	}
+	backendID, proposalModel, assessmentModel := "offline-backend", "proposal-model", "assessment-model"
+	inputs = append(inputs,
+		configkit.Input{Source: promptkit.BackendSource("example backend", map[string]promptkit.BackendConfig{
+			backendID: promptkit.LocalBackend("https://example.invalid/v1", 1),
+		})},
+		configkit.Input{Source: promptkit.ProfileSource("example profiles", map[string]promptkit.ProfileConfig{
+			"proposal-profile":   {Backend: &backendID, Model: &proposalModel},
+			"assessment-profile": {Backend: &backendID, Model: &assessmentModel},
+		})},
+	)
+	engine, err := promptkit.NewEngine(promptkit.Config{
+		Sources:    promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: inputs}}},
+		HTTPClient: &http.Client{Transport: transport},
+	}, promptkit.WithSchemaFS(assets, "schemas"))
+	if err != nil {
+		return summary{}, err
+	}
+
+	var proposalLog eventLog
+	proposal, err := engine.Run(context.Background(), promptkit.RunRequest{
+		PromptID: "example.proposal", Inputs: map[string]promptkit.ArtifactRef{"idea": promptkit.Inline("Reduce queue cleanup toil.")}, Observer: proposalLog.record,
+	})
+	if err != nil {
+		return summary{}, err
+	}
+	if !proposal.Validation.IsValid {
+		return summary{}, fmt.Errorf("proposal failed structural validation")
+	}
+	proposalTerminal, err := proposalLog.terminal()
+	if err != nil {
+		return summary{}, err
+	}
+	var assessmentLog eventLog
+	assessment, err := engine.Run(context.Background(), promptkit.RunRequest{
+		PromptID: "example.assessment", Inputs: map[string]promptkit.ArtifactRef{"proposal": promptkit.Inline(proposal.RawOutput)}, Observer: assessmentLog.record,
+	})
+	if err != nil {
+		return summary{}, err
+	}
+	if !assessment.Validation.IsValid {
+		return summary{}, fmt.Errorf("assessment failed structural validation")
+	}
+	assessmentTerminal, err := assessmentLog.terminal()
+	if err != nil {
+		return summary{}, err
+	}
+	var proposalValue struct {
+		Risk string `json:"risk"`
+	}
+	var assessmentValue struct {
+		Recommendation string `json:"recommendation"`
+	}
+	if err := json.Unmarshal([]byte(proposal.RawOutput), &proposalValue); err != nil {
+		return summary{}, err
+	}
+	if err := json.Unmarshal([]byte(assessment.RawOutput), &assessmentValue); err != nil {
+		return summary{}, err
+	}
+	decision := "rejected"
+	if proposalValue.Risk == "low" && assessmentValue.Recommendation == "accept" {
+		decision = "accepted"
+	}
+	return summary{
+		ProposalAttempts:           proposalTerminal.Attempts,
+		ProposalTransientRetries:   proposalTerminal.TransientRetries,
+		ProposalUsageComplete:      proposalTerminal.Usage.PromptTokens.Complete,
+		AssessmentAttempts:         assessmentTerminal.Attempts,
+		AssessmentTransientRetries: assessmentTerminal.TransientRetries,
+		AssessmentOutcome:          decision,
+	}, nil
+}
+
+func main() {
+	value, err := run()
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+	if err := json.NewEncoder(os.Stdout).Encode(value); err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+}

examples/go-library/observe/prompts/assessment.yaml

@@ -0,0 +1,18 @@
+prompts:
+  example.assessment:
+    "1.0.0":
+      default_profile: assessment-profile
+      description: Assess a structured proposal independently of schema validity.
+      inputs:
+        - name: proposal
+          required: true
+          content_type: application/json
+      messages:
+        - role: system
+          content: Return a concise assessment as JSON.
+        - role: user
+          content: '{{input "proposal"}}'
+      output:
+        format: json
+        validation_mode: json_schema
+        schema_path: assessment.schema.json

examples/go-library/observe/prompts/proposal.yaml

@@ -0,0 +1,18 @@
+prompts:
+  example.proposal:
+    "1.0.0":
+      default_profile: proposal-profile
+      description: Propose a bounded improvement from a synthetic idea.
+      inputs:
+        - name: idea
+          required: true
+          content_type: text/plain
+      messages:
+        - role: system
+          content: Return a concise proposal as JSON.
+        - role: user
+          content: '{{input "idea"}}'
+      output:
+        format: json
+        validation_mode: json_schema
+        schema_path: proposal.schema.json

examples/go-library/observe/schemas/assessment.schema.json

@@ -0,0 +1,10 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "type": "object",
+  "required": ["recommendation", "confidence"],
+  "properties": {
+    "recommendation": {"type": "string"},
+    "confidence": {"type": "number"}
+  },
+  "additionalProperties": false
+}

examples/go-library/observe/schemas/proposal.schema.json

@@ -0,0 +1,10 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "type": "object",
+  "required": ["title", "risk"],
+  "properties": {
+    "title": {"type": "string"},
+    "risk": {"type": "string"}
+  },
+  "additionalProperties": false
+}

examples/go-library/prepare/main.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"os"
 
+	"gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/promptkit"
 )
 
@@ -18,25 +19,45 @@ type summary struct {
 }
 
 func main() {
-	engine, err := promptkit.NewEngine(
-		promptkit.Config{},
-		promptkit.WithPromptFile("examples/go-library/prepare/prompt.yaml"),
-		promptkit.WithProfiles(promptkit.Profile{
-			ID:       "offline-example",
-			Endpoint: "https://example.invalid/v1",
-			Model:    "offline-model",
-		}),
-	)
+	endpoint, model := "https://example.invalid/v1", "offline-model"
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("example profile", map[string]promptkit.ProfileConfig{
+			"offline-example": {Endpoint: &endpoint, Model: &model},
+		})}}}},
+		Main: &configkit.Input{Source: configkit.File("examples/go-library/prepare/prompt.yaml")},
+	}})
 	if err != nil {
 		exit(err)
 	}
+	view, err := engine.InspectConfiguration()
+	if err != nil {
+		exit(err)
+	}
+	if _, found := view.Explain(configkit.Keys("profiles", "offline-example", "model")); !found {
+		exit(fmt.Errorf("configured profile model was not found"))
+	}
+	requirements, err := engine.InspectRequirements(context.Background(), promptkit.RequirementInspectionRequest{
+		PromptID: "example.prepare",
+	})
+	if err != nil {
+		exit(err)
+	}
+	if !requirements.Complete || len(requirements.Inputs) != 1 ||
+		requirements.Inputs[0].Name != "note" ||
+		requirements.Inputs[0].Kind != promptkit.RequirementRequired {
+		exit(fmt.Errorf("required note input was not discovered before preparation"))
+	}
 
-	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+	request := promptkit.RunRequest{
 		PromptID: "example.prepare",
 		Inputs: map[string]promptkit.ArtifactRef{
 			"note": promptkit.Inline("Ada finished the migration review."),
 		},
-	})
+	}
+	if err := engine.ValidateRequest(context.Background(), request); err != nil {
+		exit(err)
+	}
+	prepared, err := engine.Prepare(context.Background(), request)
 	if err != nil {
 		exit(err)
 	}

examples/go-library/prepare/prompt.yaml

@@ -1,16 +1,17 @@
-id: example.prepare
-version: "1.0.0"
-default_profile: offline-example
-description: Prepare a prompt without contacting a model provider.
-inputs:
-  - name: note
-    required: true
-    content_type: text/plain
-messages:
-  - role: system
-    content: Summarize the note in one sentence.
-  - role: user
-    content: '{{input "note"}}'
-output:
-  format: text
-  validation_mode: basic
+prompts:
+  example.prepare:
+    "1.0.0":
+      default_profile: offline-example
+      description: Prepare a prompt without contacting a model provider.
+      inputs:
+        - name: note
+          required: true
+          content_type: text/plain
+      messages:
+        - role: system
+          content: Summarize the note in one sentence.
+        - role: user
+          content: '{{input "note"}}'
+      output:
+        format: text
+        validation_mode: basic

examples/go-library/run/main.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"os"
 
+	"gitea.maximumdirect.net/eric/ConfigKit"
 	"gitea.maximumdirect.net/eric/promptkit"
 )
 
@@ -38,16 +39,13 @@ type summary struct {
 }
 
 func main() {
-	engine, err := promptkit.NewEngine(
-		promptkit.Config{},
-		promptkit.WithPromptFile("examples/go-library/run/prompt.yaml"),
-		promptkit.WithProfiles(promptkit.Profile{
-			ID:       "offline-example",
-			Endpoint: "https://example.invalid/v1",
-			Model:    "offline-model",
-		}),
-		promptkit.WithLLMClient(deterministicClient{}),
-	)
+	endpoint, model := "https://example.invalid/v1", "offline-model"
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("example profile", map[string]promptkit.ProfileConfig{
+			"offline-example": {Endpoint: &endpoint, Model: &model},
+		})}}}},
+		Main: &configkit.Input{Source: configkit.File("examples/go-library/run/prompt.yaml")},
+	}}, promptkit.WithLLMClient(deterministicClient{}))
 	if err != nil {
 		exit(err)
 	}

examples/go-library/run/prompt.yaml

@@ -1,16 +1,17 @@
-id: example.run
-version: "1.0.0"
-default_profile: offline-example
-description: Run a prompt with a deterministic injected model client.
-inputs:
-  - name: note
-    required: true
-    content_type: text/plain
-messages:
-  - role: system
-    content: Summarize the note in one sentence.
-  - role: user
-    content: '{{input "note"}}'
-output:
-  format: text
-  validation_mode: basic
+prompts:
+  example.run:
+    "1.0.0":
+      default_profile: offline-example
+      description: Run a prompt with a deterministic injected model client.
+      inputs:
+        - name: note
+          required: true
+          content_type: text/plain
+      messages:
+        - role: system
+          content: Summarize the note in one sentence.
+        - role: user
+          content: '{{input "note"}}'
+      output:
+        format: text
+        validation_mode: basic

formatting.go

@@ -18,7 +18,7 @@ func (r RunRequest) GoString() string {
 
 func (r RunRequest) redactedString() string {
 	return fmt.Sprintf(
-		"promptkit.RunRequest{PromptID:%q PromptVersion:%q ProfileID:%q APIKeySet:%t Inputs:%d Vars:%d ExecutionSet:%t ValidationSet:%t}",
+		"promptkit.RunRequest{PromptID:%q PromptVersion:%q ProfileID:%q APIKeySet:%t Inputs:%d Vars:%d ExecutionSet:%t ValidationSet:%t AppendedMessages:%d}",
 		r.PromptID,
 		r.PromptVersion,
 		r.ProfileID,
@@ -27,6 +27,7 @@ func (r RunRequest) redactedString() string {
 		len(r.Vars),
 		r.Execution != nil,
 		r.Validation != nil,
+		len(r.AppendedMessages),
 	)
 }
 

