Prepare documentation for the v0.8.0 release

README.md

@@ -33,6 +33,18 @@ boundary and constraints that framework work must preserve.
 
 ## Release Guidance
 
+Consumers upgrading from `v0.7.0` to `v0.8.0` should read 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).
 

backends.go

@@ -9,6 +9,10 @@ import (
 // backend.
 const BackendOpenRouter = backend.OpenRouterID
 
+// BackendRakestrawHome is the reserved ID of Promptkit's built-in
+// Rakestrawhome backend.
+const BackendRakestrawHome = backend.RakestrawHomeID
+
 // BackendLocal is the case-sensitive conventional ID used by [LocalBackend].
 // It is not a built-in or reserved backend and must be registered with
 // [WithBackend].
@@ -20,21 +24,25 @@ const BackendLocal = "local"
 // 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.
+	// requires a non-blank value. Built-in backend IDs are 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
+	// APIKeyEnv optionally names an environment lookup source for an API key.
-	// key. NewEngine trims it and requires the portable form
+	// NewEngine trims it and requires the portable form [A-Za-z_][A-Za-z0-9_]*.
-	// [A-Za-z_][A-Za-z0-9_]*. Store only the name, never a credential value.
+	// A direct RunRequest.APIKey takes precedence. When no usable credential is
+	// available, the built-in client omits Authorization; injected clients own
+	// their own credential-resolution behavior. 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.
+	// no defaults. NewEngine deeply copies the map and rejects excessively deep
+	// or large values for safety.
 	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
@@ -73,11 +81,11 @@ func LocalBackend(endpoint string, concurrencyLimit int) Backend {
 //
 // 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,
+// definition makes NewEngine fail with ErrInvalidConfig. Built-in IDs,
-// BackendOpenRouter cannot be replaced. The immutable registration is scoped
+// including [BackendOpenRouter] and [BackendRakestrawHome], cannot be
-// to the resulting Engine and cannot be enumerated, replaced, removed, or
+// replaced. The immutable registration is scoped to the resulting Engine and
-// mutated after construction. WithBackend does not install package-global
+// cannot be enumerated, replaced, removed, or mutated after construction.
-// state.
+// WithBackend does not install package-global state.
 func WithBackend(backend Backend) Option {
 	queueCapacity := 0
 	queueCapacitySet := backend.QueueCapacity != nil

doc.go

@@ -23,8 +23,9 @@
 // InspectProfile return copied inspection values. 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. Returned structured errors are likewise caller-owned and may
+// returned them. [CapacityError] values are caller-owned and may be mutated
-// be mutated without affecting engine state or another error.
+// 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
 //
@@ -53,9 +54,14 @@
 // Construction, inspection, handle, and error values, including [Config],
 // [Backend], [RunRequest], [ArtifactRef], [ExecutionTargetOverride], [Profile],
 // [OpenAICompatibleProfileConfig], [ProfileInspection],
-// [PromptInputDefinition], [PromptInspection], [PreparedExecution], and
+// [PromptInputDefinition], [PromptInspection], [PreparedExecution],
-// [CapacityError], do not have stable JSON representations. Direct API keys
+// [CapacityError], and [GenerationError], do not have stable JSON
-// are nevertheless excluded from JSON for every public value.
+// 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/consumers/pkg-promptkit.md

@@ -40,6 +40,36 @@ validation, and default transport behavior. Source discovery, format
 validation, and profile precedence are defined by the
 [framework format reference](../formats.md).
 
+## Supply Embedded Application Defaults
+
+Use `WithFallbackProfileFS` when an application packages profile definitions
+that should apply unless an operator provides an ordinary configured profile
+with the same ID. For example, an application can embed its defaults while
+continuing to use `ProfileDir` for operator overrides:
+
+```go
+//go:embed profiles/*.yaml
+var applicationProfiles embed.FS
+
+engine, err := promptkit.NewEngine(promptkit.Config{
+	PromptDir:  "prompts",
+	ProfileDir: operatorProfileDir,
+},
+	promptkit.WithFallbackProfileFS(applicationProfiles, "profiles"),
+)
+```
+
+Keep application-owned profile IDs and definitions in the embedded source.
+Use the ordinary configured profile source for operator overrides. Leave
+`operatorProfileDir` empty when the operator did not configure an override
+directory; a non-empty path names an authoritative higher-precedence source,
+so an unavailable or unreadable directory is an error rather than a reason to
+fall back. The
+[framework format reference](../formats.md#source-and-profile-precedence)
+owns the exact profile format and lookup order; the
+[`WithFallbackProfileFS` GoDoc](../../engine.go) owns its option contract and
+validation rules.
+
 ## Inspect A Prompt Before Preparation
 
 Use [`Engine.InspectPrompt`](../../engine.go) to check one configured prompt's
@@ -143,6 +173,29 @@ semantics. The
 [OpenAI-compatible integration contract](../integrations/openai-compatible-chat.md)
 owns the built-in client's outbound HTTP behavior.
 
+### 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 call, so it can increase latency and
+usage; `RunResult.Usage` is cumulative and `Validation.RepairAttempts` reports
+calls actually started. 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.
+
 ## Inputs, Profiles, And Overrides
 
 Use `File`, `Inline`, or `InlineWithURI` to construct request inputs. A request
@@ -152,13 +205,60 @@ 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,
+interact with prompt definitions, file-backed and application fallback
-and framework defaults.
+profiles, built-ins, schemas, and framework defaults.
 
 For programmatic profiles,
 [`OpenAICompatibleProfile`](../../profiles.go) converts ordinary
 OpenAI-compatible settings into a value accepted by `WithProfiles`.
 
+### 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
+promptkit.WithProfiles(promptkit.Profile{
+	ID:              "weather-light",
+	BaseProfileID:   "deepseek-4-flash",
+	ReasoningEffort: "high",
+})
+```
+
+Select `weather-light` in a prompt or `RunRequest.ProfileID`; it remains the
+reported selected profile. See the [profile inheritance format
+reference](../formats.md#profile-inheritance) and the
+[`Profile` GoDoc](../../types.go) for exact lookup, merging, and validation
+behavior.
+
+### 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)
+}
+```
+
+Do not register `rakestrawhome` manually. When adopting this built-in, remove
+an existing `WithBackend` registration with that exact ID; retaining it causes
+the intentional duplicate-ID configuration error. 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
@@ -174,7 +274,7 @@ if err != nil {
 
 target := inspection.EffectiveModelParams
 if target.APIKeyEnv != "" {
-	// Apply application policy for the named environment variable.
+	// This is a configured optional environment lookup source.
 } else if inspection.APIKeyRequired {
 	// Arrange a direct credential before later execution.
 }
@@ -183,9 +283,11 @@ if target.APIKeyEnv != "" {
 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. Inspection reports credential requirements but leaves the
+later execution. A reported `APIKeyEnv` is a configured optional source, while
-timing of credential enforcement to the application. The method's
+`APIKeyRequired` is the explicit local requirement. The
-[GoDoc](../../engine.go) owns its exact result and error contract.
+[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
 
@@ -393,6 +495,24 @@ status; those choices remain with the consuming application. The
 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 retry and 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 retry or presentation behavior.
+The [`GenerationError` GoDoc](../../generation_error.go) owns the exact typed
+error contract.
+
 ## Application Boundary
 
 Promptkit is an importable library. It does not own a command, inbound HTTP

docs/development.md

@@ -45,11 +45,17 @@ 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
+This section is the canonical local validation workflow for Promptkit. Run
-running the documented checks before accepting changes. Run the default Go
+every command from the repository root before accepting a change. The test
-validation from the Promptkit repository root:
+suite and maintained examples are deterministic, offline, and require no real
+provider credentials.
+
+### Tests, Analysis, Build, And Examples
+
+Run the ordinary and race-enabled suites, static analysis, the build, and both
+maintained consumer examples:
 
 ```sh
 go test ./...
@@ -57,85 +63,171 @@ go test -race ./...
 go vet ./...
 go build ./...
 go run ./examples/go-library/prepare
+go run ./examples/go-library/run
 ```
 
-Check formatting across every tracked Go file:
+Both examples must exit successfully. Review their JSON output: preparation
+must report the selected offline prompt, profile, model, and message count;
+execution must report the deterministic generated output, successful
+validation, selected offline model, and usage. Neither command may contact a
+provider or require credentials.
+
+### Go Formatting
+
+Check every tracked Go file. The final command must succeed and the captured
+list must be empty:
 
 ```sh
-gofmt -l $(git ls-files '*.go')
+unformatted=$(
+    git ls-files '*.go' |
+        while IFS= read -r go_file
+        do
+            gofmt -l "$go_file"
+        done
+)
+test -z "$unformatted"
 ```
 
-The formatting command must produce no paths. Follow every added or changed
+### Local Markdown Links
-Markdown link and confirm its target exists. Finally, check whitespace:
+
+Use the Python standard library to verify every repository-relative Markdown
+target and local heading fragment. The check is offline and prints nothing on
+success:
+
+```sh
+python3 - <<'PY'
+from pathlib import Path
+import re
+import subprocess
+import sys
+from urllib.parse import unquote
+
+root = Path.cwd().resolve()
+markdown_files = [
+    root / name
+    for name in subprocess.check_output(
+        ["git", "ls-files", "*.md"], text=True
+    ).splitlines()
+]
+link_pattern = re.compile(r"!?\[[^]]*\]\(([^)]+)\)")
+heading_pattern = re.compile(r"^#{1,6}\s+(.+?)\s*#*\s*$")
+scheme_pattern = re.compile(r"^[a-z][a-z0-9+.-]*:", re.IGNORECASE)
+
+
+def markdown_lines(path):
+    in_fence = False
+    fence = ""
+    for line in path.read_text(encoding="utf-8").splitlines():
+        stripped = line.lstrip()
+        marker = stripped[:3]
+        if marker in {"```", "~~~"}:
+            if not in_fence:
+                in_fence = True
+                fence = marker
+            elif marker == fence:
+                in_fence = False
+                fence = ""
+            continue
+        if not in_fence:
+            yield line
+
+
+anchor_cache = {}
+
+
+def anchors(path):
+    if path in anchor_cache:
+        return anchor_cache[path]
+    found = set()
+    counts = {}
+    for line in markdown_lines(path):
+        match = heading_pattern.match(line)
+        if not match:
+            continue
+        heading = re.sub(r"<[^>]+>", "", match.group(1)).replace("`", "")
+        base = re.sub(r"[^\w\- ]", "", heading.lower()).replace(" ", "-")
+        count = counts.get(base, 0)
+        counts[base] = count + 1
+        found.add(base if count == 0 else f"{base}-{count}")
+    anchor_cache[path] = found
+    return found
+
+
+failures = []
+for source in markdown_files:
+    text = "\n".join(markdown_lines(source))
+    for match in link_pattern.finditer(text):
+        target = match.group(1).strip()
+        if target.startswith("<") and target.endswith(">"):
+            target = target[1:-1]
+        if scheme_pattern.match(target) or target.startswith("//"):
+            continue
+        path_text, separator, fragment = target.partition("#")
+        destination = source if not path_text else source.parent / unquote(path_text)
+        try:
+            destination = destination.resolve()
+            destination.relative_to(root)
+        except ValueError:
+            failures.append(f"{source.relative_to(root)}: escapes repository: {target}")
+            continue
+        if not destination.exists():
+            failures.append(f"{source.relative_to(root)}: missing target: {target}")
+            continue
+        if separator and destination.suffix.lower() == ".md":
+            fragment = unquote(fragment).lower()
+            if fragment not in anchors(destination):
+                failures.append(f"{source.relative_to(root)}: missing anchor: {target}")
+
+if failures:
+    print("\n".join(failures), file=sys.stderr)
+    raise SystemExit(1)
+PY
+```
+
+### Repository Hygiene And Review
+
+Reject an active Go workspace, tracked workspace files, a vendor tree, or a
+module replacement:
+
+```sh
+case "$(go env GOWORK)" in
+    ''|off) ;;
+    *) printf '%s\n' 'an active Go workspace is not allowed' >&2; exit 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
+    exit 1
+fi
+```
+
+Check whitespace in both unstaged and staged changes. List ignored files and
+scan tracked content for common credential forms:
 
 ```sh
 git diff --check
+git diff --cached --check
+test -z "$(git ls-files --others --ignored --exclude-standard)"
+credential_pattern='-----BEGIN ([A-Z0-9]+ )?PRIV''ATE KEY-----|AKI''A[0-9A-Z]{16}|gh[pousr]_[A-Za-z0-9]{36,}|sk-[A-Za-z0-9]{32,}'
+if git grep -nEI -e "$credential_pattern" -- .
+then
+    printf '%s\n' 'possible credential found' >&2
+    exit 1
+fi
 ```
 