frozen_rendering_contract_test.go

@@ -0,0 +1,177 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+const capturedPromptDocument = `
+prompts:
+  captured:
+    "1":
+      default_profile: offline
+      session_id: '{{.session}}'
+      messages:
+        - role: user
+          content_file: body.tmpl
+      output:
+        format: text
+        validation_mode: none
+`
+
+func capturedPromptEngine(t *testing.T, source fs.FS) *promptkit.Engine {
+	t.Helper()
+	inputs, err := promptkit.FilesystemInputs("application prompts", source, "config")
+	if err != nil {
+		t.Fatalf("select prompt inputs: %v", err)
+	}
+	endpoint, model := "https://example.invalid/v1", "offline-model"
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{
+			{Inputs: inputs},
+			{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("offline profile", map[string]promptkit.ProfileConfig{
+				"offline": {Endpoint: &endpoint, Model: &model},
+			})}}},
+		},
+	}})
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+	return engine
+}
+
+func TestEngineCapturesPromptContentUntilReconstruction(t *testing.T) {
+	files := fstest.MapFS{
+		"config/prompt.yaml": &fstest.MapFile{Data: []byte(capturedPromptDocument)},
+		"config/body.tmpl":   &fstest.MapFile{Data: []byte("before {{.name}}")},
+	}
+	first := capturedPromptEngine(t, files)
+	files["config/body.tmpl"] = &fstest.MapFile{Data: []byte("after {{.name}}")}
+	request := promptkit.RunRequest{PromptID: "captured", Vars: map[string]string{"name": "Ada", "session": "session-1"}}
+	oldPrepared, err := first.Prepare(context.Background(), request)
+	if err != nil {
+		t.Fatalf("prepare with existing engine: %v", err)
+	}
+	second := capturedPromptEngine(t, files)
+	newPrepared, err := second.Prepare(context.Background(), request)
+	if err != nil {
+		t.Fatalf("prepare with rebuilt engine: %v", err)
+	}
+	if oldPrepared.Messages[0].Content != "before Ada" || newPrepared.Messages[0].Content != "after Ada" || oldPrepared.PromptHash == newPrepared.PromptHash {
+		t.Fatalf("captured content or hash did not follow engine lifetime: old=%+v new=%+v", oldPrepared, newPrepared)
+	}
+}
+
+type unreadableAfterConstructionFS struct {
+	fs.FS
+	closed atomic.Bool
+	reads  atomic.Int64
+}
+
+func (source *unreadableAfterConstructionFS) Open(name string) (fs.File, error) {
+	if source.closed.Load() {
+		source.reads.Add(1)
+		return nil, fs.ErrPermission
+	}
+	return source.FS.Open(name)
+}
+
+func TestCapturedTemplatesBindInputsIndependentlyAcrossConcurrentPreparations(t *testing.T) {
+	source := &unreadableAfterConstructionFS{FS: fstest.MapFS{
+		"config/prompt.yaml": &fstest.MapFile{Data: []byte(capturedPromptDocument)},
+		"config/body.tmpl":   &fstest.MapFile{Data: []byte(`{{.name}}: {{input "note"}}`)},
+	}}
+	engine := capturedPromptEngine(t, source)
+	source.closed.Store(true)
+
+	const calls = 24
+	var wait sync.WaitGroup
+	errs := make(chan error, calls)
+	for index := 0; index < calls; index++ {
+		wait.Add(1)
+		go func(index int) {
+			defer wait.Done()
+			name := fmt.Sprintf("caller-%d", index)
+			body := fmt.Sprintf("body-%d", index)
+			prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+				PromptID: "captured", SessionID: "direct-session",
+				Vars:   map[string]string{"name": name},
+				Inputs: map[string]promptkit.ArtifactRef{"note": promptkit.Inline(body)},
+			})
+			if err == nil && (prepared.Messages[0].Content != name+": "+body || prepared.SessionID != "direct-session") {
+				err = fmt.Errorf("unexpected prepared content: %+v", prepared)
+			}
+			errs <- err
+		}(index)
+	}
+	wait.Wait()
+	close(errs)
+	for err := range errs {
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+	if source.reads.Load() != 0 {
+		t.Fatalf("preparation reopened captured source %d times", source.reads.Load())
+	}
+
+	_, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+		PromptID: "captured", Vars: map[string]string{"name": "Ada"},
+		Inputs: map[string]promptkit.ArtifactRef{"note": promptkit.Inline("body")},
+	})
+	if !errors.Is(err, promptkit.ErrPromptRender) {
+		t.Fatalf("missing session variable = %v, want runtime render error", err)
+	}
+	_, err = engine.Prepare(context.Background(), promptkit.RunRequest{
+		PromptID: "captured", SessionID: "direct-session", Vars: map[string]string{"name": "Ada"},
+	})
+	if !errors.Is(err, promptkit.ErrPromptRender) {
+		t.Fatalf("missing input helper artifact = %v, want runtime render error", err)
+	}
+}
+
+func TestPreparationStillReadsRequestArtifactsAndSchemasAfterConstruction(t *testing.T) {
+	endpoint, model := "https://example.invalid/v1", "offline-model"
+	schemas := fstest.MapFS{}
+	engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
+		Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("profile", map[string]promptkit.ProfileConfig{
+			"offline": {Endpoint: &endpoint, Model: &model},
+		})}}}},
+		Main: &configkit.Input{Source: configkit.Mapping("structured prompt", map[string]any{
+			"prompts": map[string]any{"structured": map[string]any{"1": map[string]any{
+				"default_profile": "offline",
+				"inputs":          []any{map[string]any{"name": "note", "required": true}},
+				"messages":        []any{map[string]any{"role": "user", "content": `{{input "note"}}`}},
+				"output":          map[string]any{"format": "json", "validation_mode": "json_schema", "schema_path": "schema.json"},
+			}}},
+		})},
+	}}, promptkit.WithSchemaFS(schemas, "."))
+	if err != nil {
+		t.Fatalf("construct engine before request resources exist: %v", err)
+	}
+	schemas["schema.json"] = &fstest.MapFile{Data: []byte(`{"$schema":"https://json-schema.org/draft/2020-12/schema","type":"object"}`)}
+	artifactPath := filepath.Join(t.TempDir(), "note.txt")
+	if err := os.WriteFile(artifactPath, []byte("request artifact"), 0o600); err != nil {
+		t.Fatalf("write request artifact: %v", err)
+	}
+	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "structured", Inputs: map[string]promptkit.ArtifactRef{
+		"note": promptkit.File(artifactPath),
+	}})
+	if err != nil {
+		t.Fatalf("prepare with resources supplied after construction: %v", err)
+	}
+	if len(prepared.Messages) != 1 || prepared.Messages[0].Content != "request artifact" || prepared.StructuredOutput == nil || prepared.StructuredOutput.JSONSchema == nil {
+		t.Fatalf("runtime artifact or schema was not prepared: %+v", prepared)
+	}
+}

generation_error.go

@@ -0,0 +1,87 @@
+package promptkit
+
+import "fmt"
+
+// GenerationError reports a non-2xx response from Promptkit's built-in
+// OpenAI-compatible client during [Engine.Run] or [Engine.RunPrepared].
+//
+// Engine-produced values are immutable, caller-owned values. Use errors.Is to
+// match [ErrLLMGenerate] and errors.As with a *GenerationError target to obtain
+// this type. The four provider accessors expose untrusted provider-controlled
+// values that can contain sensitive request or schema fragments. Applications
+// must apply their own disclosure policy before logging, displaying, or
+// returning them to another caller.
+//
+// Accessors, Error, GoString, and Unwrap are safe on a nil receiver and a zero
+// value. Default and Go-syntax formatting deliberately redact provider details.
+// GenerationError has no stable JSON representation.
+type GenerationError struct {
+	statusCode      int
+	providerCode    string
+	providerType    string
+	providerMessage string
+}
+
+func newGenerationError(statusCode int, providerCode, providerType, providerMessage string) *GenerationError {
+	return &GenerationError{
+		statusCode:      statusCode,
+		providerCode:    providerCode,
+		providerType:    providerType,
+		providerMessage: providerMessage,
+	}
+}
+
+// StatusCode returns the received provider HTTP status code, or zero for a nil
+// receiver or zero value.
+func (e *GenerationError) StatusCode() int {
+	if e == nil {
+		return 0
+	}
+	return e.statusCode
+}
+
+// ProviderCode returns the normalized provider error code, if present. Its
+// value is untrusted and may contain sensitive data.
+func (e *GenerationError) ProviderCode() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerCode
+}
+
+// ProviderType returns the normalized provider error type, if present. Its
+// value is untrusted and may contain sensitive data.
+func (e *GenerationError) ProviderType() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerType
+}
+
+// ProviderMessage returns the bounded normalized provider diagnostic, if
+// present. Its value is untrusted and may contain sensitive data.
+func (e *GenerationError) ProviderMessage() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerMessage
+}
+
+// Error returns a redacted diagnostic that is not a parsing contract.
+func (e *GenerationError) Error() string {
+	if e == nil || e.statusCode == 0 {
+		return ErrLLMGenerate.Error()
+	}
+	return fmt.Sprintf("%s: provider returned HTTP status %d", ErrLLMGenerate, e.statusCode)
+}
+
+// GoString returns the same redacted diagnostic as Error.
+func (e *GenerationError) GoString() string {
+	return e.Error()
+}
+
+// Unwrap returns ErrLLMGenerate. It is safe to call on a nil receiver or zero
+// value.
+func (e *GenerationError) Unwrap() error {
+	return ErrLLMGenerate
+}

generation_error_contract_test.go

@@ -0,0 +1,229 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestBuiltInGenerationError(t *testing.T) {
+	const (
+		codeMarker    = "provider-code-marker"
+		typeMarker    = "provider-type-marker"
+		messageMarker = "provider-message-marker"
+	)
+	engine := newBuiltInGenerationErrorEngine(t, http.StatusUnprocessableEntity,
+		`{"error":{"code":"`+codeMarker+`","type":"`+typeMarker+`","message":"`+messageMarker+`"}}`)
+
+	result, err := engine.Run(context.Background(), simpleGenerationRequest())
+	if result != nil {
+		t.Fatalf("Run result = %#v, want nil", result)
+	}
+	assertGenerationError(t, err, http.StatusUnprocessableEntity, codeMarker, typeMarker, messageMarker)
+
+	preparedEngine := newBuiltInGenerationErrorEngine(t, http.StatusServiceUnavailable, `{"error":{}}`)
+	prepared, err := preparedEngine.PrepareExecution(context.Background(), simpleGenerationRequest())
+	if err != nil {
+		t.Fatalf("PrepareExecution: %v", err)
+	}
+	result, err = preparedEngine.RunPrepared(context.Background(), prepared)
+	if result != nil {
+		t.Fatalf("RunPrepared result = %#v, want nil", result)
+	}
+	assertGenerationError(t, err, http.StatusServiceUnavailable, "", "", "")
+}
+
+func TestBuiltInGenerationErrorWithAppendedMessages(t *testing.T) {
+	const messageMarker = "combined-request-provider-marker"
+	engine := newBuiltInGenerationErrorEngine(t, http.StatusUnprocessableEntity,
+		`{"error":{"message":"`+messageMarker+`"}}`)
+	request := simpleGenerationRequest()
+	request.AppendedMessages = []promptkit.RenderedMessage{
+		{Role: promptkit.RoleAssistant, Content: "previous response"},
+		{Role: promptkit.RoleUser, Content: "consumer correction"},
+	}
+
+	result, err := engine.Run(context.Background(), request)
+	if result != nil {
+		t.Fatalf("Run result = %#v, want nil", result)
+	}
+	assertGenerationError(t, err, http.StatusUnprocessableEntity, "", "", messageMarker)
+}
+
+func TestBuiltInRepairGenerationError(t *testing.T) {
+	const (
+		codeMarker    = "repair-code-marker"
+		typeMarker    = "repair-type-marker"
+		messageMarker = "repair-message-marker"
+	)
+	calls := 0
+	config := simpleConfig("message")
+	config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+		calls++
+		if calls == 1 {
+			body := `{"choices":[{"message":{"content":"not-json"}}]}`
+			return &http.Response{StatusCode: http.StatusOK, ContentLength: int64(len(body)), Body: io.NopCloser(strings.NewReader(body))}, nil
+		}
+		body := `{"error":{"code":"` + codeMarker + `","type":"` + typeMarker + `","message":"` + messageMarker + `"}}`
+		return &http.Response{StatusCode: http.StatusUnprocessableEntity, ContentLength: int64(len(body)), Body: io.NopCloser(strings.NewReader(body))}, nil
+	})}
+	engine, err := promptkit.NewEngine(config)
+	if err != nil {
+		t.Fatalf("NewEngine: %v", err)
+	}
+	req := simpleGenerationRequest()
+	req.Validation = &promptkit.OutputContract{
+		Format:         promptkit.FormatJSON,
+		ValidationMode: promptkit.ValidationJSON,
+		RepairAttempts: 1,
+	}
+
+	result, err := engine.Run(context.Background(), req)
+	if result != nil {
+		t.Fatalf("Run result = %#v, want nil", result)
+	}
+	if calls != 2 {
+		t.Fatalf("provider calls = %d, want 2", calls)
+	}
+	assertGenerationError(t, err, http.StatusUnprocessableEntity, codeMarker, typeMarker, messageMarker)
+}
+
+func TestBuiltInGenerationFailureCategories(t *testing.T) {
+	transportCause := errors.New("private transport diagnostic")
+	config := simpleConfig("message")
+	config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+		return nil, transportCause
+	})}
+	engine, err := promptkit.NewEngine(config)
+	if err != nil {
+		t.Fatal(err)
+	}
+	result, err := engine.Run(context.Background(), simpleGenerationRequest())
+	if result != nil || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, promptkit.ErrTransport) || !errors.Is(err, transportCause) {
+		t.Fatalf("transport result=%#v, error=%v", result, err)
+	}
+	if errors.Is(err, promptkit.ErrMalformedResponse) || strings.Contains(fmt.Sprintf("%#v", err), transportCause.Error()) {
+		t.Fatalf("transport category or formatting = %v", err)
+	}
+
+	for _, tc := range []struct {
+		name string
+		body string
+	}{
+		{name: "invalid framing", body: `{"choices":[]}{}`},
+		{name: "missing choices", body: `{"choices":[]}`},
+		{name: "missing content", body: `{"choices":[{"message":{}}]}`},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			engine := newBuiltInGenerationErrorEngine(t, http.StatusOK, tc.body)
+			result, err := engine.Run(context.Background(), simpleGenerationRequest())
+			if result != nil || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, promptkit.ErrMalformedResponse) || errors.Is(err, promptkit.ErrTransport) {
+				t.Fatalf("malformed result=%#v, error=%v", result, err)
+			}
+		})
+	}
+}
+
+func TestBuiltInGenerationPreservesContextIdentity(t *testing.T) {
+	for _, tc := range []struct {
+		name  string
+		cause error
+	}{
+		{name: "canceled", cause: context.Canceled},
+		{name: "deadline", cause: context.DeadlineExceeded},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			config := simpleConfig("message")
+			config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+				return nil, tc.cause
+			})}
+			engine, err := promptkit.NewEngine(config)
+			if err != nil {
+				t.Fatal(err)
+			}
+			result, err := engine.Run(context.Background(), simpleGenerationRequest())
+			if result != nil || !errors.Is(err, tc.cause) || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, promptkit.ErrTransport) {
+				t.Fatalf("context result=%#v, error=%v", result, err)
+			}
+		})
+	}
+}
+
+func TestInjectedGenerationFailureCategories(t *testing.T) {
+	for _, tc := range []struct {
+		name      string
+		client    *fakeLLMClient
+		malformed bool
+	}{
+		{name: "nil response", client: &fakeLLMClient{}, malformed: true},
+		{name: "arbitrary error", client: &fakeLLMClient{err: errors.New("injected failure")}},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			engine := newSimpleEngine(t, "message", promptkit.WithLLMClient(tc.client))
+			result, err := engine.Run(context.Background(), simpleGenerationRequest())
+			if result != nil || !errors.Is(err, promptkit.ErrLLMGenerate) || errors.Is(err, promptkit.ErrTransport) || errors.Is(err, promptkit.ErrMalformedResponse) != tc.malformed {
+				t.Fatalf("injected result=%#v, error=%v", result, err)
+			}
+			if tc.client.err != nil && !errors.Is(err, tc.client.err) {
+				t.Fatalf("injected cause lost: %v", err)
+			}
+		})
+	}
+}
+
+func assertGenerationError(t *testing.T, err error, statusCode int, code, providerType, message string) {
+	t.Helper()
+
+	if !errors.Is(err, promptkit.ErrLLMGenerate) {
+		t.Fatalf("errors.Is(%v, ErrLLMGenerate) = false", err)
+	}
+	var generationErr *promptkit.GenerationError
+	if !errors.As(err, &generationErr) || generationErr == nil {
+		t.Fatalf("error = %T, want *GenerationError", err)
+	}
+	if generationErr.StatusCode() != statusCode || generationErr.ProviderCode() != code || generationErr.ProviderType() != providerType || generationErr.ProviderMessage() != message {
+		t.Fatalf("GenerationError = %#v", generationErr)
+	}
+
+	wantFormatted := fmt.Sprintf("failed to generate output: provider returned HTTP status %d", statusCode)
+	for _, rendered := range []string{fmt.Sprintf("%v", generationErr), fmt.Sprintf("%+v", generationErr), fmt.Sprintf("%#v", generationErr)} {
+		if rendered != wantFormatted {
+			t.Fatalf("formatted error = %q, want %q", rendered, wantFormatted)
+		}
+		for _, marker := range []string{code, providerType, message} {
+			if marker != "" && strings.Contains(rendered, marker) {
+				t.Fatalf("formatted error exposed provider marker %q: %q", marker, rendered)
+			}
+		}
+	}
+}
+
+func newBuiltInGenerationErrorEngine(t *testing.T, statusCode int, body string) *promptkit.Engine {
+	t.Helper()
+
+	config := simpleConfig("message")
+	noRetries := 0
+	config.TransientRetries = &noRetries
+	config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+		return &http.Response{
+			StatusCode:    statusCode,
+			ContentLength: int64(len(body)),
+			Body:          io.NopCloser(strings.NewReader(body)),
+		}, nil
+	})}
+	engine, err := promptkit.NewEngine(config)
+	if err != nil {
+		t.Fatalf("NewEngine: %v", err)
+	}
+	return engine
+}
+
+func simpleGenerationRequest() promptkit.RunRequest {
+	return promptkit.RunRequest{PromptID: "prompt"}
+}

generation_error_internal_test.go

@@ -0,0 +1,32 @@
+package promptkit
+
+import (
+	"errors"
+	"fmt"
+	"testing"
+)
+
+func TestGenerationErrorNilAndZeroValue(t *testing.T) {
+	var nilError *GenerationError
+	zeroError := &GenerationError{}
+
+	for name, err := range map[string]*GenerationError{
+		"nil":  nilError,
+		"zero": zeroError,
+	} {
+		t.Run(name, func(t *testing.T) {
+			if err.StatusCode() != 0 || err.ProviderCode() != "" || err.ProviderType() != "" || err.ProviderMessage() != "" {
+				t.Fatalf("accessors returned provider details: %#v", err)
+			}
+			if err.Error() != "failed to generate output" || err.GoString() != "failed to generate output" {
+				t.Fatalf("redacted formatting = (%q, %q)", err.Error(), err.GoString())
+			}
+			if fmt.Sprintf("%v", err) != "failed to generate output" || fmt.Sprintf("%#v", err) != "failed to generate output" {
+				t.Fatalf("formatted error = (%q, %q)", fmt.Sprintf("%v", err), fmt.Sprintf("%#v", err))
+			}
+			if !errors.Is(err, ErrLLMGenerate) {
+				t.Fatalf("errors.Is(%v, ErrLLMGenerate) = false", err)
+			}
+		})
+	}
+}

go.mod

@@ -1,10 +1,20 @@
 module gitea.maximumdirect.net/eric/promptkit
 
-go 1.25.5
+go 1.26.0
 
 require (
+	gitea.maximumdirect.net/eric/ConfigKit v0.7.1
+	gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v3 v3.20260930.0
+	gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v3 v3.20260930.0
 	github.com/santhosh-tekuri/jsonschema/v6 v6.0.2
-	gopkg.in/yaml.v3 v3.0.1
 )
 
-require golang.org/x/text v0.14.0 // indirect
+require (
+	gitea.maximumdirect.net/eric/ReleaseKit v0.4.1 // indirect
+	github.com/yuin/goldmark/v2 v2.1.6 // indirect
+	go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect
+	golang.org/x/mod v0.40.0 // indirect
+	golang.org/x/text v0.14.0 // indirect
+)
+
+tool gitea.maximumdirect.net/eric/ReleaseKit/cmd/releasekit

go.sum

@@ -1,10 +1,22 @@
+gitea.maximumdirect.net/eric/ConfigKit v0.7.1 h1:q3HrX4WWCPt1WfMydqFEaAmCf9s6Dq+vknJj0C/mdcI=
+gitea.maximumdirect.net/eric/ConfigKit v0.7.1/go.mod h1:C82sbyKHYMV1OwaD2GJanFrwTfcZfjlW2JsGze5b0kA=
+gitea.maximumdirect.net/eric/ReleaseKit v0.4.1 h1:Qn+mdebYOzCVCEXgCbuGSO/WHmJtItwaQ8e3XwY3qtI=
+gitea.maximumdirect.net/eric/ReleaseKit v0.4.1/go.mod h1:LVodqNQ3OIKv9NBq+fFeyMByd1lHY25YOmmpah/NaNI=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v3 v3.20260930.0 h1:lzfCDbzJQRGzIlRPRprnAV9hVVL5zUQy80VMQ0kbYa0=
+gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v3 v3.20260930.0/go.mod h1:HjS0f+3q0IdHCTEz+VVAEgIwIWJFAZUjuKdAtvtqKdg=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v3 v3.20260930.0 h1:h12ZTRldAmpLAomj0oQRGteFxHLN1LPE6XDl9zJNbm8=
+gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v3 v3.20260930.0/go.mod h1:OEFBeKdjTsLzqJwLmXI2O5gG3r64FktcVJVb3mBrRfg=
 github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI=
 github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 h1:KRzFb2m7YtdldCEkzs6KqmJw4nqEVZGK7IN2kJkjTuQ=
 github.com/santhosh-tekuri/jsonschema/v6 v6.0.2/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU=
+github.com/yuin/goldmark/v2 v2.1.6 h1:igy0q9JcqovoFXFx/XjtU8sHVhcUHavPj2sA3EoKJUE=
+github.com/yuin/goldmark/v2 v2.1.6/go.mod h1:G6M4/qOFtfn01/o14BU1UR2Lo5N3S9Qo7xCuz/sHjGQ=
+go.yaml.in/yaml/v4 v4.0.0-rc.6 h1:1h7H1ohdUh93/FyE4YaDa1Zh64K6VVbjF4K6WUxMtH4=
+go.yaml.in/yaml/v4 v4.0.0-rc.6/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0=
+golang.org/x/mod v0.40.0 h1:hUv+3cXcdRHz08UmSiOob7sadHig73uo5bkXxQ/tvUs=
+golang.org/x/mod v0.40.0/go.mod h1:0/weTWkPWGBikyTWAX3dkjVztMmBA5hM0DH6BElSupE=
 golang.org/x/text v0.14.0 h1:ScX5w1eTa3QqT8oi6+ziP7dTV1S2+ALU0bI+0zXKWiQ=
 golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU=
-gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
-gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

internal/artifact/reader.go

@@ -16,10 +16,13 @@ import (
 
 var (
 	ErrUnsupportedRefType = errors.New("unsupported artifact reference type")
-	ErrMissingInlineBody  = errors.New("missing body for inline artifact")
 	ErrMissingFilePath    = errors.New("missing file path for file artifact")
+	ErrUnsupportedFile    = errors.New("file artifact path is not a regular file")
 )
 
+const fileReadChunkSize = 64 * 1024
+const fileReadInitialCapacityLimit = 1 << 20
+
 // Reader resolves artifact references into actual artifacts.
 type Reader interface {
 	Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error)
@@ -34,7 +37,7 @@ type CompositeReader struct {
 func NewCompositeReader() Reader {
 	return &CompositeReader{
 		inlineReader: &inlineReader{},
-		fileReader:   &fileReader{},
+		fileReader:   &fileReader{open: openArtifactFile},
 	}
 }
 
@@ -64,10 +67,6 @@ func (r *inlineReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domai
 	default:
 	}
 
-	if ref.Body == "" {
-		return nil, ErrMissingInlineBody
-	}
-
 	body := []byte(ref.Body)
 	return &domain.Artifact{
 		ContentType: defaults.ContentTypeTextPlain,
@@ -78,7 +77,15 @@ func (r *inlineReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domai
 	}, nil
 }
 
-type fileReader struct{}
+type artifactFile interface {
+	Read([]byte) (int, error)
+	Stat() (os.FileInfo, error)
+	Close() error
+}
+
+type fileReader struct {
+	open func(string) (artifactFile, error)
+}
 
 func (r *fileReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error) {
 	select {
@@ -91,25 +98,82 @@ func (r *fileReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.
 		return nil, ErrMissingFilePath
 	}
 
-	return readFileArtifact(ref.URI)
+	return readFileArtifact(ctx, ref.URI, r.open)
 }
 
-func readFileArtifact(path string) (*domain.Artifact, error) {
-	file, err := os.Open(path)
+func openArtifactFile(path string) (artifactFile, error) {
+	return os.Open(path)
+}
+
+func readFileArtifact(ctx context.Context, path string, open func(string) (artifactFile, error)) (*domain.Artifact, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	info, err := os.Stat(path)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+	}
+	if !info.Mode().IsRegular() {
+		return nil, fmt.Errorf("%w: %s", ErrUnsupportedFile, path)
+	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+
+	file, err := open(path)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
 	}
 	defer file.Close()
 
-	data, err := io.ReadAll(file)
+	openedInfo, err := file.Stat()
 	if err != nil {
-		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+		return nil, fmt.Errorf("failed to inspect opened file %s: %w", path, err)
+	}
+	if !openedInfo.Mode().IsRegular() {
+		return nil, fmt.Errorf("%w: %s", ErrUnsupportedFile, path)
+	}
+
+	capacity := 0
+	if size := openedInfo.Size(); size > 0 {
+		capacity = int(min(size, int64(fileReadInitialCapacityLimit)))
+	}
+	data := make([]byte, 0, capacity)
+	chunk := make([]byte, fileReadChunkSize)
+	for {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		n, readErr := file.Read(chunk)
+		if n > 0 {
+			data = append(data, chunk[:n]...)
+		}
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		if errors.Is(readErr, io.EOF) {
+			break
+		}
+		if readErr != nil {
+			return nil, fmt.Errorf("failed to read file %s: %w", path, readErr)
+		}
+	}
+	if cap(data)-len(data) > len(data) {
+		// A stale size hint should not make a small completed artifact retain a
+		// disproportionately large backing array.
+		compact := make([]byte, len(data))
+		copy(compact, data)
+		data = compact
 	}
 
 	contentType := mime.TypeByExtension(filepath.Ext(path))
 	if contentType == "" {
 		contentType = defaults.ContentTypeTextPlain
 	}
+	hash := fmt.Sprintf("%x", sha256.Sum256(data))
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
 
 	return &domain.Artifact{
 		Name:        filepath.Base(path),
@@ -117,6 +181,6 @@ func readFileArtifact(path string) (*domain.Artifact, error) {
 		Body:        data,
 		URI:         path,
 		Size:        int64(len(data)),
-		Hash:        fmt.Sprintf("%x", sha256.Sum256(data)),
+		Hash:        hash,
 	}, nil
 }