-Documentation-only work does not require unrelated new tests, but it still
+Inspect `git status --short --untracked-files=all` and the complete diff before
-requires link validation and `git diff --check`. Run the Go validation whenever
+accepting a change. The status may contain only the intended source changes
-documentation changes commands, examples, generated output, or another
+during development. Reject credentials, private keys, environment files,
-behavior checked by the module.
+generated binaries, test or coverage output, downloaded assets, template
+residue, and any other artifact that does not belong in source control. The
+credential scan catches common forms but does not replace inspection of the
+actual change.
 
-## Focused Validation
+After committing the accepted change, require a clean candidate:
-
-Use focused checks while iterating, then run the complete validation sequence
-before accepting the change. The root package supports:
 
 ```sh
-go test .
+test -z "$(git status --porcelain)"
-go vet .
-go build .
 ```
-
-Filter tests by name without assuming a fixed internal package layout:
-
-```sh
-go test ./... -run 'TestName'
-```
-
-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.

docs/formats.md

@@ -9,9 +9,10 @@ explains how to select these sources and invoke the engine. The
 owns the resulting outbound wire behavior.
 
 Prompt and profile sources recursively discover files ending in `.yaml` or
-`.yml`. YAML decoding is strict: unknown fields are errors for the selected
+`.yml`. Each prompt-definition and profile file contains exactly one YAML
-definition. Definitions are selected by their YAML `id`, not their file name
+document; comments and trailing whitespace are allowed. YAML decoding is
-or directory.
+strict: unknown fields are errors for the selected definition. Definitions are
+selected by their YAML `id`, not their file name or directory.
 
 ## Prompt Definitions
 
@@ -86,9 +87,16 @@ Each message has a non-empty `role` and exactly one of:
 - `content`, containing an inline Go template; or
 - `content_file`, naming a file whose contents are the Go template.
 
-For directory and `fs.FS` prompt sources, `content_file` resolves relative to
+`content_file` must be a relative path. It resolves from the directory that
-the prompt file and remains within the source root. `WithPromptFile` also
+contains the prompt file and must remain within the configured prompt source
-resolves it relative to that file.
+root; parent components are allowed only when the resolved target remains
+inside that root. Absolute paths and paths that escape the root are rejected.
+Operating-system directory and single-file sources also reject symlink targets
+outside the root, while injected `fs.FS` sources apply containment in that
+filesystem's relative path namespace. For `WithPromptFile`, the source root is
+the directory containing the selected prompt file. Promptkit uses the parsed
+path text exactly after checking separately that it is not blank, so leading
+and trailing whitespace can name real filesystem entries.
 
 Request variables are the template data, so a variable named `audience` is
 referenced as `{{.audience}}`. The `{{input "note"}}` helper renders the body
@@ -118,7 +126,7 @@ outbound integration determines its wire representation.
 | `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. |
+| `repair_attempts` | no | Integer from zero through three; omitted means zero. A positive value requires `basic`, `json`, or `json_schema` validation. |
 
 The validation modes behave as follows:
 
@@ -128,14 +136,31 @@ The validation modes behave as follows:
 - `json_schema` requires valid JSON that satisfies the selected schema.
 
 `format` controls output artifact metadata. JSON Schema mode also supplies the
-schema to compatible model clients as structured-output metadata. The public
+schema to compatible model clients as structured-output metadata. Plain `json`
-engine does not install an output repairer, so its validation is single-pass
+validation accepts every valid JSON value and does not request a provider-native
-even when a positive `repair_attempts` value is present.
+JSON-object constraint.
+
+`repair_attempts` counts additional generation calls after a failed validation.
+Zero is single-pass. With a positive eligible budget, Promptkit stops at the
+first valid candidate. If the budget is exhausted, it returns the final
+candidate and its complete failed validation result; generation and operational
+validation failures remain errors. `none` never permits repair.
 
 A request-level `OutputContract` replaces the complete prompt output contract.
 It does not merge individual fields. If its format is empty, Promptkit uses
 `text`.
 
+## Built-In Backends
+
+Every engine provides these reserved OpenAI-compatible backend IDs. Consumers
+must not register either ID with `WithBackend`; exact registration and
+reservation behavior belongs to the [`Backend` GoDoc](../backends.go).
+
+| ID | Base endpoint | API-key environment variable | Active generation limit | Default queue capacity |
+| --- | --- | --- | ---: | ---: |
+| `openrouter` | `https://openrouter.ai/api/v1` | `OPENROUTER_API_KEY` | 16 | 1024 |
+| `rakestrawhome` | `https://inference.ai.rakestrawhome.com/v1` | `RAKESTRAWHOME_INFERENCE_API_KEY` | 4 | 1024 |
+
 ## Profile Definitions
 