internal/artifact/reader_benchmark_test.go

@@ -0,0 +1,180 @@
+package artifact
+
+import (
+	"bytes"
+	"context"
+	"crypto/sha256"
+	"fmt"
+	"mime"
+	"os"
+	"path/filepath"
+	"runtime"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+var artifactFileReadBenchmarkSink *domain.Artifact
+
+func BenchmarkArtifactFileRead(b *testing.B) {
+	for _, fixture := range makeArtifactReadBenchmarkFiles(b) {
+		b.Run(fixture.name, func(b *testing.B) {
+			reader := &fileReader{open: openArtifactFile}
+			ref := domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: fixture.path}
+			checkArtifactFileRead(b, reader, ref, fixture.body)
+
+			b.ReportAllocs()
+			b.SetBytes(int64(len(fixture.body)))
+			b.ResetTimer()
+			for range b.N {
+				artifact, err := reader.Read(context.Background(), ref)
+				if err != nil {
+					b.Fatalf("read artifact: %v", err)
+				}
+				artifactFileReadBenchmarkSink = artifact
+			}
+			b.StopTimer()
+			runtime.KeepAlive(artifactFileReadBenchmarkSink)
+		})
+	}
+}
+
+// BenchmarkRetainedArtifactFileRead measures live artifact bodies when an
+// opened file reports a size larger than the content that reaches EOF.
+func BenchmarkRetainedArtifactFileRead(b *testing.B) {
+	for _, fixture := range []struct {
+		name string
+		body []byte
+	}{
+		{name: "small", body: bytes.Repeat([]byte("x"), 256<<10)},
+		{name: "empty", body: []byte{}},
+	} {
+		b.Run(fixture.name, func(b *testing.B) {
+			benchmarkRetainedArtifactFileRead(b, fixture.body)
+		})
+	}
+}
+
+func benchmarkRetainedArtifactFileRead(b *testing.B, body []byte) {
+	path := filepath.Join(b.TempDir(), "overreported.txt")
+	if err := os.WriteFile(path, body, 0o600); err != nil {
+		b.Fatal(err)
+	}
+	mime.TypeByExtension(filepath.Ext(path))
+
+	reader := &fileReader{open: func(path string) (artifactFile, error) {
+		file, err := os.Open(path)
+		if err != nil {
+			return nil, err
+		}
+		return &sizeReportingArtifactFile{artifactFile: file, reportedSize: 16 << 20}, nil
+	}}
+	ref := domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: path}
+	checkArtifactFileRead(b, reader, ref, body)
+
+	const retainedBatchSize = 4
+	retained := make([]*domain.Artifact, retainedBatchSize)
+	artifactFileReadBenchmarkSink = nil
+	runtime.GC()
+	var setupMemory runtime.MemStats
+	runtime.ReadMemStats(&setupMemory)
+	runtime.KeepAlive(reader)
+	runtime.KeepAlive(ref)
+
+	b.ReportAllocs()
+	b.SetBytes(int64(len(body)))
+	b.ResetTimer()
+	for iteration := 0; iteration < b.N; iteration++ {
+		artifact, err := reader.Read(context.Background(), ref)
+		if err != nil {
+			b.Fatalf("read retained artifact: %v", err)
+		}
+		retained[iteration%len(retained)] = artifact
+	}
+	b.StopTimer()
+	for i, artifact := range retained {
+		if artifact != nil {
+			continue
+		}
+		artifact, err := reader.Read(context.Background(), ref)
+		if err != nil {
+			b.Fatalf("fill retained artifact batch: %v", err)
+		}
+		retained[i] = artifact
+	}
+	runtime.GC()
+	var retainedMemory runtime.MemStats
+	runtime.ReadMemStats(&retainedMemory)
+	b.ReportMetric(float64(int64(retainedMemory.HeapAlloc)-int64(setupMemory.HeapAlloc)), "retained-B")
+	runtime.KeepAlive(retained)
+	runtime.KeepAlive(reader)
+	runtime.KeepAlive(ref)
+}
+
+type artifactReadBenchmarkFile struct {
+	name string
+	path string
+	body []byte
+}
+
+func makeArtifactReadBenchmarkFiles(b *testing.B) []artifactReadBenchmarkFile {
+	b.Helper()
+	fixtures := []struct {
+		name string
+		size int
+	}{
+		{name: "small_4KiB", size: 4 << 10},
+		{name: "1MiB", size: 1 << 20},
+		{name: "large_4MiB", size: 4 << 20},
+	}
+	results := make([]artifactReadBenchmarkFile, 0, len(fixtures))
+	for _, fixture := range fixtures {
+		body := bytes.Repeat([]byte("x"), fixture.size)
+		path := filepath.Join(b.TempDir(), fixture.name+".txt")
+		if err := os.WriteFile(path, body, 0o600); err != nil {
+			b.Fatal(err)
+		}
+		mime.TypeByExtension(filepath.Ext(path))
+		results = append(results, artifactReadBenchmarkFile{
+			name: fixture.name,
+			path: path,
+			body: body,
+		})
+	}
+	return results
+}
+
+func checkArtifactFileRead(b *testing.B, reader Reader, ref domain.ArtifactRef, want []byte) {
+	b.Helper()
+	artifact, err := reader.Read(context.Background(), ref)
+	if err != nil {
+		b.Fatalf("check artifact read: %v", err)
+	}
+	if !bytes.Equal(artifact.Body, want) || artifact.Size != int64(len(want)) {
+		b.Fatalf("artifact size=%d body length=%d, want %d", artifact.Size, len(artifact.Body), len(want))
+	}
+	wantHash := fmt.Sprintf("%x", sha256.Sum256(want))
+	if artifact.Hash != wantHash {
+		b.Fatalf("artifact hash=%q, want %q", artifact.Hash, wantHash)
+	}
+}
+
+type sizeReportingArtifactFile struct {
+	artifactFile
+	reportedSize int64
+}
+
+func (file *sizeReportingArtifactFile) Stat() (os.FileInfo, error) {
+	info, err := file.artifactFile.Stat()
+	if err != nil {
+		return nil, err
+	}
+	return reportedArtifactFileInfo{FileInfo: info, size: file.reportedSize}, nil
+}
+
+type reportedArtifactFileInfo struct {
+	os.FileInfo
+	size int64
+}
+
+func (info reportedArtifactFileInfo) Size() int64 { return info.size }

internal/artifact/reader_fifo_linux_test.go

@@ -0,0 +1,43 @@
+//go:build linux
+
+package artifact
+
+import (
+	"context"
+	"errors"
+	"path/filepath"
+	"syscall"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestFileReaderRejectsFIFOBeforeOpen(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "artifact.fifo")
+	if err := syscall.Mkfifo(path, 0o600); err != nil {
+		t.Fatalf("create fifo: %v", err)
+	}
+
+	type result struct {
+		artifact *domain.Artifact
+		err      error
+	}
+	done := make(chan result, 1)
+	go func() {
+		artifact, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{
+			Type: domain.ArtifactRefFile,
+			URI:  path,
+		})
+		done <- result{artifact: artifact, err: err}
+	}()
+
+	select {
+	case got := <-done:
+		if got.artifact != nil || !errors.Is(got.err, ErrUnsupportedFile) {
+			t.Fatalf("artifact=%#v err=%v, want nil/ErrUnsupportedFile", got.artifact, got.err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("FIFO read blocked instead of rejecting the non-regular file")
+	}
+}

internal/artifact/reader_test.go

@@ -1,8 +1,11 @@
 package artifact
 
 import (
+	"bytes"
 	"context"
+	"crypto/sha256"
 	"errors"
+	"fmt"
 	"os"
 	"path/filepath"
 	"strings"
@@ -11,54 +14,97 @@ import (
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 )
 
-func TestCompositeReader_Read(t *testing.T) {
+func TestCompositeReaderRejectsUnsupportedReferences(t *testing.T) {
+	_, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{
+		Type: domain.ArtifactRefType("unsupported"),
+		URI:  "unsupported://bucket/key",
+	})
+	if !errors.Is(err, ErrUnsupportedRefType) {
+		t.Fatalf("expected ErrUnsupportedRefType, got %v", err)
+	}
+}
+
+func TestCompositeReaderSourceParityAndOpaqueHashes(t *testing.T) {
 	reader := NewCompositeReader()
-	ctx := context.Background()
+	hashes := make(map[string]string)
+	tests := []struct {
+		name    string
+		content string
+	}{
+		{name: "empty", content: ""},
+		{name: "ordinary", content: "same content"},
+		{name: "changed", content: "changed content"},
+	}
 
-	t.Run("inline artifact", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefInline,
-			Body: "hello world",
-		}
-		art, err := reader.Read(ctx, ref)
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if string(art.Body) != "hello world" {
-			t.Errorf("expected 'hello world', got %s", string(art.Body))
-		}
-		if art.ContentType != "text/plain" {
-			t.Errorf("expected text/plain content type, got %q", art.ContentType)
-		}
-		if art.Hash != "b94d27b9934d3e08a52e52d7da7dabfac484efe37a5380ee9088f7ace2efcde9" {
-			t.Errorf("unexpected hash: %s", art.Hash)
-		}
-		if art.Size != int64(len(ref.Body)) {
-			t.Errorf("expected size %d, got %d", len(ref.Body), art.Size)
-		}
-	})
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			filePath := filepath.Join(t.TempDir(), "artifact.txt")
+			if err := os.WriteFile(filePath, []byte(tc.content), 0o600); err != nil {
+				t.Fatal(err)
+			}
 
-	t.Run("inline artifact missing body", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefInline,
-			Body: "",
-		}
-		_, err := reader.Read(ctx, ref)
-		if !errors.Is(err, ErrMissingInlineBody) {
-			t.Errorf("expected ErrMissingInlineBody, got %v", err)
-		}
-	})
+			sources := []struct {
+				name    string
+				ref     domain.ArtifactRef
+				wantURI string
+			}{
+				{
+					name: "inline",
+					ref:  domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: tc.content},
+				},
+				{
+					name:    "inline with uri",
+					ref:     domain.ArtifactRef{Type: domain.ArtifactRefInline, URI: "memory://input", Body: tc.content},
+					wantURI: "memory://input",
+				},
+				{
+					name:    "file",
+					ref:     domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filePath},
+					wantURI: filePath,
+				},
+			}
 
-	t.Run("unsupported ref type", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefType("unsupported"),
-			URI:  "unsupported://bucket/key",
-		}
-		_, err := reader.Read(ctx, ref)
-		if !errors.Is(err, ErrUnsupportedRefType) {
-			t.Error("expected error for unsupported type")
-		}
-	})
+			var sourceHash string
+			for _, source := range sources {
+				t.Run(source.name, func(t *testing.T) {
+					first, err := reader.Read(context.Background(), source.ref)
+					if err != nil {
+						t.Fatalf("first read: %v", err)
+					}
+					second, err := reader.Read(context.Background(), source.ref)
+					if err != nil {
+						t.Fatalf("second read: %v", err)
+					}
+					if string(first.Body) != tc.content || first.Size != int64(len(tc.content)) {
+						t.Fatalf("body=%q size=%d, want %q/%d", first.Body, first.Size, tc.content, len(tc.content))
+					}
+					if first.URI != source.wantURI {
+						t.Fatalf("URI = %q, want %q", first.URI, source.wantURI)
+					}
+					if first.Hash == "" || first.Hash != second.Hash {
+						t.Fatalf("hashes are not non-empty and stable: %q/%q", first.Hash, second.Hash)
+					}
+					if sourceHash == "" {
+						sourceHash = first.Hash
+					} else if first.Hash != sourceHash {
+						t.Fatalf("equal content hashes differ: %q/%q", sourceHash, first.Hash)
+					}
+					if source.ref.Type == domain.ArtifactRefFile {
+						if first.Name != filepath.Base(filePath) || !strings.HasPrefix(first.ContentType, "text/plain") {
+							t.Fatalf("unexpected file metadata: %+v", first)
+						}
+					} else if first.ContentType != "text/plain" {
+						t.Fatalf("inline content type = %q", first.ContentType)
+					}
+				})
+			}
+			hashes[tc.name] = sourceHash
+		})
+	}
+
+	if hashes["empty"] == hashes["ordinary"] || hashes["ordinary"] == hashes["changed"] {
+		t.Fatalf("changed content did not change opaque hash: %#v", hashes)
+	}
 }
 
 func TestCompositeReaderCopiesInlineData(t *testing.T) {
@@ -87,94 +133,219 @@ func TestCompositeReaderCopiesInlineData(t *testing.T) {
 	}
 }
 
-func TestCompositeReaderHonorsCancellation(t *testing.T) {
-	ctx, cancel := context.WithCancel(context.Background())
-	cancel()
+func TestCompositeReaderHonorsPreCancellation(t *testing.T) {
+	filePath := filepath.Join(t.TempDir(), "artifact.txt")
+	if err := os.WriteFile(filePath, []byte("ignored"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	tests := []struct {
+		name string
+		ref  domain.ArtifactRef
+	}{
+		{name: "inline", ref: domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: "ignored"}},
+		{name: "file", ref: domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filePath}},
+	}
 
-	_, err := NewCompositeReader().Read(ctx, domain.ArtifactRef{
-		Type: domain.ArtifactRefInline,
-		Body: "ignored",
-	})
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("expected context cancellation, got %v", err)
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			ctx, cancel := context.WithCancel(context.Background())
+			cancel()
+
+			artifact, err := NewCompositeReader().Read(ctx, tc.ref)
+			if artifact != nil || !errors.Is(err, context.Canceled) {
+				t.Fatalf("artifact=%#v err=%v, want nil/context.Canceled", artifact, err)
+			}
+		})
 	}
 }
 
-func TestFileReader_Read(t *testing.T) {
-	content := []byte("test file content")
-	filePath := filepath.Join(t.TempDir(), "artifact.txt")
-	if err := os.WriteFile(filePath, content, 0o600); err != nil {
-		t.Fatal(err)
-	}
-
+func TestFileReaderFailuresAndMetadata(t *testing.T) {
 	reader := NewCompositeReader()
-	ctx := context.Background()
-
-	t.Run("file artifact loading", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefFile,
-			URI:  filePath,
-		}
-		art, err := reader.Read(ctx, ref)
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if string(art.Body) != string(content) {
-			t.Errorf("expected %s, got %s", string(content), string(art.Body))
-		}
-		if art.Name != filepath.Base(filePath) {
-			t.Errorf("expected name %q, got %q", filepath.Base(filePath), art.Name)
-		}
-		if !strings.HasPrefix(art.ContentType, "text/plain") {
-			t.Errorf("expected text content type, got %q", art.ContentType)
-		}
-		if art.URI != filePath {
-			t.Errorf("expected URI %q, got %q", filePath, art.URI)
-		}
-		if art.Size != int64(len(content)) {
-			t.Errorf("expected size %d, got %d", len(content), art.Size)
-		}
-		if art.Hash != "60f5237ed4049f0382661ef009d2bc42e48c3ceb3edb6600f7024e7ab3b838f3" {
-			t.Errorf("unexpected hash: %s", art.Hash)
-		}
-	})
 
 	t.Run("missing file path", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefFile,
-			URI:  "",
-		}
-		_, err := reader.Read(ctx, ref)
+		_, err := reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefFile})
 		if !errors.Is(err, ErrMissingFilePath) {
-			t.Errorf("expected ErrMissingFilePath, got %v", err)
+			t.Fatalf("expected ErrMissingFilePath, got %v", err)
 		}
 	})
 
 	t.Run("missing file", func(t *testing.T) {
-		ref := domain.ArtifactRef{
+		_, err := reader.Read(context.Background(), domain.ArtifactRef{
 			Type: domain.ArtifactRefFile,
 			URI:  filepath.Join(t.TempDir(), "missing.txt"),
-		}
-		if _, err := reader.Read(ctx, ref); err == nil {
+		})
+		if err == nil {
 			t.Fatal("expected missing file error")
 		}
 	})
 
+	t.Run("directory rejected before open", func(t *testing.T) {
+		artifact, err := reader.Read(context.Background(), domain.ArtifactRef{
+			Type: domain.ArtifactRefFile,
+			URI:  t.TempDir(),
+		})
+		if artifact != nil || !errors.Is(err, ErrUnsupportedFile) {
+			t.Fatalf("artifact=%#v err=%v, want nil/ErrUnsupportedFile", artifact, err)
+		}
+	})
+
+	t.Run("non-regular opened target rejected", func(t *testing.T) {
+		filePath := filepath.Join(t.TempDir(), "artifact.txt")
+		if err := os.WriteFile(filePath, []byte("content"), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		directoryInfo, err := os.Stat(t.TempDir())
+		if err != nil {
+			t.Fatal(err)
+		}
+		fileReader := &fileReader{open: func(path string) (artifactFile, error) {
+			file, err := os.Open(path)
+			if err != nil {
+				return nil, err
+			}
+			return &reportedInfoFile{artifactFile: file, info: directoryInfo}, nil
+		}}
+
+		artifact, err := fileReader.Read(context.Background(), domain.ArtifactRef{
+			Type: domain.ArtifactRefFile,
+			URI:  filePath,
+		})
+		if artifact != nil || !errors.Is(err, ErrUnsupportedFile) {
+			t.Fatalf("artifact=%#v err=%v, want nil/ErrUnsupportedFile", artifact, err)
+		}
+	})
+
 	t.Run("unknown extension uses text fallback", func(t *testing.T) {
-		path := filepath.Join(t.TempDir(), "artifact.unknownextension")
-		if err := os.WriteFile(path, content, 0o600); err != nil {
+		filePath := filepath.Join(t.TempDir(), "artifact.unknownextension")
+		if err := os.WriteFile(filePath, []byte("content"), 0o600); err != nil {
 			t.Fatal(err)
 		}
 
-		art, err := reader.Read(ctx, domain.ArtifactRef{
+		artifact, err := reader.Read(context.Background(), domain.ArtifactRef{
 			Type: domain.ArtifactRefFile,
-			URI:  path,
+			URI:  filePath,
 		})
 		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
+			t.Fatalf("read artifact: %v", err)
 		}
-		if art.ContentType != "text/plain" {
-			t.Errorf("expected text/plain fallback, got %q", art.ContentType)
+		if artifact.ContentType != "text/plain" {
+			t.Fatalf("content type = %q", artifact.ContentType)
 		}
 	})
 }
+
+func TestFileReaderUsesOpenedSizeOnlyAsACapacityHint(t *testing.T) {
+	tests := []struct {
+		name         string
+		body         []byte
+		reportedSize int64
+	}{
+		{name: "overreported size", body: []byte("actual file content"), reportedSize: 128},
+		{name: "overreported size with empty content", body: []byte{}, reportedSize: 16 << 20},
+		{name: "underreported size", body: bytes.Repeat([]byte("x"), fileReadChunkSize+17), reportedSize: 1},
+		{name: "very large hint with small content", body: []byte("small actual content"), reportedSize: int64(^uint64(0) >> 1)},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			filePath := filepath.Join(t.TempDir(), "artifact.txt")
+			if err := os.WriteFile(filePath, tc.body, 0o600); err != nil {
+				t.Fatal(err)
+			}
+			reader := &fileReader{open: func(path string) (artifactFile, error) {
+				file, err := os.Open(path)
+				if err != nil {
+					return nil, err
+				}
+				info, err := file.Stat()
+				if err != nil {
+					file.Close()
+					return nil, err
+				}
+				return &reportedInfoFile{
+					artifactFile: file,
+					info:         reportedFileSizeInfo{FileInfo: info, size: tc.reportedSize},
+				}, nil
+			}}
+
+			artifact, err := reader.Read(context.Background(), domain.ArtifactRef{
+				Type: domain.ArtifactRefFile,
+				URI:  filePath,
+			})
+			if err != nil {
+				t.Fatalf("read artifact: %v", err)
+			}
+			if !bytes.Equal(artifact.Body, tc.body) {
+				t.Fatalf("artifact body length=%d, want %d", len(artifact.Body), len(tc.body))
+			}
+			if artifact.Size != int64(len(tc.body)) {
+				t.Fatalf("artifact size=%d, want %d", artifact.Size, len(tc.body))
+			}
+			if artifact.Name != filepath.Base(filePath) || artifact.URI != filePath || artifact.ContentType == "" {
+				t.Fatalf("unexpected artifact metadata: name=%q URI=%q content type=%q", artifact.Name, artifact.URI, artifact.ContentType)
+			}
+			wantHash := fmt.Sprintf("%x", sha256.Sum256(tc.body))
+			if artifact.Hash != wantHash {
+				t.Fatalf("artifact hash=%q, want %q", artifact.Hash, wantHash)
+			}
+		})
+	}
+}
+
+func TestFileReaderCancelsAfterReadProgress(t *testing.T) {
+	filePath := filepath.Join(t.TempDir(), "artifact.bin")
+	content := bytes.Repeat([]byte("x"), fileReadChunkSize*2)
+	if err := os.WriteFile(filePath, content, 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	var opened *cancelAfterProgressFile
+	reader := &fileReader{open: func(path string) (artifactFile, error) {
+		file, err := os.Open(path)
+		if err != nil {
+			return nil, err
+		}
+		opened = &cancelAfterProgressFile{artifactFile: file, cancel: cancel}
+		return opened, nil
+	}}
+
+	artifact, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filePath})
+	if artifact != nil || !errors.Is(err, context.Canceled) {
+		t.Fatalf("artifact=%#v err=%v, want nil/context.Canceled", artifact, err)
+	}
+	if opened == nil || opened.reads != 1 {
+		t.Fatalf("read count = %v, want one progressing read", opened)
+	}
+}
+
+type reportedInfoFile struct {
+	artifactFile
+	info os.FileInfo
+}
+
+func (f *reportedInfoFile) Stat() (os.FileInfo, error) {
+	return f.info, nil
+}
+
+type reportedFileSizeInfo struct {
+	os.FileInfo
+	size int64
+}
+
+func (info reportedFileSizeInfo) Size() int64 { return info.size }
+
+type cancelAfterProgressFile struct {
+	artifactFile
+	cancel context.CancelFunc
+	reads  int
+}
+
+func (f *cancelAfterProgressFile) Read(buffer []byte) (int, error) {
+	n, err := f.artifactFile.Read(buffer)
+	if n > 0 {
+		f.reads++
+		f.cancel()
+	}
+	return n, err
+}

internal/backend/registry.go

@@ -5,70 +5,40 @@ package backend
 import (
 	"errors"
 	"fmt"
-	"net/url"
-	"regexp"
-	"sort"
-	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
-	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 )
 
 const (
-	// OpenRouterID is the reserved ID of Promptkit's built-in OpenRouter
-	// backend.
+	// OpenRouterID identifies the maintained OpenRouter backend.
 	OpenRouterID = "openrouter"
-
-	openRouterEndpoint  = "https://openrouter.ai/api/v1"
-	openRouterAPIKeyEnv = "OPENROUTER_API_KEY"
-
-	openRouterConcurrencyLimit = 16
-	defaultQueueCapacity       = 1024
+	// RakestrawHomeID identifies the maintained Rakestrawhome backend.
+	RakestrawHomeID = "rakestrawhome"
 )
 
 // ErrBackendNotFound identifies a registry lookup for an unknown backend ID.
 var ErrBackendNotFound = errors.New("backend not found")
 
-var environmentVariableName = regexp.MustCompile(`^[A-Za-z_][A-Za-z0-9_]*$`)
-
 // Registry is an immutable collection of validated backend definitions.
 type Registry struct {
 	backends map[string]domain.Backend
 }
 
-// NewRegistry constructs a registry containing the built-in OpenRouter
-// definition followed by the supplied additions. Every ID must be unique.
-func NewRegistry(additions []domain.Backend) (*Registry, error) {
-	registry := &Registry{
-		backends: make(map[string]domain.Backend, len(additions)+1),
-	}
-
-	definitions := make([]domain.Backend, 0, len(additions)+1)
-	definitions = append(definitions, domain.Backend{
-		ID:               OpenRouterID,
-		Endpoint:         openRouterEndpoint,
-		APIKeyEnv:        openRouterAPIKeyEnv,
-		ConcurrencyLimit: openRouterConcurrencyLimit,
-	})
-	definitions = append(definitions, additions...)
-
-	for _, definition := range definitions {
-		definition.ID = strings.TrimSpace(definition.ID)
-		if definition.ID == "" {
-			return nil, errors.New("backend ID must not be blank")
-		}
-		if _, exists := registry.backends[definition.ID]; exists {
-			return nil, fmt.Errorf("backend ID %q is already registered", definition.ID)
-		}
-
-		normalized, err := normalizeBackend(definition)
+// NewEffectiveRegistry stores already-validated effective backend definitions
+// under their exact configuration keys. Configuration validation and
+// normalization precede this ownership boundary.
+func NewEffectiveRegistry(definitions map[string]domain.Backend) (*Registry, error) {
+	registry := &Registry{backends: make(map[string]domain.Backend, len(definitions))}
+	for id, definition := range definitions {
+		params, err := jsonvalue.CopyMap(definition.ExtraParams)
 		if err != nil {
-			return nil, err
+			return nil, fmt.Errorf("backend %q extra parameters: %w", id, err)
 		}
-		registry.backends[normalized.ID] = normalized
+		definition.ID = id
+		definition.ExtraParams = params
+		registry.backends[id] = definition
 	}
-
 	return registry, nil
 }
 
@@ -107,106 +77,3 @@ func (r *Registry) CapacityPolicies() map[string]domain.BackendCapacityPolicy {
 	}
 	return policies
 }
-
-func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
-	definition.Endpoint = strings.TrimSpace(definition.Endpoint)
-	if err := validateEndpoint(definition.Endpoint); err != nil {
-		return domain.Backend{}, fmt.Errorf("backend %q endpoint: %w", definition.ID, err)
-	}
-
-	definition.APIKeyEnv = strings.TrimSpace(definition.APIKeyEnv)
-	if definition.APIKeyEnv != "" && !environmentVariableName.MatchString(definition.APIKeyEnv) {
-		return domain.Backend{}, fmt.Errorf(
-			"backend %q api key environment variable %q is invalid",
-			definition.ID,
-			definition.APIKeyEnv,
-		)
-	}
-
-	if definition.ConcurrencyLimit < 0 {
-		return domain.Backend{}, fmt.Errorf(
-			"backend %q concurrency limit must not be negative",
-			definition.ID,
-		)
-	}
-	if definition.QueueCapacity < 0 {
-		return domain.Backend{}, fmt.Errorf(
-			"backend %q queue capacity must not be negative",
-			definition.ID,
-		)
-	}
-	if definition.ConcurrencyLimit == 0 {
-		if definition.QueueCapacitySet {
-			return domain.Backend{}, fmt.Errorf(
-				"backend %q queue capacity requires a positive concurrency limit",
-				definition.ID,
-			)
-		}
-		definition.QueueCapacity = 0
-	} else {
-		if !definition.QueueCapacitySet {
-			definition.QueueCapacity = defaultQueueCapacity
-			definition.QueueCapacitySet = true
-		}
-		maxInt := int(^uint(0) >> 1)
-		if definition.QueueCapacity > maxInt-definition.ConcurrencyLimit {
-			return domain.Backend{}, fmt.Errorf(
-				"backend %q total capacity overflows int",
-				definition.ID,
-			)
-		}
-	}
-
-	keys := make([]string, 0, len(definition.ExtraParams))
-	for key := range definition.ExtraParams {
-		keys = append(keys, key)
-	}
-	sort.Strings(keys)
-	for _, key := range keys {
-		if key == "" {
-			return domain.Backend{}, fmt.Errorf("backend %q extra parameter key must not be empty", definition.ID)
-		}
-		if llm.IsReservedOpenAIChatRequestField(key) {
-			return domain.Backend{}, fmt.Errorf(
-				"backend %q extra parameter %q collides with a reserved request field",
-				definition.ID,
-				key,
-			)
-		}
-	}
-
-	extraParams, err := jsonvalue.CopyMap(definition.ExtraParams)
-	if err != nil {
-		return domain.Backend{}, fmt.Errorf("backend %q extra parameters: %w", definition.ID, err)
-	}
-	definition.ExtraParams = extraParams
-	return definition, nil
-}
-
-func validateEndpoint(endpoint string) error {
-	if endpoint == "" {
-		return errors.New("must not be blank")
-	}
-	if strings.Contains(endpoint, "#") {
-		return errors.New("must not contain a fragment")
-	}
-
-	parsed, err := url.Parse(endpoint)
-	if err != nil {
-		return fmt.Errorf("must be a valid URL: %w", err)
-	}
-	scheme := strings.ToLower(parsed.Scheme)
-	if scheme != "http" && scheme != "https" {
-		return errors.New("must use http or https")
-	}
-	if !parsed.IsAbs() || parsed.Hostname() == "" {
-		return errors.New("must be absolute and include a host")
-	}
-	if parsed.User != nil {
-		return errors.New("must not contain user information")
-	}
-	if parsed.RawQuery != "" || parsed.ForceQuery {
-		return errors.New("must not contain a query string")
-	}
-	return nil
-}

internal/backend/registry_test.go

@@ -2,367 +2,55 @@ package backend_test
 
 import (
 	"errors"
-	"strings"
+	"reflect"
 	"testing"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/backend"
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 )
 