 A profile supplies model execution settings:
@@ -154,11 +179,21 @@ extra_params:
   provider_option: enabled
 ```
 
+A derived profile can use a named base and override only the settings it owns:
+
+```yaml
+id: local-summary-fast
+base_profile: local-summary
+timeout_seconds: 30
+reasoning_effort: low
+```
+
 | Field | Required | Meaning |
 | --- | --- | --- |
-| `id` | yes | Non-empty profile identifier. IDs must be unique within one source. |
+| `id` | yes | Profile identifier, trimmed before selection and publication. It must be non-empty after trimming and unique within one source after normalization. |
+| `base_profile` | no | One optional parent profile ID. A derived profile may inherit target fields from it. |
 | `backend` | unless `endpoint` is present | Backend registry ID. It is trimmed and registry membership is checked when the profile is prepared or inspected. |
-| `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. |
+| `endpoint` | unless `backend` is present | OpenAI-compatible base URL, including an API version path when required. A nonempty value is trimmed and must be absolute HTTP or HTTPS with a host and without user information, a query, or a fragment. 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. |
@@ -166,16 +201,21 @@ extra_params:
 | `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. |
+| `api_key_env` | no | Optional environment-variable lookup source for an 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`.
 
+A standalone profile must provide a model and at least one of `backend` or
+`endpoint`. A derived profile may omit those target fields because its selected
+base chain can provide them. Local parsing still validates a derived profile's
+own ID, supplied endpoint, execution-setting bounds, and `extra_params`.
+
 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
+The engine always provides the built-in `openrouter` and `rakestrawhome` IDs.
-engine-scoped IDs with
+Consumers can add engine-scoped IDs with
 [`WithBackend`](../backends.go); exact registration validation belongs to its
 GoDoc.
 
@@ -183,30 +223,33 @@ GoDoc.
 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).
+Excessively deep or large JSON-shaped values are rejected for safety.
 
 ### Defaults And Overrides
 
 Execution settings resolve in this order:
 
-1. framework defaults;
+1. the framework timeout baseline;
 2. the selected backend, when the profile names one;
 3. the selected profile; and
 4. request `ExecutionTargetOverride` values.
 
-The framework defaults are:
+The framework baseline is:
 
 | Setting | Default |
 | --- | --- |
-| `temperature` | `0` |
+| `temperature` | Unspecified and omitted from compatible provider requests unless a profile or runtime override selects it. |
-| `max_tokens` | `0` |
+| `max_tokens` | Unspecified and omitted from compatible provider requests unless a profile or runtime override selects it. |
-| `top_p` | `1` |
+| `top_p` | Unspecified and omitted from compatible provider requests unless a profile or runtime override selects it. |
 | `timeout_seconds` | `600` |
 
-Numeric zero in a file or in-memory profile means that the profile does not
+Numeric zero in a file or in-memory profile does not select a numeric value.
-replace the framework default. Numeric request overrides use pointers, so an
+For `temperature`, `max_tokens`, and `top_p`, it leaves the provider control
-explicit zero is preserved. In particular, an explicit request
+unspecified. For `timeout_seconds`, it retains the framework deadline. Numeric
-`timeout_seconds` of zero disables the per-generation deadline while leaving
+request overrides use pointers, so an explicit zero is retained and sent to
-the caller context and transport timeout intact.
+compatible providers. 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
@@ -231,50 +274,79 @@ default.
 Profile sources resolve matching IDs in this order:
 
 1. in-memory profiles supplied with `WithProfiles`;
-2. a profile file, `fs.FS`, or configured profile directory; and
+2. the ordinary configured source selected by a profile file, `fs.FS`, or
-3. embedded built-in profiles.
+   configured profile directory;
+3. application fallback profiles supplied with `WithFallbackProfileFS`; and
+4. embedded built-in profiles.
 
-A higher-precedence source falls back only when the profile is absent. An
+A profile source supplies a complete definition; definitions and their fields
-invalid matching profile is an error and does not fall back. In-memory
+are not merged across sources. A higher-precedence source falls back only when
-`Profile` values follow the same ranges as YAML profiles. They use
+the requested profile ID is absent. An invalid matching profile is an error and
-`APIKeyRequired` for request-scoped credentials instead of `api_key_env`.
+does not fall back. In-memory `Profile` values follow the same ranges as YAML
-Preparation and exact profile inspection use this same source precedence.
+profiles. They use `APIKeyRequired` for request-scoped credentials instead of
+`api_key_env`. Preparation and exact profile inspection use this same source
+precedence.
+
+When a selected definition names `base_profile`, every profile ID in that
+chain is looked up through this same precedence order. A higher-precedence
+definition therefore shadows a lower-precedence definition of the same base
+ID, including a built-in. References are not source-qualified.
+
+### Profile Inheritance
+
+Promptkit resolves one linear base chain of at most 32 profiles, including the
+selected profile. It merges settings from the root base to the selected leaf.
+The leaf's `id` remains the selected profile identity. Nonblank string fields
+(`backend`, `endpoint`, `model`, `service_tier`, `reasoning_effort`, and
+`api_key_env`) and nonzero numeric fields replace inherited values. A nonempty
+`extra_params` map replaces the complete inherited map rather than merging
+keys, and `APIKeyRequired: true` remains true through the chain. Backend and
+endpoint are independent: replacing one does not clear the other.
+
+There is no profile-level clearing syntax. Blank strings, zero numbers, false,
+and empty maps remain unspecified and inherit from a base. Use existing
+presence-aware request overrides where an execution needs an explicit zero or
+empty reasoning setting.
+
+An absent directly selected profile reports the ordinary not-found error. Once
+the selected profile exists, a missing base, cycle, overlong chain, or
+incomplete resolved target is a profile-load failure. Ordinary operations
+resolve chains afresh; prepared execution retains the fully resolved target.
 
 ## Built-In Profile Catalog
 
-Every built-in selects the `openrouter` backend. The engine's built-in backend
+Every built-in profile selects one maintained built-in backend and inherits
-registry supplies `https://openrouter.ai/api/v1` and the environment-variable
+that backend's connection and credential metadata. Profile files do not repeat
-name `OPENROUTER_API_KEY`, so individual profiles contain only model and
+those values. A configured, application fallback, or in-memory profile with
-generation settings. Built-in profile files do not repeat those connection
+the same profile ID takes precedence.
-values. A custom or in-memory profile with the same profile ID takes
-precedence.
 
-| Provider | ID | Model |
+| Provider | ID | Backend | Model |
-| --- | --- | --- |
+| --- | --- | --- | --- |
-| aion-labs | `aion-2` | `aion-labs/aion-2.0` |
+| aion-labs | `aion-2` | `openrouter` | `aion-labs/aion-2.0` |
-| anthropic | `claude-fable-latest` | `~anthropic/claude-fable-latest` |
+| anthropic | `claude-fable-latest` | `openrouter` | `~anthropic/claude-fable-latest` |
-| anthropic | `claude-haiku-latest` | `~anthropic/claude-haiku-latest` |
+| anthropic | `claude-haiku-latest` | `openrouter` | `~anthropic/claude-haiku-latest` |
-| anthropic | `claude-opus-latest` | `~anthropic/claude-opus-latest` |
+| anthropic | `claude-opus-latest` | `openrouter` | `~anthropic/claude-opus-latest` |
-| anthropic | `claude-sonnet-latest` | `~anthropic/claude-sonnet-latest` |
+| anthropic | `claude-sonnet-latest` | `openrouter` | `~anthropic/claude-sonnet-latest` |
-| deepseek | `deepseek-3-2` | `deepseek/deepseek-v3.2` |
+| deepseek | `deepseek-3-2` | `openrouter` | `deepseek/deepseek-v3.2` |
-| deepseek | `deepseek-4-flash` | `deepseek/deepseek-v4-flash` |
+| deepseek | `deepseek-4-flash` | `openrouter` | `deepseek/deepseek-v4-flash` |
-| deepseek | `deepseek-4-pro` | `deepseek/deepseek-v4-pro` |
+| deepseek | `deepseek-4-pro` | `openrouter` | `deepseek/deepseek-v4-pro` |
-| google | `gemini-2-flash` | `google/gemini-2.5-flash` |
+| google | `gemini-2-flash` | `openrouter` | `google/gemini-2.5-flash` |
-| google | `gemini-2-flash-lite` | `google/gemini-2.5-flash-lite` |
+| google | `gemini-2-flash-lite` | `openrouter` | `google/gemini-2.5-flash-lite` |
-| google | `gemini-2-pro` | `google/gemini-2.5-pro` |
+| google | `gemini-2-pro` | `openrouter` | `google/gemini-2.5-pro` |
-| google | `gemini-3-flash-lite` | `google/gemini-3.1-flash-lite` |
+| google | `gemini-3-flash-lite` | `openrouter` | `google/gemini-3.1-flash-lite` |
-| google | `gemini-flash-latest` | `~google/gemini-flash-latest` |
+| google | `gemini-flash-latest` | `openrouter` | `~google/gemini-flash-latest` |
-| google | `gemini-pro-latest` | `~google/gemini-pro-latest` |
+| google | `gemini-pro-latest` | `openrouter` | `~google/gemini-pro-latest` |
-| google | `gemma-4-31b` | `google/gemma-4-31b-it:exacto` |
+| google | `gemma-4-31b` | `openrouter` | `google/gemma-4-31b-it:exacto` |
-| minimax | `minimax-m2` | `minimax/minimax-m2.5` |
+| google | `rakestrawhome-gemma-4-31b` | `rakestrawhome` | `google/gemma-4-31b-it` |
-| minimax | `minimax-m3` | `minimax/minimax-m3` |
+| minimax | `minimax-m2` | `openrouter` | `minimax/minimax-m2.5` |
-| mistral | `mistral-large-2512` | `mistralai/mistral-large-2512` |
+| minimax | `minimax-m3` | `openrouter` | `minimax/minimax-m3` |
-| mistral | `mistral-medium-3-5` | `mistralai/mistral-medium-3-5` |
+| mistral | `mistral-large-2512` | `openrouter` | `mistralai/mistral-large-2512` |
-| mistral | `mistral-small-3` | `mistralai/mistral-small-3.2-24b-instruct` |
+| mistral | `mistral-medium-3-5` | `openrouter` | `mistralai/mistral-medium-3-5` |
-| mistral | `mistral-small-4` | `mistralai/mistral-small-2603` |
+| mistral | `mistral-small-3` | `openrouter` | `mistralai/mistral-small-3.2-24b-instruct` |
-| nvidia | `nemotron-3-ultra` | `nvidia/nemotron-3-ultra-550b-a55b` |
+| mistral | `mistral-small-4` | `openrouter` | `mistralai/mistral-small-2603` |
-| openai | `gpt-5-mini` | `openai/gpt-5.4-mini` |
+| nvidia | `nemotron-3-ultra` | `openrouter` | `nvidia/nemotron-3-ultra-550b-a55b` |
-| openai | `gpt-5-nano` | `openai/gpt-5.4-nano` |
+| openai | `gpt-5-mini` | `openrouter` | `openai/gpt-5.4-mini` |
+| openai | `gpt-5-nano` | `openrouter` | `openai/gpt-5.4-nano` |
 
 ## Schemas
 
@@ -293,16 +365,25 @@ schema produces a failed validation result.
 Credential values belong at the request or environment boundary, never in
 prompt, profile, schema, or example files:
 
-- a file profile names an environment variable with `api_key_env`;
+- a backend or file profile can name an optional environment lookup source
-- an in-memory profile may set `APIKeyRequired`;
+  with `APIKeyEnv` or `api_key_env`;
-- a request can provide a direct `APIKey` or override `APIKeyEnv`; and
+- an in-memory profile may set `APIKeyRequired` as an explicit local
+  requirement;
+- a request can provide a direct `APIKey` or override the optional `APIKeyEnv`
+  source; 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.
+Named environment sources are optional: when the selected source is absent,
+empty, or whitespace-only, the built-in client omits the `Authorization`
+header and handles the provider response normally. `APIKeyRequired` is the
+only explicit local availability requirement. Promptkit validates required
+credential availability during preparation and rechecks it when a prepared
+execution runs. Injected clients receive resolved source metadata but define
+their own credential-resolution behavior.
 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.

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
+and request values into one execution target. Endpoint configuration is trimmed
-target overrides the client's configured base URL. After trailing slashes are
+and must be an absolute HTTP or HTTPS URL with a host and without user
-removed, `/chat/completions` is appended. Generation fails before sending when
+information, a query, or a fragment. A non-empty endpoint from the target
-neither source supplies an endpoint.
+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
+A usable API key supplied directly on the execution target takes precedence.
-precedence. Otherwise, when an API-key environment-variable name is supplied,
+Otherwise, when an API-key environment-variable name is supplied, the client
-the client reads that variable and requires a non-empty value. The selected
+reads and trims that variable. A bearer header is sent only when the resolved
-key is sent as `Authorization: Bearer <key>`. No authorization header is sent
+direct or environment credential is non-empty. When neither source is usable,
-when neither mechanism is configured.
+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
@@ -53,12 +61,14 @@ never also sent as a session header.
 
 The client conditionally includes:
 
-- `temperature`, `max_tokens`, and `top_p` when non-zero or explicitly
+- `temperature`, `max_tokens`, and `top_p` only when selected by a profile or
-  present;
+  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
@@ -81,13 +91,47 @@ request fields.
 
 ## Response Handling
 
-Any 2xx response is decoded as an OpenAI-compatible chat response. The client
+Any 2xx response body is limited to 16 MiB (16,777,216 bytes). A larger
-returns the first choice's non-empty message content and maps prompt,
+declared `Content-Length` is rejected before the body is read, and streamed,
-completion, total, cached, and cache-write token counts.
+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
+The bounded body must contain exactly one OpenAI-compatible JSON response
-responses. For a non-2xx status, the error includes the status code but never
+object followed only by JSON whitespace and EOF. The client returns the first
-the provider response body.
+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.
+
+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.
 
 ## Timeout And Cancellation
 
@@ -102,5 +146,7 @@ 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
+request. Caller cancellation retains `context.Canceled`; caller, generation,
-`http.Client`.
+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

@@ -56,7 +56,10 @@ reserve another bounded slot.
 
 `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.
+default repairer receive the same wrapper. Their requests retain the same
+effective backend, credential, numeric-presence metadata, and structured-output
+settings, so 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
@@ -71,7 +74,9 @@ 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.
+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.
 
 ## Cancellation And Release
 

docs/internal/llm.md

@@ -25,16 +25,21 @@ and request precedence. 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
+Construction trims and validates a nonempty configured base URL and clones any
-`http.Client` so Promptkit can apply its timeout default without mutating the
+supplied `http.Client` so Promptkit can apply its timeout default without
-caller's client. Generation then:
+mutating the caller's client. An empty configured base remains valid because a
+resolved request target may supply the endpoint. Generation 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;
+4. composes `/chat/completions` through parsed URL path operations;
-5. performs the outbound request under the applicable deadlines; and
+5. resolves authentication;
-6. decodes the first response choice and token usage.
+6. performs the outbound request under the applicable deadlines; and
+7. decodes one strictly framed, size-bounded successful response object and
+   maps its first choice and token usage, or decodes bounded structured
+   non-success detail.
 
 `internal/llm` owns the set of reserved OpenAI-compatible request fields used
 when validating extra parameters. Backend registration consumes the same rule
@@ -50,12 +55,14 @@ 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 that the frozen credential
+Before backend admission, the runner rechecks a frozen credential
-environment-variable name is available. The handle does not retain the
+environment-variable name only when the target explicitly requires a
-environment value; the model client resolves the value visible when generation
+credential. The handle does not retain the environment value; the model client
-begins. A direct request key remains in private execution state only until the
+resolves the value visible when generation begins. For optional sources with no
-claimed execution finishes or an unclaimed handle is discarded. Exact public
+usable value, the built-in client omits `Authorization` and continues to the
-ownership and redaction semantics belong 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
@@ -63,11 +70,47 @@ ownership and redaction semantics belong to the
 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
+Invalid nonempty configured endpoints are configuration failures. A missing or
-reported as request execution failures. The runner classifies these identities
+invalid final selected endpoint is an invalid generation request and is
-without depending on HTTP status mapping.
+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.
+
+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
 
@@ -75,7 +118,15 @@ 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
+cancellation, endpoint selection and composition, pre-transport rejection, and
-also verifies that resolved backend settings reach this client without
+bounded single-document successful-response framing, closure, and
-serializing backend identity. All use local test servers or test transports;
+response-body suppression. The focused
-the default suite makes no live or paid provider requests.
+[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/overview.md

@@ -11,23 +11,23 @@ contributor workflow and validation.
 
 | Component | Implemented responsibility | References |
 | --- | --- | --- |
-| Root `promptkit` package | Provides the supported engine facade, source, backend-registration, and injection options, public request, result, prompt-inspection, and profile-inspection values, opaque prepared-execution handles, profile construction, extension interfaces, value conversion, redacted formatting, typed capacity errors, public error mapping, and engine-local assembly. | [Package GoDoc](../../doc.go), [prepared execution](../../prepared_execution.go), [backend API](../../backends.go), [engine assembly](../../engine.go) |