-const validEndpoint = "https://backend.example/v1"
-
-func TestRegistryIncludesExactOpenRouterDefinition(t *testing.T) {
-	registry, err := backend.NewRegistry(nil)
-	if err != nil {
-		t.Fatalf("construct registry: %v", err)
-	}
-
-	definition, err := registry.GetBackend(backend.OpenRouterID)
-	if err != nil {
-		t.Fatalf("look up OpenRouter: %v", err)
-	}
-	if definition.ID != "openrouter" ||
-		definition.Endpoint != "https://openrouter.ai/api/v1" ||
-		definition.APIKeyEnv != "OPENROUTER_API_KEY" ||
-		definition.ConcurrencyLimit != 16 ||
-		definition.QueueCapacity != 1024 ||
-		!definition.QueueCapacitySet ||
-		definition.ExtraParams != nil {
-		t.Fatalf("unexpected OpenRouter definition: %#v", definition)
-	}
-	policies := registry.CapacityPolicies()
-	if len(policies) != 1 ||
-		policies["openrouter"] != (domain.BackendCapacityPolicy{
-			ConcurrencyLimit: 16,
-			QueueCapacity:    1024,
-		}) {
-		t.Fatalf("unexpected OpenRouter capacity policies: %#v", policies)
-	}
-}
-
-func TestRegistryNormalizesUniqueAdditionsAndIsolatesMutations(t *testing.T) {
-	nested := map[string]int{"limit": 2}
-	extraParams := map[string]any{
-		"count":  int64(7),
-		"nested": nested,
-	}
-	registry, err := backend.NewRegistry([]domain.Backend{
-		{
-			ID:               " custom ",
-			Endpoint:         " https://custom.example/openai/v1 ",
-			APIKeyEnv:        " CUSTOM_API_KEY ",
-			ExtraParams:      extraParams,
-			ConcurrencyLimit: 3,
-			QueueCapacity:    2,
-			QueueCapacitySet: true,
-		},
-		{
-			ID:       "Custom",
-			Endpoint: validEndpoint,
+func TestEffectiveRegistryOwnsDefinitionsAndCapacityPolicies(t *testing.T) {
+	params := map[string]any{"nested": map[string]any{"enabled": true}}
+	registry, err := backend.NewEffectiveRegistry(map[string]domain.Backend{
+		" local.v1 ": {
+			Endpoint: "https://backend.example/v1", ExtraParams: params,
+			ConcurrencyLimit: 2, QueueCapacity: 3, QueueCapacitySet: true,
 		},
 	})
 	if err != nil {
-		t.Fatalf("construct registry: %v", err)
+		t.Fatal(err)
 	}
-
-	nested["limit"] = 99
-	extraParams["added"] = true
-
-	got, err := registry.GetBackend("custom")
-	if err != nil {
-		t.Fatalf("look up custom backend: %v", err)
+	params["nested"].(map[string]any)["enabled"] = false
+	selected, err := registry.GetBackend(" local.v1 ")
+	if err != nil || selected.ID != " local.v1 " || selected.ExtraParams["nested"].(map[string]any)["enabled"] != true {
+		t.Fatalf("effective backend was not captured under its literal ID: %+v, %v", selected, err)
 	}
-	if got.ID != "custom" ||
-		got.Endpoint != "https://custom.example/openai/v1" ||
-		got.APIKeyEnv != "CUSTOM_API_KEY" ||
-		got.ConcurrencyLimit != 3 ||
-		got.QueueCapacity != 2 ||
-		!got.QueueCapacitySet {
-		t.Fatalf("unexpected normalized definition: %#v", got)
+	selected.ExtraParams["nested"].(map[string]any)["enabled"] = false
+	again, err := registry.GetBackend(" local.v1 ")
+	if err != nil || again.ExtraParams["nested"].(map[string]any)["enabled"] != true {
+		t.Fatalf("lookup exposed registry values: %+v, %v", again, err)
 	}
-	if count, ok := got.ExtraParams["count"].(int64); !ok || count != 7 {
-		t.Fatalf("integer type or value changed: %#v", got.ExtraParams["count"])
-	}
-	gotNested, ok := got.ExtraParams["nested"].(map[string]int)
-	if !ok || gotNested["limit"] != 2 {
-		t.Fatalf("container type or value changed: %#v", got.ExtraParams["nested"])
-	}
-	if _, exists := got.ExtraParams["added"]; exists {
-		t.Fatalf("registry retained caller map: %#v", got.ExtraParams)
-	}
-
-	gotNested["limit"] = 100
-	got.ExtraParams["added"] = true
-	again, err := registry.GetBackend("custom")
-	if err != nil {
-		t.Fatalf("look up custom backend again: %v", err)
-	}
-	if again.ExtraParams["nested"].(map[string]int)["limit"] != 2 {
-		t.Fatalf("lookup exposed registry nested map: %#v", again.ExtraParams)
-	}
-	if _, exists := again.ExtraParams["added"]; exists {
-		t.Fatalf("lookup exposed registry map: %#v", again.ExtraParams)
-	}
-
-	if _, err := registry.GetBackend("Custom"); err != nil {
-		t.Fatalf("backend IDs should be case-sensitive: %v", err)
-	}
-
 	policies := registry.CapacityPolicies()
-	if len(policies) != 2 {
-		t.Fatalf("unexpected capacity policy count: %#v", policies)
-	}
-	policies["custom"] = domain.BackendCapacityPolicy{}
-	delete(policies, backend.OpenRouterID)
-	againPolicies := registry.CapacityPolicies()
-	if againPolicies["custom"] != (domain.BackendCapacityPolicy{
-		ConcurrencyLimit: 3,
-		QueueCapacity:    2,
+	if !reflect.DeepEqual(policies, map[string]domain.BackendCapacityPolicy{
+		" local.v1 ": {ConcurrencyLimit: 2, QueueCapacity: 3},
 	}) {
-		t.Fatalf("capacity policy map mutated registry state: %#v", againPolicies)
+		t.Fatalf("capacity policy = %+v", policies)
 	}
-	if _, ok := againPolicies[backend.OpenRouterID]; !ok {
-		t.Fatalf("capacity policy deletion mutated registry state: %#v", againPolicies)
+	delete(policies, " local.v1 ")
+	if len(registry.CapacityPolicies()) != 1 {
+		t.Fatal("capacity policy map exposed registry state")
 	}
 }
 
-func TestNewRegistryNormalizesCapacityPolicy(t *testing.T) {
-	maxInt := int(^uint(0) >> 1)
-	tests := []struct {
-		name       string
-		definition domain.Backend
-		want       domain.BackendCapacityPolicy
-		wantSet    bool
-		wantError  bool
-	}{
-		{
-			name:       "unlimited when omitted",
-			definition: domain.Backend{},
-		},
-		{
-			name: "default queue",
-			definition: domain.Backend{
-				ConcurrencyLimit: 2,
-			},
-			want: domain.BackendCapacityPolicy{
-				ConcurrencyLimit: 2,
-				QueueCapacity:    1024,
-			},
-			wantSet: true,
-		},
-		{
-			name: "explicit zero queue",
-			definition: domain.Backend{
-				ConcurrencyLimit: 2,
-				QueueCapacitySet: true,
-			},
-			want: domain.BackendCapacityPolicy{
-				ConcurrencyLimit: 2,
-			},
-			wantSet: true,
-		},
-		{
-			name: "negative concurrency limit",
-			definition: domain.Backend{
-				ConcurrencyLimit: -1,
-			},
-			wantError: true,
-		},
-		{
-			name: "negative queue capacity",
-			definition: domain.Backend{
-				ConcurrencyLimit: 1,
-				QueueCapacity:    -1,
-				QueueCapacitySet: true,
-			},
-			wantError: true,
-		},
-		{
-			name: "queue without limit",
-			definition: domain.Backend{
-				QueueCapacitySet: true,
-			},
-			wantError: true,
-		},
-		{
-			name: "total overflow",
-			definition: domain.Backend{
-				ConcurrencyLimit: maxInt,
-				QueueCapacity:    1,
-				QueueCapacitySet: true,
-			},
-			wantError: true,
-		},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			tc.definition.ID = "custom"
-			tc.definition.Endpoint = validEndpoint
-			registry, err := backend.NewRegistry([]domain.Backend{tc.definition})
-			if tc.wantError {
-				if err == nil {
-					t.Fatal("expected invalid capacity policy error")
-				}
-				return
-			}
-			if err != nil {
-				t.Fatalf("construct registry: %v", err)
-			}
-
-			definition, err := registry.GetBackend("custom")
-			if err != nil {
-				t.Fatalf("look up custom backend: %v", err)
-			}
-			if definition.ConcurrencyLimit != tc.want.ConcurrencyLimit ||
-				definition.QueueCapacity != tc.want.QueueCapacity ||
-				definition.QueueCapacitySet != tc.wantSet {
-				t.Fatalf("normalized capacity=(%d, %d, %t), want (%d, %d, %t)",
-					definition.ConcurrencyLimit,
-					definition.QueueCapacity,
-					definition.QueueCapacitySet,
-					tc.want.ConcurrencyLimit,
-					tc.want.QueueCapacity,
-					tc.wantSet,
-				)
-			}
-			policies := registry.CapacityPolicies()
-			got, ok := policies["custom"]
-			if ok != tc.wantSet || got != tc.want {
-				t.Fatalf("capacity policy=(%#v, %t), want (%#v, %t)", got, ok, tc.want, tc.wantSet)
-			}
-		})
-	}
-}
-
-func TestNewRegistryRejectsDuplicateIDs(t *testing.T) {
-	tests := []struct {
-		name      string
-		additions []domain.Backend
-		wantID    string
-	}{
-		{
-			name: "built-in collision after normalization",
-			additions: []domain.Backend{{
-				ID: " openrouter ",
-			}},
-			wantID: "openrouter",
-		},
-		{
-			name: "consumer collision after normalization",
-			additions: []domain.Backend{
-				{ID: "custom", Endpoint: validEndpoint},
-				{ID: " custom ", Endpoint: "https://other.example/v1"},
-			},
-			wantID: "custom",
-		},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			_, err := backend.NewRegistry(tc.additions)
-			if err == nil {
-				t.Fatal("expected duplicate ID error")
-			}
-			if !strings.Contains(err.Error(), tc.wantID) {
-				t.Fatalf("expected error to identify %q, got %v", tc.wantID, err)
-			}
-		})
-	}
-}
-
-func TestNewRegistryValidatesIDs(t *testing.T) {
-	for _, id := range []string{"", " \t\n "} {
-		t.Run(id, func(t *testing.T) {
-			_, err := backend.NewRegistry([]domain.Backend{{
-				ID:       id,
-				Endpoint: validEndpoint,
-			}})
-			if err == nil {
-				t.Fatal("expected blank ID error")
-			}
-		})
-	}
-}
-
-func TestNewRegistryValidatesEndpoints(t *testing.T) {
-	tests := []struct {
-		name     string
-		endpoint string
-	}{
-		{name: "blank", endpoint: ""},
-		{name: "relative", endpoint: "/v1"},
-		{name: "missing host", endpoint: "https:///v1"},
-		{name: "unsupported scheme", endpoint: "ftp://backend.example/v1"},
-		{name: "user information", endpoint: "https://user@backend.example/v1"},
-		{name: "query", endpoint: "https://backend.example/v1?mode=chat"},
-		{name: "empty query", endpoint: "https://backend.example/v1?"},
-		{name: "fragment", endpoint: "https://backend.example/v1#chat"},
-		{name: "empty fragment", endpoint: "https://backend.example/v1#"},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			_, err := backend.NewRegistry([]domain.Backend{{
-				ID:       "custom",
-				Endpoint: tc.endpoint,
-			}})
-			if err == nil {
-				t.Fatal("expected invalid endpoint error")
-			}
-		})
-	}
-}
-
-func TestNewRegistryValidatesEnvironmentVariableNames(t *testing.T) {
-	for _, name := range []string{"1API_KEY", "API-KEY", "API KEY", "ÅPI_KEY"} {
-		t.Run(name, func(t *testing.T) {
-			_, err := backend.NewRegistry([]domain.Backend{{
-				ID:        "custom",
-				Endpoint:  validEndpoint,
-				APIKeyEnv: name,
-			}})
-			if err == nil {
-				t.Fatal("expected invalid environment-variable name error")
-			}
-		})
-	}
-}
-
-func TestNewRegistryRejectsInvalidAndReservedExtraParameters(t *testing.T) {
-	tests := []struct {
-		name        string
-		extraParams map[string]any
-	}{
-		{name: "unsupported value", extraParams: map[string]any{"value": make(chan int)}},
-		{name: "reserved key", extraParams: map[string]any{"model": "override"}},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			_, err := backend.NewRegistry([]domain.Backend{{
-				ID:          "custom",
-				Endpoint:    validEndpoint,
-				ExtraParams: tc.extraParams,
-			}})
-			if err == nil {
-				t.Fatal("expected invalid extra parameters error")
-			}
-		})
-	}
-}
-
-func TestRegistryLookupReportsNotFound(t *testing.T) {
-	registry, err := backend.NewRegistry(nil)
+func TestEffectiveRegistryMissingBackend(t *testing.T) {
+	registry, err := backend.NewEffectiveRegistry(nil)
 	if err != nil {
-		t.Fatalf("construct registry: %v", err)
+		t.Fatal(err)
 	}
-
-	_, err = registry.GetBackend("missing")
-	if !errors.Is(err, backend.ErrBackendNotFound) {
-		t.Fatalf("expected ErrBackendNotFound, got %v", err)
+	if _, err := registry.GetBackend("missing"); !errors.Is(err, backend.ErrBackendNotFound) {
+		t.Fatalf("missing backend error = %v", err)
 	}
-	if !strings.Contains(err.Error(), "missing") {
-		t.Fatalf("expected error to identify backend, got %v", err)
+	if _, err := registry.GetBackend("local.v1"); !errors.Is(err, backend.ErrBackendNotFound) {
+		t.Fatalf("trimmed lookup unexpectedly matched: %v", err)
 	}
 }

internal/capacity/client.go

@@ -28,13 +28,13 @@ func (c *client) Generate(
 	ctx context.Context,
 	req domain.GenerateRequest,
 ) (*domain.GenerateResponse, error) {
-	pool := c.manager.getPool(req.Target.BackendID)
-	if pool == nil {
+	if c.manager.getPool(req.Target.BackendID) == nil {
 		return c.next.Generate(ctx, req)
 	}
-	if err := pool.acquire(ctx); err != nil {
+	release, err := c.manager.Acquire(ctx, req.Target.BackendID)
+	if err != nil {
 		return nil, err
 	}
-	defer pool.releaseActive()
+	defer release()
 	return c.next.Generate(ctx, req)
 }

internal/capacity/manager.go

@@ -107,6 +107,26 @@ func (m *Manager) getPool(backendID string) *pool {
 	return m.pools[backendID]
 }
 
+// Limited reports whether a backend uses an active-generation pool.
+func (m *Manager) Limited(backendID string) bool { return m.getPool(backendID) != nil }
+
+// Acquire obtains one active-generation permit for a backend. Its release
+// function is idempotent. Backends without a configured pool are unlimited.
+func (m *Manager) Acquire(ctx context.Context, backendID string) (func(), error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	pool := m.getPool(backendID)
+	if pool == nil {
+		return releaseNothing, nil
+	}
+	if err := pool.acquire(ctx); err != nil {
+		return nil, err
+	}
+	var once sync.Once
+	return func() { once.Do(pool.releaseActive) }, nil
+}
+
 func (p *pool) acquire(ctx context.Context) error {
 	p.mu.Lock()
 	if err := ctx.Err(); err != nil {

internal/capacity/manager_test.go

@@ -60,6 +60,26 @@ func TestNewManagerRejectsInvalidPolicies(t *testing.T) {
 	}
 }
 
+func TestManagerAcquireReleaseIsIdempotent(t *testing.T) {
+	manager, err := NewManager(map[string]domain.BackendCapacityPolicy{
+		"limited": {ConcurrencyLimit: 1},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	release, err := manager.Acquire(context.Background(), "limited")
+	if err != nil {
+		t.Fatal(err)
+	}
+	release()
+	release()
+	other, err := manager.Acquire(context.Background(), "limited")
+	if err != nil {
+		t.Fatal(err)
+	}
+	other()
+}
+
 func TestManagerAdmissionIsBoundedAndReleaseIsIdempotent(t *testing.T) {
 	policies := map[string]domain.BackendCapacityPolicy{
 		"limited": {
@@ -124,6 +144,11 @@ func TestManagerAdmissionHonorsContextAndUnlimitedBackends(t *testing.T) {
 	if err != nil {
 		t.Fatalf("construct manager: %v", err)
 	}
+	release, err := manager.Admit(context.Background(), "limited")
+	if err != nil {
+		t.Fatalf("fill limited pool: %v", err)
+	}
+	defer release()
 
 	ctx, cancel := context.WithCancel(context.Background())
 	cancel()

internal/catalog/catalog.go

@@ -0,0 +1,52 @@
+// Package catalog selects the maintained configuration assets supplied with
+// PromptKit. ConfigKit owns their composition and effective validation.
+package catalog
+
+import (
+	"context"
+	"fmt"
+	"io/fs"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	openrouter "gitea.maximumdirect.net/eric/promptkit-backend-openrouter/v3"
+	rakestrawhome "gitea.maximumdirect.net/eric/promptkit-backend-rakestrawhome/v3"
+	"gitea.maximumdirect.net/eric/promptkit/internal/backend"
+	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
+)
+
+// Underlays selects external catalogs followed by PromptKit-owned profiles.
+// Neither group is interpreted before the final effective configuration loads.
+func Underlays() ([]configkit.Underlay, error) {
+	openrouterInputs, err := assetInputs("OpenRouter catalog", openrouter.FS(), openrouter.Root)
+	if err != nil {
+		return nil, err
+	}
+	rakestrawhomeInputs, err := assetInputs("Rakestrawhome catalog", rakestrawhome.FS(), rakestrawhome.Root)
+	if err != nil {
+		return nil, err
+	}
+	catalogInputs := append(openrouterInputs, rakestrawhomeInputs...)
+	owned := configkit.Input{Source: configkit.Mapping("PromptKit-owned profiles", map[string]any{
+		"profiles": map[string]any{
+			"promptkit-openrouter-standard": map[string]any{"extends": "gemini-flash-latest"},
+			"promptkit-openrouter-auto":     map[string]any{"backend": backend.OpenRouterID, "model": "openrouter/auto"},
+		},
+	})}
+	return []configkit.Underlay{{Inputs: catalogInputs}, {Inputs: []configkit.Input{owned}}}, nil
+}
+
+func assetInputs(label string, filesystem fs.FS, root string) ([]configkit.Input, error) {
+	files, err := filecatalog.FindFSYAMLFiles(context.Background(), filesystem, root)
+	if err != nil {
+		return nil, fmt.Errorf("discover %s configuration: %w", label, err)
+	}
+	namespace := configkit.NewFilesystem(label, filesystem)
+	inputs := make([]configkit.Input, 0, len(files))
+	for _, name := range files {
+		inputs = append(inputs, configkit.Input{
+			Source:     namespace.File(name),
+			Filesystem: &configkit.FilesystemContext{FS: &namespace, Root: root},
+		})
+	}
+	return inputs, nil
+}

internal/catalog/catalog_test.go

@@ -0,0 +1,68 @@
+package catalog
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit/internal/configuration"
+)
+
+func TestPublishedCatalogsResolveThroughEffectiveConfiguration(t *testing.T) {
+	underlays, err := Underlays()
+	if err != nil {
+		t.Fatal(err)
+	}
+	state, err := configuration.Load(configkit.Input{Source: configkit.NoSource()}, underlays)
+	if err != nil {
+		t.Fatal(err)
+	}
+	openrouter, err := state.GetBackend("openrouter")
+	if err != nil || openrouter.Endpoint != "https://openrouter.ai/api/v1" || openrouter.ConcurrencyLimit != 16 || openrouter.QueueCapacity != 1024 {
+		t.Fatalf("OpenRouter backend: %+v, %v", openrouter, err)
+	}
+	rakestrawhome, err := state.GetBackend("rakestrawhome")
+	if err != nil || rakestrawhome.Endpoint != "https://inference.ai.rakestrawhome.com/v1" || rakestrawhome.ConcurrencyLimit != 4 {
+		t.Fatalf("Rakestrawhome backend: %+v, %v", rakestrawhome, err)
+	}
+	for id, want := range map[string]string{
+		"promptkit-openrouter-standard": "~google/gemini-flash-latest",
+		"promptkit-openrouter-auto":     "openrouter/auto",
+		"rakestrawhome-gemma-4-31b":     "google/gemma-4-31b-it",
+	} {
+		profile, err := state.GetProfile(context.Background(), id)
+		if err != nil || profile.ResolvedTarget.Model != want {
+			t.Fatalf("profile %s: %+v, %v", id, profile, err)
+		}
+	}
+}
+
+func TestConsumerOverlayReplacesCatalogDefaultsBeforeValidation(t *testing.T) {
+	underlays, err := Underlays()
+	if err != nil {
+		t.Fatal(err)
+	}
+	underlays = append(underlays, configkit.Underlay{Inputs: []configkit.Input{{Source: configkit.Mapping("operator", map[string]any{
+		"backends": map[string]any{"openrouter": map[string]any{"endpoint": "invalid lower value"}},
+	})}}})
+	main := configkit.Input{Source: configkit.Mapping("application", map[string]any{
+		"backends": map[string]any{"openrouter": map[string]any{"endpoint": "https://alternate.example/v1"}},
+		"profiles": map[string]any{"gemini-flash-latest": map[string]any{"model": "consumer-model"}},
+	})}
+	state, err := configuration.Load(main, underlays)
+	if err != nil {
+		t.Fatalf("replaced lower-layer value failed final validation: %v", err)
+	}
+	profile, err := state.GetProfile(context.Background(), "promptkit-openrouter-standard")
+	if err != nil || profile.ResolvedTarget.Model != "consumer-model" || profile.ResolvedTarget.Endpoint != "https://alternate.example/v1" {
+		t.Fatalf("maintained profile did not observe final configuration: %+v, %v", profile, err)
+	}
+	_, err = configuration.Load(configkit.Input{Source: configkit.Mapping("invalid final", map[string]any{
+		"backends": map[string]any{"openrouter": map[string]any{"endpoint": "invalid final value"}},
+	})}, underlays[:2])
+	var loadError *configkit.LoadError
+	if !errors.As(err, &loadError) {
+		t.Fatalf("invalid final backend did not produce a ConfigKit load error: %v", err)
+	}
+}