+| Root `promptkit` package | Provides the supported engine facade, source, backend-registration, and injection options, public request, result, prompt-inspection, and profile-inspection values, opaque prepared-execution handles, profile construction, extension interfaces, value conversion, redacted formatting, typed capacity and generation error mapping, engine-local profile-source assembly including application fallbacks, and bounded output-repair assembly. | [Package GoDoc](../../doc.go), [prepared execution](../../prepared_execution.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/backend` | Constructs each engine's immutable registry from the maintained built-in definitions 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 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, profiles, execution targets, rendering, generation, and validation. | [Domain declarations](../../internal/domain/domain.go) |
+| `internal/domain` | Defines internal framework values for requests, artifacts, prompt definitions, 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), [endpoint invariant](../../internal/domain/endpoint.go) |
 | `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 and prepared-schema trees while preserving supported concrete value types. | [JSON values](../../internal/jsonvalue/jsonvalue.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` | 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` | Loads strictly decoded, locally validated execution profiles from filesystem and `fs.FS` sources, overlays raw sources with error-preserving fallback, and resolves inherited profiles. | [Framework formats](../formats.md), [profile repositories](../../internal/profile/filesystem_repository.go), [internal sources](sources.md#profiles-and-built-ins) |
-| `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/profile/builtin` | Embeds the built-in profile catalog, whose entries select maintained built-in backends. | [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/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 and creates frozen validation plans for prepared execution. | [Framework formats](../formats.md#schemas), [internal sources and validation](sources.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 response decoding, authentication, deadline handling, and ownership of the OpenAI-compatible reserved request-field policy. | [Internal model client](llm.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/usecase` | Resolves prompt definitions and hashes, profiles, backends, and targets for exact inspection and request settings for preparation, and coordinates ordinary execution and one-attempt prepared execution across internal sources, rendering, artifact loading, generation, validation, capacity, and optional repair. | [Internal runner](runner.md), [prepared-execution implementation](../../internal/usecase/prepared_execution.go) |
+| `internal/usecase` | Resolves prompt definitions and hashes, profiles, backends, and targets for exact inspection and request settings for preparation, and coordinates ordinary execution and one-attempt prepared execution across internal sources, rendering, artifact loading, operation-local validation plans, generation, capacity, and bounded repair. | [Internal runner](runner.md), [prepared-execution implementation](../../internal/usecase/prepared_execution.go) |
 
 The root package assembles these internal components without exposing their
 representations. Consumers depend only on the root facade.

docs/internal/runner.md

@@ -20,10 +20,13 @@ and override semantics consumed by the runner.
 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
+admitter, and a model client wrapped by the same capacity manager. Validation
-documents are loaded through the validator's optional schema-loader interface.
+plans and provider-facing schema metadata come from the validator's preparation
-An output repairer can be injected internally, but the ordinary runner
+interface.
-constructor does not enable one.
+The root engine supplies one default output repairer through the explicit
+runner constructor, using the same capacity-wrapped client 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.
@@ -69,14 +72,16 @@ performs only the work needed to validate routing and admission:
 5. resolve application-neutral defaults, backend defaults, profile values,
    and explicit request overrides in that order;
 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;
@@ -87,7 +92,10 @@ 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 stores the same
+kind of plan only in its private payload.
 
 Pointer-based numeric overrides preserve an explicit zero. Invalid negative or
 out-of-range values fail as invalid requests. Endpoint overrides do not change
@@ -118,8 +126,17 @@ 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
+and validates that artifact with the plan compiled during completion. Invalid
-result; an inability to perform validation is an operational error.
+generated content remains a validation result; an inability to generate or
+perform validation is an operational error.
+
+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
@@ -127,22 +144,37 @@ 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.
 
-When an internal repairer is present, a JSON or JSON Schema content failure can
+After a failed `basic`, JSON, or JSON Schema validation with a positive frozen
-trigger bounded repair attempts. Repair receives the effective execution
+budget, the installed repairer can make a bounded corrective call. Each request
-target and session ID, validation errors, prior output, and structured-output
+starts with a fresh copy of the complete original rendered messages, includes
-specification. The default repairer uses the same wrapped client as initial
+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, and structured-output specification.
+The same request constructor supplies those common fields to initial and repair
+generation. The default repairer uses the same wrapped client 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
+second bounded admission, and repaired outputs use the operation's existing
-option.
+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
+effective target and presence metadata, including backend identity and direct
-repair attempt.
+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 usage.
 
 ## Failure Categories
 
@@ -165,7 +197,8 @@ 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.
+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 source loading, 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,
@@ -177,8 +210,9 @@ 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,
+bounded repair progression, initial/repair request parity, cumulative usage,
-error categories, artifact metadata, usage, and timing. The
+shared initial/repair capacity, credentials and redaction, error categories,
+artifact metadata, and timing. The
 [capacity subsystem document](capacity.md) identifies the focused pool,
 waiter, and wrapped-client tests.
 

docs/internal/sources.md

@@ -12,9 +12,21 @@ validation modes, built-in catalog, and source precedence.
 
 ## Prompt Definitions
 
-`internal/promptdef` discovers YAML deterministically, decodes and validates
+`internal/promptdef` uses one source-neutral flow for prompt selection and
-definitions, selects an ID and optional version, and resolves file-backed
+normalization. That flow scans normalized YAML ID and version metadata,
-message content within the selected operating-system or `fs.FS` source.
+requires one strictly decoded document per file, classifies errors for the
+selected definition, detects duplicates, and normalizes the exact match.
+Small operating-system and `fs.FS` adapters own discovery, byte reads, display
+paths, content opening, and root containment. Each lookup remains a
+point-in-time scan: definitions and catalogs are not cached, and file-backed
+message content is opened only for the exact selected candidate.
+
+Operating-system sources enforce containment against canonical roots and
+targets so symlinks cannot escape. Injected `fs.FS` sources enforce containment
+in their clean relative path namespace. A single-file source uses the selected
+prompt file's containing directory as its root. Every content path must be
+relative and is opened from its exact parsed text after a separate blank check;
+contained parent components and whitespace-bearing names remain valid.
 
 Exact prompt inspection performs one point-in-time lookup through that same
 repository and validates referenced message content before returning declared
@@ -27,35 +39,69 @@ duplicate detection, and source containment:
 
 ## Profiles And Built-Ins
 
-`internal/profile` loads and validates execution profiles from an
+`internal/profile` loads, locally validates, overlays, and resolves execution
-operating-system filesystem or an `fs.FS`. It supports a primary repository
+profiles from an operating-system filesystem or an `fs.FS`. A file contains
-with fallback only when the primary reports that a profile is absent. Strict
+exactly one YAML document and its trimmed YAML `id` is its only selection
-YAML decoding recognizes the optional `backend` field, trims its value, and
+identity; filenames do not confer authority. Each point lookup reads discovered
-requires a model plus at least one non-blank backend or endpoint. Loading does
+files once for their metadata and reuses the selected file's bytes for strict
-not check registry membership because the available registry belongs to the
+decoding; unrelated profiles are not fully decoded. Strict selected decoding
+recognizes `base_profile` and the optional `backend` field, trims their values,
+and permits inherited target fields only when a base is named. File-backed
+`extra_params` values are validated and defensively copied through the shared
+bounded JSON-value owner before a profile is published. OpenAI-compatible
+reserved-field policy remains with the model-client and backend-registry owners.
+
+The overlay repository consults the next repository only when the
+higher-precedence repository reports that a profile is absent. A reliably
+selected malformed profile stops fallback, while an unrelated malformed file
+does not become authoritative through its filename. Loading does not check
+backend registry membership because the available registry belongs to the
 assembled engine; the runner checks membership during preparation and exact
 profile inspection.
 
-Exact profile inspection performs one point-in-time lookup through those
+The root engine assembles one raw composite catalog in precedence order:
-profile sources and checks the resolved target without reading prompt, input,
+in-memory profiles, one ordinary configured source, an application fallback
-or schema sources. It does not retain that lookup for a later execution.
+source, then the embedded built-in catalog. An explicit file or `fs.FS` profile
+source replaces `Config.ProfileDir` within the ordinary configured-source
+category. One outer resolving repository wraps that complete raw catalog, so
+each base lookup observes the same precedence and shadowing rules.
 
-`internal/profile/builtin` embeds the maintained built-in profile catalog and
+The resolving repository traverses every selected chain afresh, retains no
-can place a caller-selected repository ahead of that catalog. Every embedded
+cache, detects cycles, limits a chain to 32 profiles, merges root-to-leaf into a
-profile selects `openrouter` and inherits its endpoint and credential
+new caller-owned value, and validates the final target before publishing it. It
-environment-variable name from the built-in backend registry rather than
+does not check backend registry membership. Exact `base_profile` syntax, merge
-repeating those values. Profile behavior is owned by the
+rules, and consumer-visible failure behavior belong to the [framework format
-[profile repository tests](../../internal/profile/repository_test.go), while
+reference](../formats.md#profile-inheritance).
-catalog completeness, the backend-selection invariant, duplicate IDs, and
+
+Exact profile inspection performs one point-in-time resolved lookup through
+those profile sources and checks the final target without reading prompt, input,
+or schema sources. It does not retain that lookup for a later execution.
+Prepared execution instead freezes the fully resolved target; a later ordinary
+operation performs a fresh traversal.
+
+`internal/profile/builtin` embeds the maintained built-in profile catalog.
+Every embedded profile selects a maintained built-in backend and inherits that
+backend's endpoint and credential environment-variable name from the built-in
+backend registry rather than repeating those values. Profile loading and
 overlay behavior are owned by the
+[profile repository tests](../../internal/profile/repository_test.go), while
+catalog completeness, the backend-selection invariant, and duplicate IDs are
+owned by the
 [built-in repository tests](../../internal/profile/builtin/repository_test.go).
 
 ## Ordinary Artifacts
 
-`internal/artifact` resolves inline references and unrestricted,
+`internal/artifact` accepts explicitly typed inline references even when their
-caller-selected file paths. It copies content into an artifact, records
+body is empty. It also resolves unrestricted, caller-selected paths only when
-metadata and a content hash, applies a content-type fallback, and honors
+they identify regular operating-system files, checking that condition before
-context cancellation.
+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. 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
@@ -67,8 +113,17 @@ 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
+artifacts and variables. Within one render, each referenced artifact body is
-control into the rendered prompt. The
+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.
+
+Session and message parsing and execution remain synchronous. The renderer
+checks cancellation before and after each parse and execution boundary,
+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 carries
+message roles, session IDs, and cache control into the rendered prompt. The
 [renderer tests](../../internal/prompt/renderer_test.go) own rendering behavior.
 
 ## Schemas And Output Validation
@@ -78,22 +133,35 @@ 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.
 
-For executable preparation, the built-in validators create a frozen validation
+Every preparation operation creates one operation-local validation plan. None,
-plan. None, basic, and JSON modes retain the effective output contract without
+basic, and JSON modes retain the effective output contract without source
-source access. JSON Schema mode loads the root document, resolves and compiles
+access. JSON Schema mode loads the root document once, resolves and compiles
-every transitive reference during preparation, and retains the compiled
+each transitive reference, and retains the compiled validator. Schema compiler
-validator. The provider-facing structured-output metadata uses that same
+resources use canonical escaped file or private-scheme URLs; loaders decode
-captured root document.
+their paths once and enforce the configured source boundary. The
+provider-facing structured-output metadata uses the root document captured by
+the same plan.
 
-`PrepareExecution` also completes prompt and profile selection, artifact
+Schema preparation and execution remain synchronous. Promptkit checks
-loading and hashing, session and message rendering, and target resolution.
+cancellation before and after source resolution, JSON decoding, compilation,
-`RunPrepared` uses the retained source-derived state and validation plan; it
+and validation, and between bounded schema-read chunks. Once cancellation is
-does not reopen prompt, profile, input, or schema sources and does not rerender
+observed it returns the context error without publishing a partial plan or
-the request. By contrast, ordinary `Prepare` produces a preparation value only:
+validation result, even when a compiler or validator has just returned a
-a later `Run` performs its own source resolution and preparation.
+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 prompt, profile, input, or
+schema sources or rerendering the request. A later ordinary `Run` always
+performs fresh source resolution and preparation.
 
 The [validator tests](../../internal/validate/standard_validator_test.go) own
 basic, JSON, JSON Schema, source resolution, schema loading, compilation,
-frozen-reference behavior, and content-failure behavior. Prepared execution
+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/policy/architecture.md

@@ -19,10 +19,10 @@ results, public values, extension interfaces, profiles, and error sentinels.
 
 The implemented internal components consist of:
 
-- `internal/domain`, which owns framework data values shared by later internal
+- `internal/domain`, which owns framework data values and source-neutral
-  components;
+  invariants shared by later internal components;
 - `internal/backend`, which owns validated immutable OpenAI-compatible backend
-  definitions and the built-in OpenRouter definition;
+  definitions and the maintained built-in definitions;
 - `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
@@ -92,6 +92,13 @@ 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

docs/policy/testing.md

@@ -50,19 +50,18 @@ Examples of appropriate seams include clocks, randomness, subprocesses, remote A
 ## Test execution requirements
 
 Promptkit currently uses maintainer-run validation rather than hosted CI.
-Maintainers run the repository-documented test, vet, build, formatting,
+Maintainers run the complete local workflow in the
-documentation-link, and repository-hygiene checks before accepting changes.
+[development guide](../development.md#maintainer-validation) before accepting
+changes. That guide is the canonical owner of exact commands, formatting,
+documentation-link validation, and repository-hygiene checks.
 Introducing hosted CI later would supplement, not silently redefine, this
 documented validation model.
 
-The complete test sequence includes ordinary and race-enabled package tests.
+Maintainer validation must include ordinary and race-enabled package tests,
-The maintained offline consumer workflow is also run from the repository root:
+static analysis, a complete build, and execution of both maintained offline
-
+consumer examples. The preparation example protects assembled preparation and
-```sh
+inspection behavior. The execution example separately protects assembled
-go test ./...
+`Run`, injected-client, validation, usage, and result behavior.
-go test -race ./...
-go run ./examples/go-library/prepare
-```
 
 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 +87,9 @@ 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
+- The maintained offline preparation and execution examples protect distinct
-  assembled consumer workflow without contacting a model provider.
+  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

docs/release.md

@@ -106,48 +106,12 @@ gitea.maximumdirect.net/eric/promptkit 1.25.5
 promptkit gitea.maximumdirect.net/eric/promptkit
 ```
 
-Run the complete maintainer validation required by the
+As a release prerequisite, run the complete
-[development guide](development.md):
+[maintainer validation workflow](development.md#maintainer-validation) against
-
+the clean candidate. Do not substitute a partial command list: the development
-```sh
+guide owns the tests, race checks, analysis, build, both offline examples,
-go test ./...
+formatting, Markdown links, generated-output and credential review, and
-go test -race ./...
+repository hygiene. Record the successful workflow result with the candidate.
-go vet ./...
-go build ./...
-go run ./examples/go-library/prepare
-```
-
-Check every tracked Go file. This command must produce no output:
-
-```sh
-unformatted=$(
-    git ls-files '*.go' |
-        while IFS= read -r go_file
-        do
-            gofmt -l "$go_file"
-        done
-)
-test -z "$unformatted"
-```
-
-Follow every maintained Markdown link and confirm that its local or published
-target exists. Review the repository for generated binaries, test or coverage
-output, credentials, template residue, downloaded assets, and other files that
-do not belong in source control.
-
-Recheck module and repository hygiene, whitespace, and the clean checkout:
-
-```sh
-test -z "$(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)"
-```
 
 ## Write The Release Note
 

docs/releases/v0.4.0.md

@@ -126,9 +126,9 @@ 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 ordinary configured and built-in profile precedence and
+Inspection applies the engine's profile source precedence and resolves any
-resolves any selected backend. It does not load a prompt, render content,
+selected backend. It does not load a prompt, render content, reserve capacity,
-reserve capacity, or contact a model.
+or contact a model.
 
 See the
 [profile-inspection consumer guide](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work)

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/roadmap/deferred.md

@@ -0,0 +1,61 @@
+# 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. Experience from downstream
+adoption should establish whether an engine-wide operation would add enough
+value to justify its broader contract.
+
+Promptkit could provide an explicit offline operation that discovers and
+structurally validates configured prompt, profile, and schema sources without
+model generation. The normal `NewEngine` path would remain lazy.
+
+An eager operation would need coherent handling for duplicate prompt IDs and
+versions, strict YAML decoding, referenced content files, profile/backend
+membership, 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.
 
@@ -33,7 +38,8 @@ consumers.
 
 ## Ideas
 
-No ideas currently await selection.
+No ideas are currently awaiting selection. Active feature work belongs in its
+focused roadmap rather than this catalog.
 