internal/configuration/benchmark_test.go

@@ -0,0 +1,295 @@
+package configuration
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+const benchmarkDefinitionCount = 128
+
+var (
+	constructionResult       *loaded
+	selectedPrompt           *domain.PromptDefinition
+	selectedPromptHash       string
+	selectedPromptInspection *domain.PromptInspection
+	selectedProfile          *domain.ExecutionProfile
+	selectedProfileIdentity  string
+)
+
+type configurationShape struct {
+	main             configkit.Input
+	underlays        []configkit.Underlay
+	inputEntries     int
+	effectiveEntries int
+	tiers            int
+}
+
+func benchmarkPrompt(profileID string) map[string]any {
+	return map[string]any{
+		"default_profile": profileID,
+		"messages":        []any{map[string]any{"role": "user", "content": "Hello {{.name}}"}},
+		"output":          map[string]any{"format": "text", "validation_mode": "none"},
+	}
+}
+
+func independentDefinitions(count int) configurationShape {
+	profiles := make(map[string]any, count)
+	prompts := make(map[string]any, count)
+	for index := range count {
+		profileID := fmt.Sprintf("profile-%03d", index)
+		promptID := fmt.Sprintf("prompt-%03d", index)
+		profiles[profileID] = map[string]any{"endpoint": "https://example.test/v1", "model": profileID}
+		prompts[promptID] = map[string]any{"1": benchmarkPrompt(profileID)}
+	}
+	return configurationShape{
+		main:             configkit.Input{Source: configkit.Mapping("independent definitions", map[string]any{"profiles": profiles, "prompts": prompts})},
+		inputEntries:     count * 2,
+		effectiveEntries: count * 2,
+		tiers:            1,
+	}
+}
+
+func fileBackedDefinitions(count int) configurationShape {
+	profiles := make(map[string]any, count)
+	prompts := make(map[string]any, count)
+	for index := range count {
+		profileID := fmt.Sprintf("profile-%03d", index)
+		promptID := fmt.Sprintf("prompt-%03d", index)
+		profiles[profileID] = map[string]any{"endpoint": "https://example.test/v1", "model": profileID}
+		prompt := benchmarkPrompt(profileID)
+		prompt["messages"] = []any{map[string]any{"role": "user", "content_file": "body.md"}}
+		prompts[promptID] = map[string]any{"1": prompt}
+	}
+	namespace := configkit.NewFilesystem("benchmark content", fstest.MapFS{
+		"assets/body.md": &fstest.MapFile{Data: []byte("Hello {{.name}}")},
+	})
+	return configurationShape{
+		main: configkit.Input{
+			Source:     configkit.Mapping("file-backed definitions", map[string]any{"profiles": profiles, "prompts": prompts}),
+			Filesystem: &configkit.FilesystemContext{FS: &namespace, Root: ".", BaseDir: "assets"},
+		},
+		inputEntries: count * 2, effectiveEntries: count * 2, tiers: 1,
+	}
+}
+
+func sharedBaseDefinitions(count int) configurationShape {
+	profiles := make(map[string]any, count+1)
+	profiles["base"] = map[string]any{
+		"endpoint": "https://example.test/v1", "model": "shared-model",
+		"extra_params": map[string]any{"provider": map[string]any{"mode": "shared", "limit": 4}},
+	}
+	for index := range count {
+		profiles[fmt.Sprintf("child-%03d", index)] = map[string]any{"extends": "base", "temperature": float64(index%10) / 10}
+	}
+	return configurationShape{
+		main: configkit.Input{Source: configkit.Mapping("inherited definitions", map[string]any{
+			"profiles": profiles,
+			"prompts":  map[string]any{"shared": map[string]any{"1": benchmarkPrompt(fmt.Sprintf("child-%03d", count-1))}},
+		})},
+		inputEntries:     count + 2,
+		effectiveEntries: count + 2,
+		tiers:            1,
+	}
+}
+
+func layeredDefinitions(count int) configurationShape {
+	base := make(map[string]any, count)
+	operator := make(map[string]any, count)
+	programmatic := make(map[string]any, count)
+	final := make(map[string]any, count)
+	for index := range count {
+		id := fmt.Sprintf("profile-%03d", index)
+		base[id] = map[string]any{"endpoint": "https://example.test/v1", "model": "base-model"}
+		operator[id] = map[string]any{"model": "operator-model"}
+		programmatic[id] = map[string]any{"extra_params": map[string]any{"provider": map[string]any{"mode": "programmatic"}}}
+		final[id] = map[string]any{"temperature": float64(index%10) / 10}
+	}
+	return configurationShape{
+		underlays: []configkit.Underlay{
+			{Inputs: []configkit.Input{{Source: configkit.Mapping("fallback", map[string]any{"profiles": base})}}},
+			{Inputs: []configkit.Input{{Source: configkit.Mapping("operator", map[string]any{"profiles": operator})}}},
+			{Inputs: []configkit.Input{{Source: configkit.Mapping("programmatic", map[string]any{"profiles": programmatic})}}},
+		},
+		main: configkit.Input{Source: configkit.Mapping("main", map[string]any{
+			"profiles": final,
+			"prompts":  map[string]any{"layered": map[string]any{"1": benchmarkPrompt(fmt.Sprintf("profile-%03d", count-1))}},
+		})},
+		inputEntries:     count*4 + 1,
+		effectiveEntries: count + 1,
+		tiers:            4,
+	}
+}
+
+func BenchmarkConfigurationConstruction(b *testing.B) {
+	for _, scenario := range []struct {
+		name  string
+		shape configurationShape
+	}{
+		{name: "independent", shape: independentDefinitions(benchmarkDefinitionCount)},
+		{name: "file_backed", shape: fileBackedDefinitions(64)},
+		{name: "shared_base", shape: sharedBaseDefinitions(benchmarkDefinitionCount)},
+		{name: "layered", shape: layeredDefinitions(benchmarkDefinitionCount)},
+	} {
+		b.Run(scenario.name, func(b *testing.B) {
+			b.ReportAllocs()
+			for b.Loop() {
+				result, err := load(scenario.shape.main, scenario.shape.underlays)
+				if err != nil {
+					b.Fatal(err)
+				}
+				constructionResult = result
+			}
+			b.ReportMetric(float64(scenario.shape.inputEntries), "input-entries/op")
+			b.ReportMetric(float64(scenario.shape.effectiveEntries), "effective-entries/op")
+			b.ReportMetric(float64(scenario.shape.tiers), "tiers/op")
+		})
+	}
+}
+
+func BenchmarkConfigurationSelection(b *testing.B) {
+	shape := independentDefinitions(benchmarkDefinitionCount)
+	result, err := load(shape.main, nil)
+	if err != nil {
+		b.Fatal(err)
+	}
+	b.Run("prompt", func(b *testing.B) {
+		b.ReportAllocs()
+		for b.Loop() {
+			definition, _, err := result.prompt("prompt-127", "1")
+			if err != nil {
+				b.Fatal(err)
+			}
+			selectedPrompt = definition
+		}
+		b.ReportMetric(float64(shape.effectiveEntries), "effective-entries/op")
+	})
+	b.Run("profile", func(b *testing.B) {
+		b.ReportAllocs()
+		for b.Loop() {
+			profile, err := result.GetProfile(context.Background(), "profile-127")
+			if err != nil {
+				b.Fatal(err)
+			}
+			selectedProfile = profile
+		}
+		b.ReportMetric(float64(shape.effectiveEntries), "effective-entries/op")
+	})
+}
+
+func inspectionExtraParams(entries int) map[string]any {
+	providers := make(map[string]any, entries)
+	for index := range entries {
+		providers[fmt.Sprintf("provider-%03d", index)] = map[string]any{
+			"enabled": true,
+			"labels":  []any{fmt.Sprintf("tier-%d", index%4), fmt.Sprintf("region-%d", index%3)},
+			"limits":  map[string]any{"burst": index + 1, "period": "minute"},
+			"routes":  []any{map[string]any{"weight": index + 1, "fallback": true}},
+		}
+	}
+	return map[string]any{"provider_options": providers}
+}
+
+func inspectionDefinitions(extraParams map[string]any) configurationShape {
+	profiles := map[string]any{
+		"small": map[string]any{"model": "inspection-model", "extra_params": extraParams},
+	}
+	prompt := benchmarkPrompt("small")
+	prompt["inputs"] = []any{map[string]any{"name": "transcript", "required": true}}
+	prompts := map[string]any{
+		"inspection": map[string]any{"1": prompt},
+	}
+	return configurationShape{
+		main: configkit.Input{Source: configkit.Mapping("inspection definitions", map[string]any{
+			"profiles": profiles,
+			"prompts":  prompts,
+		})},
+		inputEntries: 2, effectiveEntries: 2, tiers: 1,
+	}
+}
+
+func BenchmarkInspectionSelection(b *testing.B) {
+	for _, scenario := range []struct {
+		name    string
+		entries int
+	}{
+		{name: "small_extra", entries: 2},
+		{name: "nested_extra", entries: 64},
+	} {
+		shape := inspectionDefinitions(inspectionExtraParams(scenario.entries))
+		result, err := load(shape.main, nil)
+		if err != nil {
+			b.Fatal(err)
+		}
+		ctx := context.Background()
+		definition, hash, err := result.GetPromptSnapshot(ctx, "inspection", "1")
+		if err != nil || definition.ID != "inspection" || definition.Version != "1" || hash == "" ||
+			len(definition.Inputs) != 1 || definition.Inputs[0].Name != "transcript" || !definition.Inputs[0].Required {
+			b.Fatalf("prompt lookup fixture is invalid: definition=%+v hash=%q err=%v", definition, hash, err)
+		}
+		profile, err := result.GetProfile(ctx, "small")
+		if err != nil || profile.ID != "small" || profile.ResolvedTarget.Model != "inspection-model" ||
+			len(profile.ResolvedTarget.ExtraParams["provider_options"].(map[string]any)) != scenario.entries {
+			b.Fatalf("profile lookup fixture is invalid: profile=%+v err=%v", profile, err)
+		}
+		inspection, err := result.GetPromptInspection(ctx, "inspection", "1")
+		if err != nil || inspection.PromptID != "inspection" || inspection.PromptVersion != "1" ||
+			len(inspection.Inputs) != 1 || inspection.Inputs[0].Name != "transcript" {
+			b.Fatalf("prompt inspection fixture is invalid: inspection=%+v err=%v", inspection, err)
+		}
+		identity, err := result.GetProfileIdentity(ctx, "small")
+		if err != nil || identity != "small" {
+			b.Fatalf("profile identity fixture is invalid: identity=%q err=%v", identity, err)
+		}
+
+		b.Run("prompt/"+scenario.name, func(b *testing.B) {
+			b.ReportAllocs()
+			for b.Loop() {
+				definition, hash, err := result.GetPromptSnapshot(ctx, "inspection", "1")
+				if err != nil {
+					b.Fatal(err)
+				}
+				selectedPrompt = definition
+				selectedPromptHash = hash
+			}
+			b.ReportMetric(float64(scenario.entries), "extra-entries/op")
+		})
+		b.Run("prompt_inspection/"+scenario.name, func(b *testing.B) {
+			b.ReportAllocs()
+			for b.Loop() {
+				inspection, err := result.GetPromptInspection(ctx, "inspection", "1")
+				if err != nil {
+					b.Fatal(err)
+				}
+				selectedPromptInspection = inspection
+			}
+			b.ReportMetric(float64(scenario.entries), "extra-entries/op")
+		})
+		b.Run("profile/"+scenario.name, func(b *testing.B) {
+			b.ReportAllocs()
+			for b.Loop() {
+				profile, err := result.GetProfile(ctx, "small")
+				if err != nil {
+					b.Fatal(err)
+				}
+				selectedProfile = profile
+			}
+			b.ReportMetric(float64(scenario.entries), "extra-entries/op")
+		})
+		b.Run("profile_identity/"+scenario.name, func(b *testing.B) {
+			b.ReportAllocs()
+			for b.Loop() {
+				identity, err := result.GetProfileIdentity(ctx, "small")
+				if err != nil {
+					b.Fatal(err)
+				}
+				selectedProfileIdentity = identity
+			}
+			b.ReportMetric(float64(scenario.entries), "extra-entries/op")
+		})
+	}
+}

internal/configuration/inspection.go

@@ -0,0 +1,66 @@
+package configuration
+
+import (
+	"io"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+)
+
+// Inspection is the read-only subset of the retained ConfigKit result used by
+// the facade. It does not expose typed configuration or resource contexts.
+type Inspection interface {
+	Metadata() configkit.Metadata
+	Sources() []configkit.SourceInfo
+	Diagnostics() []configkit.Diagnostic
+	Explain(configkit.Path) (configkit.Explanation, bool)
+	WriteEffective(io.Writer) error
+	WriteExplanation(io.Writer, configkit.Path) error
+	capturedResult() *configkit.Result[document]
+}
+
+type inspectionResult struct {
+	*configkit.Result[document]
+}
+
+func (inspection inspectionResult) capturedResult() *configkit.Result[document] {
+	return inspection.Result
+}
+
+// Inspection returns only the retained result's inspection operations.
+func (state *loaded) Inspection() Inspection {
+	if state == nil {
+		return nil
+	}
+	return inspectionResult{Result: state.result}
+}
+
+// Diff compares two captured ConfigKit results without reconstructing them.
+func Diff(left, right Inspection) ([]configkit.Difference, error) {
+	var leftResult, rightResult *configkit.Result[document]
+	if left != nil {
+		leftResult = left.capturedResult()
+	}
+	if right != nil {
+		rightResult = right.capturedResult()
+	}
+	return configkit.Diff(leftResult, rightResult)
+}
+
+func secretProviderPath(path configkit.Path) bool {
+	pointer := path.String()
+	if pointer == "" {
+		return false
+	}
+	parts := strings.Split(strings.TrimPrefix(pointer, "/"), "/")
+	if len(parts) < 4 || (parts[0] != "backends" && parts[0] != "profiles") || parts[2] != "extra_params" {
+		return false
+	}
+	key := strings.ToLower(strings.ReplaceAll(strings.ReplaceAll(parts[len(parts)-1], "~1", "/"), "~0", "~"))
+	switch key {
+	case "api_key", "apikey", "authorization", "credential", "credentials", "password", "secret", "token", "access_token":
+		return true
+	default:
+		return false
+	}
+}

internal/configuration/listing.go

@@ -0,0 +1,29 @@
+package configuration
+
+import (
+	"context"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func (state *loaded) ListProfiles(ctx context.Context) ([]string, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	profiles := append([]string(nil), state.profileIDs...)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	return profiles, nil
+}
+
+func (state *loaded) ListPrompts(ctx context.Context) ([]domain.PromptIdentity, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	prompts := append([]domain.PromptIdentity(nil), state.promptIDs...)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	return prompts, nil
+}

internal/configuration/listing_test.go

@@ -0,0 +1,142 @@
+package configuration
+
+import (
+	"context"
+	"errors"
+	"reflect"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/ConfigKit"
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
+	state, err := Load(mapping("identities", map[string]any{
+		"backends": map[string]any{
+			"local": map[string]any{"endpoint": "http://localhost:8080/v1"},
+		},
+		"profiles": map[string]any{
+			"z/profile":       map[string]any{"backend": "local", "model": "z-model"},
+			"partial/profile": map[string]any{"backend": "local"},
+			"a/profile":       map[string]any{"backend": "local", "model": "a-model"},
+			" z/profile ":     map[string]any{"backend": "local", "model": "spaced-model"},
+		},
+		"prompts": map[string]any{
+			"z/prompt": map[string]any{"v1": listingPromptDefinition()},
+			"a/prompt": map[string]any{"v1": listingPromptDefinition()},
+			" z/prompt ": map[string]any{
+				"2":  listingPromptDefinition(),
+				"10": listingPromptDefinition(),
+			},
+		},
+	}), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	profiles, err := state.ListProfiles(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	wantProfiles := []string{" z/profile ", "a/profile", "partial/profile", "z/profile"}
+	if !reflect.DeepEqual(profiles, wantProfiles) {
+		t.Fatalf("profile IDs = %#v, want %#v", profiles, wantProfiles)
+	}
+
+	prompts, err := state.ListPrompts(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	wantPrompts := []domain.PromptIdentity{
+		{ID: " z/prompt ", Version: "10"},
+		{ID: " z/prompt ", Version: "2"},
+		{ID: "a/prompt", Version: "v1"},
+		{ID: "z/prompt", Version: "v1"},
+	}
+	if !reflect.DeepEqual(prompts, wantPrompts) {
+		t.Fatalf("prompt identities = %#v, want %#v", prompts, wantPrompts)
+	}
+}
+
+func TestLoadedListsAreIndependent(t *testing.T) {
+	state, err := Load(mapping("identities", map[string]any{
+		"profiles": map[string]any{"profile": map[string]any{}},
+		"prompts":  map[string]any{"example": map[string]any{"v1": listingPromptDefinition()}},
+	}), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	profiles, err := state.ListProfiles(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	profiles = append(profiles, "caller profile")
+	profiles[0] = "changed profile"
+
+	prompts, err := state.ListPrompts(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	prompts[0].ID = "changed prompt"
+	prompts[0].Version = "changed version"
+
+	profiles, err = state.ListProfiles(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !reflect.DeepEqual(profiles, []string{"profile"}) {
+		t.Fatalf("caller mutation changed profile inventory: %#v", profiles)
+	}
+	prompts, err = state.ListPrompts(context.Background())
+	if err != nil {
+		t.Fatal(err)
+	}
+	wantPrompts := []domain.PromptIdentity{{ID: "example", Version: "v1"}}
+	if !reflect.DeepEqual(prompts, wantPrompts) {
+		t.Fatalf("caller mutation changed prompt inventory: %#v", prompts)
+	}
+}
+
+func TestLoadedListsEmptyInventory(t *testing.T) {
+	state, err := Load(configkit.Input{Source: configkit.NoSource()}, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	profiles, err := state.ListProfiles(context.Background())
+	if err != nil || len(profiles) != 0 {
+		t.Fatalf("empty profile listing = %#v, %v", profiles, err)
+	}
+	prompts, err := state.ListPrompts(context.Background())
+	if err != nil || len(prompts) != 0 {
+		t.Fatalf("empty prompt listing = %#v, %v", prompts, err)
+	}
+}
+
+func TestLoadedListsHonorCanceledContext(t *testing.T) {
+	state, err := Load(mapping("identities", map[string]any{
+		"profiles": map[string]any{"partial": map[string]any{}},
+		"prompts":  map[string]any{"example": map[string]any{"v1": listingPromptDefinition()}},
+	}), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	profiles, err := state.ListProfiles(ctx)
+	if !errors.Is(err, context.Canceled) || len(profiles) != 0 {
+		t.Fatalf("canceled profile listing = %#v, %v", profiles, err)
+	}
+	prompts, err := state.ListPrompts(ctx)
+	if !errors.Is(err, context.Canceled) || len(prompts) != 0 {
+		t.Fatalf("canceled prompt listing = %#v, %v", prompts, err)
+	}
+}
+
+func listingPromptDefinition() map[string]any {
+	return map[string]any{
+		"messages": []any{map[string]any{"role": "user", "content": "Hello."}},
+		"output":   map[string]any{"format": "text", "validation_mode": "basic"},
+	}
+}