 ## Entry Format
 

docs/roadmap/notarius-promptkit-wishlist.md

@@ -1,245 +0,0 @@
-# Notarius PromptKit Wishlist
-
-## Purpose
-
-This document records features and interface changes that would be useful
-additions to PromptKit from the perspective of the maintainers of Notarius, a
-downstream application that consumes PromptKit.
-
-PromptKit now provides the capabilities Notarius currently needs. The
-remaining deferred ideas are optional opportunities to improve checkpointing
-and operational observability.
-
-The examples are API sketches intended to communicate the desired capability,
-not prescriptive names or finalized Go contracts.
-
-## Priority 1: Atomic Execution With Prepared Details
-
-**Disposition:** Implemented through [`Engine.PrepareExecution` and
-`Engine.RunPrepared`](../../engine.go). See the
-[consumer guidance](../consumers/pkg-promptkit.md#prepare-now-and-execute-the-same-snapshot-later).
-A separate `RunDetailed` method is not cataloged.
-
-### Downstream need
-
-Notarius needs both:
-
-- the completed `RunResult`; and
-- the rendered messages, effective output contract, hashes, and other
-  preparation details exposed by `PreparedRun`.
-
-Notarius uses the prepared details to construct redaction-aware debug bundles
-and retain enough information to diagnose model behavior.
-
-### Implemented behavior
-
-Notarius can prepare one frozen execution snapshot, retain a caller-owned and
-credential-redacted `Details` value for its debug bundle, and execute the same
-snapshot through `RunPrepared`. The opaque handle is engine-bound and
-single-use; an unused handle can be released with `Discard`. The consumer
-guide and exported GoDoc own the exact lifecycle and failure contracts.
-
-### Value to Notarius
-
-This removes duplicate work from PromptKit-backed calls and ensures that
-retained debug material corresponds atomically to the actual execution.
-
-## Priority 2: Prompt-Independent Profile Inspection
-
-**Disposition:** Implemented as
-[`Engine.InspectProfile`](../../engine.go). See the
-[consumer guidance](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work).
-
-### Downstream need
-
-Notarius validates configured pipeline profile IDs before beginning a run. It
-needs to determine whether:
-
-- a profile exists;
-- its referenced backend is registered;
-- its execution target can be resolved; and
-- it declares a credential requirement that the application may need to
-  enforce.
-
-This validation should not require model generation.
-
-### Previous integration
-
-Before profile inspection was available, Notarius constructed a synthetic
-prompt using `testing/fstest.MapFS`, supplied a dummy transcript, and called
-`Engine.Prepare` solely to exercise profile and backend resolution.
-
-### Value to Notarius
-
-The implemented interface eliminates a synthetic production-only prompt
-fixture and establishes a direct, supported contract for configuration-time
-profile and backend validation.
-
-## Priority 3: Semantic Execution-Target Fingerprints
-
-**Disposition:** Deferred pending a separate semantic-equality design for
-resolved execution targets.
-
-### Downstream need
-
-Notarius checkpoints model-backed pipeline stages. A checkpoint must not be
-reused when generation-affecting PromptKit configuration changes.
-
-Notarius therefore needs a stable equality signal for the effective profile
-and backend target used by a pipeline.
-
-### Current integration
-
-Notarius currently constructs this identity itself from:
-
-- a manually maintained marker for the PromptKit release and built-in profile
-  catalog;
-- raw hashes of configured profile files; and
-- a separate hash of the configured conventional local-backend endpoint.
-
-This is safe but conservative and coupled to PromptKit details. Raw file
-hashing also invalidates checkpoints for semantically irrelevant YAML changes,
-such as comments or formatting.
-
-### Requested capability
-
-Expose an opaque semantic digest for a resolved profile and its effective
-generation target. It could be returned by the proposed profile-resolution
-API:
-
-```go
-type ResolvedProfile struct {
-	ProfileID       string
-	BackendID       string
-	EffectiveTarget ExecutionTarget
-	ExecutionDigest string
-}
-```
-
-Alternatively, PromptKit could expose a dedicated method such as
-`ProfileExecutionDigest(profileID)`.
-
-### Desired equality semantics
-
-The digest should change when generation-affecting state changes, including:
-
-- resolved model and endpoint;
-- backend routing identity;
-- backend request defaults and extra parameters;
-- profile generation parameters; and
-- the semantic identity of any selected built-in profile.
-
-The digest should not incorporate:
-
-- credential values;
-- concurrency or queue capacity;
-- filesystem source paths;
-- YAML comments or formatting; or
-- other settings that affect scheduling or source representation without
-  changing the generation target.
-
-The credential environment-variable name may need to participate if changing
-it can select a materially different provider account or target. PromptKit
-should define this deliberately while continuing to exclude the resolved
-secret value.
-
-### Design considerations
-
-- Treat the digest as an opaque equality value rather than a public encoding
-  of internal structures.
-- Document which categories of change affect equality.
-- Include a versioned semantic marker internally so PromptKit can deliberately
-  invalidate old digests when its resolution semantics change.
-- Prefer a per-profile digest over a digest of every profile known to an
-  engine. Notarius generally knows which profiles a resolved pipeline uses.
-- Do not require consumers to know PromptKit's built-in catalog version.
-
-### Value to Notarius
-
-This would let Notarius remove its PromptKit release marker and raw
-profile-source fingerprinting, reduce unnecessary checkpoint invalidation, and
-delegate execution-target equality to the component that owns target
-resolution.
-
-## Priority 4: Structured Capacity Errors
-
-**Disposition:** Implemented behavior. See the consumer guide's
-[Handle Errors](../consumers/pkg-promptkit.md#handle-errors) section.
-
-### Downstream need
-
-Notarius translates PromptKit backend-capacity rejection into a
-provider-neutral application error. When multiple backends are active,
-operators would benefit from knowing which backend rejected admission without
-parsing an error string or exposing endpoint details.
-
-### Implemented behavior
-
-PromptKit now retains the broad capacity classification while allowing
-Notarius to obtain the selected backend ID without parsing diagnostic text.
-The consumer guide owns the application workflow, including retry and backoff
-policy.
-
-### Value to Notarius
-
-This improves operational diagnostics and future metrics while preserving
-the provider-neutral error boundary used by Notarius.
-
-## Capabilities PromptKit Already Provides Well
-
-The current PromptKit boundary is sufficient for Notarius's implemented
-behavior. In particular, PromptKit already provides:
-
-- filesystem, `fs.FS`, and programmatic prompt, profile, and schema sources;
-- offline preparation without model execution;
-- structured output and content validation;
-- direct session propagation;
-- tri-state per-run reasoning overrides;
-- selected profile, backend, model, endpoint, effective parameters, hashes, and
-  token-usage provenance;
-- endpoint-only profiles;
-- the conventional `local` backend helper;
-- arbitrary engine-scoped `Backend` registrations;
-- backend authentication environment names, extra parameters, concurrency
-  limits, and queue-capacity policies;
-- provider-client and artifact-reader extension interfaces;
-- context cancellation; and
-- useful public error sentinels, including profile absence and capacity
-  exhaustion.
-
-The wishlist does not imply that Notarius needs PromptKit to broaden its core
-responsibilities. It primarily asks for more direct access to information and
-operations that PromptKit already computes internally.
-
-## Responsibilities That Should Remain In Notarius
-
-The following concerns belong to the downstream application and should not
-move into PromptKit for the sake of Notarius:
-
-- pipeline staging, dependencies, and generated references;
-- application-wide scheduling across providers and backends;
-- module and validation retry policy;
-- checkpoints, resume, and recomputation;
-- durable run artifacts and manifests;
-- D&D prompts, schemas, extractors, validators, and normalizers;
-- Notarius configuration-file parsing and precedence;
-- domain-specific prompt-cache prefix policy; and
-- application-specific redaction, retention, and debug-bundle policy.
-
-PromptKit's complete `Backend` API already supports custom IDs, multiple local
-endpoints, authentication, extra parameters, and explicit queue policies.
-Whether Notarius exposes those capabilities in its own configuration is an
-application-policy decision, not an upstream PromptKit gap.
-
-## Suggested Upstream Sequence
-
-For downstream adoption and any remaining upstream work, the useful order is:
-
-1. Adopt prepared execution for atomic details and results.
-2. Add a semantic execution-target digest, preferably alongside profile
-   inspection.
-3. Use the implemented typed capacity error where backend admission diagnostics
-   are needed.
-
-The first removes the concrete execution workaround. The second would improve
-checkpoint correctness and reduce coupling. The third is operational polish.

docs/roadmap/weatherreporter-promptkit-wishlist.md

@@ -1,331 +0,0 @@
-# Weatherreporter PromptKit Wishlist
-
-## Purpose
-
-This document records features and interface changes that would be useful
-additions to PromptKit from the perspective of the maintainers of
-Weatherreporter, a downstream application planning to replace its Scriptorium
-CLI integration with PromptKit.
-
-PromptKit now provides the capabilities Weatherreporter needs for the
-migration. The remaining deferred ideas are optional opportunities to validate
-configuration earlier and improve durable failure diagnostics.
-
-The examples are API sketches intended to communicate the desired capability,
-not prescriptive names or finalized Go contracts. The related
-[Notarius PromptKit wishlist](notarius-promptkit-wishlist.md) proposes several
-overlapping features from another downstream consumer's perspective.
-
-## Priority 1: Executable Preparation Handles
-
-**Disposition:** Implemented as [`Engine.PrepareExecution` and
-`Engine.RunPrepared`](../../engine.go). See the
-[consumer guidance](../consumers/pkg-promptkit.md#prepare-now-and-execute-the-same-snapshot-later).
-
-### Downstream need
-
-Weatherreporter treats prompt preparation as a durable preflight boundary. It
-needs to:
-
-1. prepare the exact request that will be executed;
-2. persist a safe preparation record before starting the provider call; and
-3. execute without reloading or rerendering prompt, profile, schema, or input
-   sources.
-
-Persisting preflight before generation leaves useful evidence when a provider
-call fails or the process is interrupted during generation.
-
-### Implemented behavior
-
-Weatherreporter can prepare one frozen execution snapshot, persist a
-caller-owned and credential-redacted `Details` value, and execute that same
-snapshot through `RunPrepared`. The opaque handle is engine-bound and
-single-use; an unused handle can be released with `Discard`. The consumer
-guide and exported GoDoc own the exact lifecycle, credential, cancellation,
-and capacity contracts.
-
-### Value to Weatherreporter
-
-This preserves Weatherreporter's durable preflight behavior, removes duplicate
-work, eliminates the source-consistency window, and ensures that persisted
-provenance describes the actual execution.
-
-## Priority 2: Prompt-Definition Inspection
-
-**Disposition:** Implemented as
-[`Engine.InspectPrompt`](../../engine.go). See the
-[consumer guidance](../consumers/pkg-promptkit.md#inspect-a-prompt-before-preparation).
-
-### Downstream need
-
-Weatherreporter has a fixed registry of seven report definitions. Each report
-selects a prompt ID and one of two output workflows:
-
-- direct Markdown; or
-- structured generated text followed by application-owned domain validation
-  and Markdown template rendering.
-
-Weatherreporter will embed the PromptKit prompt definitions and private
-response schemas that implement those reports. It needs to validate that the
-report registry and embedded prompt corpus agree before weather collection or
-provider execution.
-
-### Previous integration option
-
-Before prompt inspection was available, Weatherreporter could maintain
-synthetic data-package fixtures and call `Engine.Prepare` for every report
-prompt during tests. Runtime validation could also occur through the ordinary
-per-report preparation stage.
-
-This required complete placeholder inputs and profile resolution when the
-application primarily wanted to inspect prompt identity and declared contracts.
-
-### Value to Weatherreporter
-
-The implemented interface lets Weatherreporter directly verify that every
-report prompt exists, requires the curated `data_package` input, and declares
-the expected Markdown or JSON Schema output contract. It reduces synthetic
-test setup and moves failures ahead of weather collection.
-
-## Priority 3: Prompt-Independent Profile Inspection
-
-**Disposition:** Implemented as
-[`Engine.InspectProfile`](../../engine.go). See the
-[consumer guidance](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work).
-
-### Downstream need
-
-Weatherreporter will allow operators to select an external PromptKit profile
-source and may allow an explicit profile override. It needs to reject a missing
-profile, unknown backend, or malformed execution target before collecting
-weather data or writing report artifacts, and to apply its own policy to a
-reported credential requirement.
-
-### Value to Weatherreporter
-
-The implemented interface improves fail-fast configuration validation and gives
-operator-facing errors direct profile and backend context. It remains optional
-for the initial migration.
-
-## Priority 4: Eager Source Validation
-
-**Disposition:** Deferred until prompt and profile inspection have been used
-to determine whether a broader engine-wide validation operation is still
-needed.
-
-### Downstream need
-
-PromptKit deliberately defers reading and validating filesystem and `fs.FS`
-prompt, profile, and schema content until a request needs it. Weatherreporter
-has a small fixed embedded prompt corpus and one optional external profile
-source. It would benefit from an explicit offline validation operation for
-tests, startup diagnostics, and configuration checks.
-
-### Current integration option
-
-Weatherreporter can prepare every report prompt with fixture inputs and inspect
-any explicit profiles individually. That provides strong coverage but requires
-consumer-maintained traversal and synthetic material.
-
-### Requested capability
-
-Consider an opt-in source-validation operation:
-
-```go
-type SourceValidationOptions struct {
-	RequireCredentials bool
-}
-
-func (e *Engine) ValidateSources(
-	ctx context.Context,
-	opts SourceValidationOptions,
-) error
-```
-
-The operation should eagerly discover and structurally validate the configured
-prompt, profile, and schema sources without model generation.
-
-### Design considerations
-
-- Keep deferred validation as the normal `NewEngine` behavior.
-- Make eager validation an explicit consumer choice.
-- Validate duplicate IDs and versions, strict YAML decoding, referenced content
-  files, profile/backend membership, schema syntax, and schema references.
-- Distinguish structural credential declarations from current environment
-  availability.
-- Do not read or expose credential values when credential availability is not
-  requested.
-- Preserve source-specific public error identities and useful path context.
-- Respect context cancellation during filesystem discovery and schema work.
-- Consider whether exact prompt and profile inspection APIs already provide a
-  smaller sufficient surface before adding an engine-wide operation.
-
-### Value to Weatherreporter
-
-This would simplify offline corpus checks and catch malformed operator profile
-sources before report work begins. It is helpful but lower priority than exact
-prompt and profile inspection.
-
-## Priority 5: Structured Generation Errors
-
-**Disposition:** Deferred pending stronger downstream demand and a narrower
-design that does not duplicate prepared provenance or impose HTTP-specific
-fields on injected model clients.
-
-### Downstream need
-
-Weatherreporter preserves redacted, inspectable failure receipts for report
-runs. When model generation fails operationally, it needs to classify the
-failure and retain safe execution context without parsing error prose.
-
-Prompt preparation already supplies selected profile, backend, and model
-identity. Provider status classification would add useful operator context,
-especially when the built-in OpenAI-compatible client receives a non-success
-HTTP status.
-
-### Current integration option
-
-PromptKit exposes `ErrLLMGenerate` and preserves injected client errors through
-`errors.Is`. Weatherreporter can reliably classify generation failure and use
-its preparation record for profile, backend, and model provenance. Any further
-diagnostic detail remains a redacted error string.
-
-### Requested capability
-
-Consider a typed generation error that continues to match `ErrLLMGenerate`:
-
-```go
-type GenerationError struct {
-	BackendID string
-	Model     string
-	StatusCode int
-}
-```
-
-The exact fields may differ. The useful contract is safe structured context
-available through `errors.As`, while `errors.Is(err, ErrLLMGenerate)` remains
-compatible.
-
-### Design considerations
-
-- Include only fields that PromptKit knows reliably and can expose safely.
-- Treat an HTTP status as optional because injected model clients may not use
-  HTTP.
-- Do not expose provider response bodies, endpoints, credential environment
-  names, credential values, request content, or generated content.
-- Do not make a structured error a second source of prompt/profile provenance
-  already present in a prepared execution.
-- Preserve injected client error identity.
-- Keep retry and backoff policy with the consuming application.
-
-### Value to Weatherreporter
-
-This would improve durable failure receipts and troubleshooting, particularly
-for built-in transport failures. It is not required if preparation details and
-the existing sentinel remain available.
-
-## Lower-Priority Shared Wishlist Items
-
-### Structured Capacity Errors
-
-**Disposition:** Implemented behavior. See the consumer guide's
-[Handle Errors](../consumers/pkg-promptkit.md#handle-errors) section.
-
-PromptKit now exposes the stable backend ID for capacity rejection without
-requiring Weatherreporter to parse error text.
-
-Weatherreporter currently generates batch reports sequentially and constructs
-one engine per invocation, so engine-local capacity exhaustion is unlikely in
-the initial design. The typed error becomes more valuable if report generation
-later becomes concurrent or PromptKit engines become longer-lived.
-It should not block adoption.
-
-### Semantic Execution-Target Fingerprints
-
-**Disposition:** Deferred pending a separate semantic-equality design for
-resolved execution targets.
-
-The semantic target digest proposed by the
-[Notarius wishlist](notarius-promptkit-wishlist.md#priority-3-semantic-execution-target-fingerprints)
-would provide a compact equality signal for audit metadata.
-
-Weatherreporter does not currently reuse LLM-dependent checkpoints. Its Recent
-Changes behavior compares deterministic module snapshots rather than generated
-reports, so the digest has no immediate cache-correctness role. Existing
-PromptKit result metadata is sufficient for the initial integration. A digest
-would still be useful provenance and future-proofing, but it is not a
-migration priority.
-
-## Capabilities PromptKit Already Provides Well
-
-PromptKit already provides the essential Weatherreporter integration surface:
-
-- importable in-process engine construction;
-- filesystem, `fs.FS`, and programmatic prompt, profile, and schema sources;
-- offline preparation without model execution;
-- prepared execution handles for a durable preflight boundary;
-- exact prompt and profile inspection;
-- versioned prompt selection;
-- text, Markdown, JSON, and JSON Schema output contracts;
-- single-pass output validation with raw output retained after completed
-  validation failure;
-- inline input artifacts with provenance URIs and input hashes;
-- selected profile, backend, model, effective target, prompt hashes, timing,
-  and token-usage provenance;
-- endpoint-only profiles and engine-scoped backend registration;
-- injected model-client and artifact-reader interfaces;
-- caller cancellation, generation timeout, and transport timeout behavior; and
-- public error sentinels for configuration, prompt, profile, artifact,
-  validation, capacity, and generation failures; and
-- structured backend identity for capacity rejection.
-
-These capabilities are sufficient for Weatherreporter to adopt PromptKit
-without waiting for new upstream work.
-
-## Responsibilities That Should Remain In Weatherreporter
-
-The following concerns belong to Weatherreporter and should not move into
-PromptKit:
-
-- report definitions, valid periods, batches, and output naming;
-- application prompt content and private report response schemas;
-- deterministic weather facts, modules, and Recent Changes;
-- curated `data_package` construction and persistence;
-- generated-text domain validation and Markdown template rendering;
-- managed artifact paths, atomic writes, metadata, and inspection commands;
-- preparation, execution, raw-output, and failure-receipt schemas;
-- CLI configuration loading and precedence;
-- debug enablement, redaction, placement, sensitivity, and retention;
-- distributor notification;
-- batch continuation and any future retry policy; and
-- application-level compatibility and migration policy.
-
-## Suggested Upstream Sequence
-
-For downstream adoption and any remaining upstream work, the useful order is:
-
-1. Adopt the implemented executable preparation handles.
-2. Consider eager source validation after evaluating whether the two exact
-   inspection APIs are sufficient.
-3. Add structured generation errors.
-4. Use structured capacity errors and consider semantic execution-target
-   fingerprints as lower-priority operational improvements.
-
-The first item removes the material integration workaround. Prompt and profile
-inspection improve fail-fast validation. The remaining items are optional
-ergonomic and diagnostic improvements.
-
-## Adoption Sequencing
-
-Weatherreporter should not wait for the deferred wishlist items. The current
-PromptKit interface is sufficient when Weatherreporter:
-
-- embeds immutable prompt and schema assets;
-- supplies immutable inline data-package bytes;
-- constructs one engine per CLI invocation;
-- prepares an execution, persists selected `Details`, and calls
-  `RunPrepared`; and
-- keeps PromptKit behind a weatherreporter-owned adapter contract.
-
-Prompt inspection, profile inspection, source validation, structured errors,
-capacity details, and semantic fingerprints should not gate adoption.

engine.go

@@ -53,9 +53,9 @@ var (
 	// an execution profile or resolve its backend, except for the profile
 	// not-found case represented by ErrProfileNotFound.
 	ErrProfileLoad = errors.New("failed to load execution profile")
-	// ErrAPIKeyEnvMissing identifies an APIKeyEnv whose environment variable is
+	// ErrAPIKeyEnvMissing identifies an explicitly required APIKeyEnv whose
-	// unset or empty when no direct RunRequest.APIKey takes precedence. Such an
+	// environment variable is unset or empty after direct RunRequest.APIKey
-	// error also matches ErrInvalidRequest.
+	// 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.
@@ -69,8 +69,9 @@ var (
 	// 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
+	// response. A built-in OpenAI-compatible non-2xx response is available as a
-	// through errors.Is.
+	// [GenerationError]. Errors returned by an injected LLMClient remain
+	// available through errors.Is.
 	ErrLLMGenerate = errors.New("failed to generate output")
 	// ErrValidation identifies an operational failure to load or compile a
 	// schema or validate output. A completed validation whose Status is
@@ -98,9 +99,10 @@ type Config struct {
 	// 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
+	// ProfileDir is an optional ordinary configured source whose profiles take
-	// embedded built-in profiles. An empty value selects only built-ins unless
+	// precedence over application fallback and embedded built-in profiles. An
-	// profile options are also supplied.
+	// empty value selects the lower-precedence sources unless a profile-source
+	// option supplies the ordinary source.
 	ProfileDir string
 	// SchemaDir is the root for JSON Schema files. An empty value uses the
 	// current directory. WithSchemaFS or WithSchemaFile replaces this source.
@@ -119,12 +121,12 @@ type Config struct {
 // 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,
+// each prompt-source, ordinary-profile-source, fallback-profile-source,
-// model-client, and artifact-reader category, the last non-nil valid option
+// in-memory-profile, schema-source, model-client, and artifact-reader
-// replaces earlier options in that category. WithBackend is the additive
+// category, the last non-nil valid option replaces earlier options in that
-// exception: unique registrations accumulate, and a repeated backend ID is an
+// category. WithBackend is the additive exception: unique registrations
-// error rather than a replacement. An invalid option fails construction even
+// accumulate, and a repeated backend ID is an error rather than a replacement.
-// if a later option would replace it.
+// An invalid option fails construction even if a later option would replace it.
 type Option interface {
 	apply(*engineOptions) error
 }
@@ -140,11 +142,13 @@ type engineOptions struct {
 	artifactReader        artifactadapter.Reader
 	promptDefs            promptdef.Repository
 	profiles              profile.Repository
+	fallbackProfiles      profile.Repository
 	memoryProfiles        profile.Repository
 	backends              []domain.Backend
 	validator             validate.Validator
 	promptSource          bool
 	profileSource         bool
+	fallbackProfileSource bool
 	memorySource          bool
 	validatorSource       bool
 	artifactSource        bool
@@ -216,7 +220,7 @@ func WithPromptFile(path string) Option {
 		if err != nil {
 			return err
 		}
-		options.promptDefs = promptdef.NewFSRepository(fsys, root)
+		options.promptDefs = promptdef.NewFileRepository(fsys, root, filepath.Dir(path))
 		options.promptSource = true
 		return nil
 	})
@@ -224,12 +228,12 @@ func WithPromptFile(path string) Option {
 
 // WithProfileFS loads execution profiles from fsys under root.
 //
-// Profiles from this source overlay built-in profiles. Profile YAML must use
+// Profiles from this ordinary configured source take precedence over
-// api_key_env for environment-based credentials; raw API keys are rejected.
+// application fallback and built-in profiles. Profile YAML must use api_key_env
-// fsys must be non-nil and root must be non-empty; otherwise NewEngine fails
+// for environment-based credentials; raw API keys are rejected. fsys must be
-// with ErrInvalidConfig. This option replaces Config.ProfileDir and earlier
+// non-nil and root must be non-empty; otherwise NewEngine fails with
-// file or FS profile-source options, but remains below WithProfiles in
+// ErrInvalidConfig. This option replaces Config.ProfileDir and earlier file or
-// precedence.
+// 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 {
@@ -246,11 +250,12 @@ func WithProfileFS(fsys fs.FS, root string) Option {
 
 // 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
+// The profile takes precedence over application fallback and built-in profiles.
-// environment-based credentials; raw API keys are rejected. path must name an
+// Profile YAML must use api_key_env for environment-based credentials; raw API
-// existing non-directory file when NewEngine applies the option. This option
+// keys are rejected. path must name an existing non-directory file when
-// replaces Config.ProfileDir and earlier file or FS profile-source options,
+// NewEngine applies the option. This option replaces Config.ProfileDir and
-// but remains below WithProfiles in precedence.
+// 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)
@@ -263,13 +268,48 @@ func WithProfileFile(path string) Option {
 	})
 }
 
-// WithProfiles configures in-memory profiles that take precedence over
+// WithFallbackProfileFS supplies application-owned fallback profile
-// configured profile files and built-in profiles.
+// definitions from fsys under root.
 //
-// NewEngine validates and copies every profile. IDs must be unique within one
+// Profile lookup checks, in order, profiles supplied by WithProfiles; the
-// call. An invalid profile, duplicate ID, or unsupported ExtraParams value
+// ordinary configured source selected by WithProfileFile, WithProfileFS, or
-// makes construction fail with ErrInvalidConfig. Repeating WithProfiles
+// Config.ProfileDir; this fallback source; and Promptkit's embedded built-in
-// replaces the complete earlier in-memory set rather than merging it.
+// profiles. Each source supplies a complete profile definition; profile fields
+// are not merged between sources. Only an absent profile ID proceeds to the
+// next source. A matching read, parse, duplicate, validation, or credential
+// format failure stops resolution.
+//
+// Files use the ordinary strict profile YAML and api_key_env credential rules.
+// Loading and validation are lazy: NewEngine validates this option's arguments
+// but does not read profile files. fsys must be non-nil and root must be
+// nonblank; otherwise NewEngine returns an error matching ErrInvalidConfig.
+// Repeating this option replaces the earlier valid fallback source.
+//
+// This option controls profile-definition lookup, not provider or generation
+// failover.
+func WithFallbackProfileFS(fsys fs.FS, root string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		if fsys == nil {
+			return ErrInvalidConfig
+		}
+		if strings.TrimSpace(root) == "" {
+			return ErrInvalidConfig
+		}
+		options.fallbackProfiles = profile.NewFSRepository(fsys, root)
+		options.fallbackProfileSource = true
+		return nil
+	})
+}
+
+// WithProfiles configures in-memory profiles that take precedence over
+// ordinary configured, application fallback, and built-in profiles.
+//
+// NewEngine locally validates and copies every profile. IDs must be unique
+// within one call. An invalid local definition, duplicate ID, or unsupported
+// ExtraParams value makes construction fail with ErrInvalidConfig. A derived
+// profile's base reference and resolved target completeness are checked when it
+// is selected or inspected. 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)
@@ -350,13 +390,7 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 		promptDefs = promptdef.NewFilesystemRepository(cfg.PromptDir)
 	}
 
-	profiles := builtin.NewRepositoryWithDirectory(cfg.ProfileDir)
+	profiles := newProfileRepository(cfg.ProfileDir, options)
-	if options.profileSource {
-		profiles = builtin.NewRepositoryWithPrimary(options.profiles)
-	}
-	if options.memorySource {
-		profiles = profile.NewOverlayRepository(options.memoryProfiles, profiles)
-	}
 
 	backendRegistry, err := backend.NewRegistry(options.backends)
 	if err != nil {
@@ -396,7 +430,7 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 	}
 
 	return &Engine{
-		runner: usecase.NewRunner(
+		runner: usecase.NewRunnerWithRepairer(
 			promptDefs,
 			profiles,
 			backendRegistry,
@@ -404,27 +438,47 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 			prompt.NewGoRenderer(),
 			llmClient,
 			validator,
+			usecase.NewDefaultOutputRepairer(llmClient),
 			capacityManager,
 		),
 	}, nil
 }
 
+func newProfileRepository(profileDir string, options engineOptions) profile.Repository {
+	repository := builtin.NewRepository()
+
+	if options.fallbackProfileSource {
+		repository = profile.NewOverlayRepository(options.fallbackProfiles, repository)
+	}
+
+	if options.profileSource {
+		repository = profile.NewOverlayRepository(options.profiles, repository)
+	} else if strings.TrimSpace(profileDir) != "" {
+		repository = profile.NewOverlayRepository(profile.NewFilesystemRepository(profileDir), repository)
+	}
+
+	if options.memorySource {
+		repository = profile.NewOverlayRepository(options.memoryProfiles, repository)
+	}
+
+	return profile.NewResolvingRepository(repository)
+}
+
 func fileSource(name string) (fs.FS, string, error) {
-	cleanName := strings.TrimSpace(name)
+	if strings.TrimSpace(name) == "" {
-	if cleanName == "" {
 		return nil, "", ErrInvalidConfig
 	}
-	dir := filepath.Dir(cleanName)
+	dir := filepath.Dir(name)
-	base := filepath.Base(cleanName)
+	base := filepath.Base(name)
-	if base == "." || base == string(filepath.Separator) || strings.TrimSpace(base) == "" {
+	if base == "." || base == string(filepath.Separator) {
 		return nil, "", ErrInvalidConfig
 	}
-	info, err := os.Stat(cleanName)
+	info, err := os.Stat(name)
 	if err != nil {
-		return nil, "", fmt.Errorf("%w: failed to access source file %q: %v", ErrInvalidConfig, cleanName, err)
+		return nil, "", fmt.Errorf("%w: failed to access source file %q: %v", ErrInvalidConfig, name, err)
 	}
 	if info.IsDir() {
-		return nil, "", fmt.Errorf("%w: source path %q must be a file", ErrInvalidConfig, cleanName)
+		return nil, "", fmt.Errorf("%w: source path %q must be a file", ErrInvalidConfig, name)
 	}
 	return os.DirFS(dir), filepath.ToSlash(base), nil
 }
@@ -481,10 +535,10 @@ func (e *Engine) InspectPrompt(
 //
 // InspectProfile trims surrounding whitespace from profileID and looks up the
 // resulting nonblank ID exactly and case-sensitively through the engine's
-// ordinary in-memory, configured-source, and built-in profile precedence. It
+// in-memory, ordinary configured-source, application fallback, and built-in
-// applies framework defaults, the selected backend, and then the selected
+// profile precedence. It applies the framework timeout baseline, selected
-// profile to EffectiveModelParams without a request override. BackendID is
+// backend, and then selected profile to EffectiveModelParams without a request
-// empty for an endpoint-only profile.
+// 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
@@ -586,21 +640,25 @@ func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*Prepare
 // generated output.
 //
 // A content-validation failure is a successful run whose
-// RunResult.Validation has Status ValidationFailed. An inability to perform
+// RunResult.Validation has Status ValidationFailed. When its output contract
-// validation returns an error matching ErrValidation and no partial result.
+// has a positive repair budget, a failed eligible validation can make bounded
-// The public Engine does not perform output repair, so validation is
+// additional model calls and stops at the first valid candidate. Exhaustion
-// single-pass even when OutputContract.RepairAttempts is positive.
+// 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.
 //
 // Run can return every error category documented by [Engine.Prepare], plus
 // ErrCapacityExceeded and ErrLLMGenerate. 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. Errors from injected clients remain
+// ErrInvalidRequest or ErrLLMGenerate. A built-in OpenAI-compatible non-2xx
-// available through errors.Is. Cancellation while waiting for model-generation
+// response is discoverable as [GenerationError]. Errors from injected clients
-// capacity matches both ErrLLMGenerate and the context error. Cancellation
+// remain available through errors.Is. Cancellation while waiting for
-// otherwise follows the active collaborator's documented behavior. A nil
+// model-generation capacity matches both ErrLLMGenerate and the context error.
-// Engine returns ErrInvalidConfig. Run returns no partial result on 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 {
 		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
@@ -631,16 +689,17 @@ func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
 //
 // The supplied context governs this execution attempt independently of the
 // preparation context. It covers credential revalidation, admission,
-// generation, validation, and any internal repair. Result timing begins after
+// generation, validation, and any bounded output repair. Result timing begins
-// the claim and excludes preparation and consumer-held delay.
+// 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 completed content-validation rejection is returned
+// ErrCapacityExceeded. A built-in OpenAI-compatible non-2xx response is
-// in RunResult, not as an operational error. An operational error returns no
+// discoverable as [GenerationError]. A completed content-validation rejection,
-// partial RunResult.
+// including repair exhaustion, is returned in RunResult, not as an operational
+// error. An operational error returns no partial RunResult.
 func (e *Engine) RunPrepared(ctx context.Context, prepared *PreparedExecution) (*RunResult, error) {
 	if e == nil || e.runner == nil {
 		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)

errors.go

@@ -6,6 +6,7 @@ import (
 	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/capacity"
+	"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"
@@ -21,6 +22,19 @@ func mapPublicError(err error) error {
 		return &CapacityError{BackendID: internalCapacityError.BackendID}
 	}
 	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
 	}

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"
 )
 
@@ -48,3 +49,27 @@ func TestMapPublicErrorTranslatesCapacityError(t *testing.T) {
 		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)
+	}
+}

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 +1,133 @@
+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(), generationErrorRunRequest())
+	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(), generationErrorRunRequest())
+	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 TestBuiltInRepairGenerationError(t *testing.T) {
+	const (
+		codeMarker    = "repair-code-marker"
+		typeMarker    = "repair-type-marker"
+		messageMarker = "repair-message-marker"
+	)
+	calls := 0
+	config := contractConfig(frameworkSchemaDir)
+	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 := generationErrorRunRequest()
+	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 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 := contractConfig(frameworkSchemaDir)
+	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 generationErrorRunRequest() promptkit.RunRequest {
+	return promptkit.RunRequest{
+		PromptID: frameworkMarkdownSummaryPromptID,
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline("Rin opens the gate."),
+			"glossary":   promptkit.Inline("gate: A guarded passage."),
+		},
+	}
+}

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)
+			}
+		})
+	}
+}

internal/artifact/reader.go

@@ -16,10 +16,12 @@ 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
+
 // Reader resolves artifact references into actual artifacts.
 type Reader interface {
 	Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error)
@@ -34,7 +36,7 @@ type CompositeReader struct {
 func NewCompositeReader() Reader {
 	return &CompositeReader{
 		inlineReader: &inlineReader{},
-		fileReader:   &fileReader{},
+		fileReader:   &fileReader{open: openArtifactFile},
 	}
 }
 
@@ -64,10 +66,6 @@ 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 +76,15 @@ 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 +97,71 @@ 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) {
+func openArtifactFile(path string) (artifactFile, error) {
-	file, err := os.Open(path)
+	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)
+	}
+
+	data := make([]byte, 0)
+	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)
+		}
 	}
 
 	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 +169,6 @@ 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_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

@@ -1,6 +1,7 @@
 package artifact
 
 import (
+	"bytes"
 	"context"
 	"errors"
 	"os"
@@ -11,54 +12,97 @@ import (
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 )
 
-func TestCompositeReader_Read(t *testing.T) {
+func TestCompositeReaderRejectsUnsupportedReferences(t *testing.T) {
-	reader := NewCompositeReader()
+	_, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{
-	ctx := context.Background()
-
-	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)
-		}
-	})
-
-	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)
-		}
-	})
-
-	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")
+		t.Fatalf("expected ErrUnsupportedRefType, got %v", err)
+	}
+}
+
+func TestCompositeReaderSourceParityAndOpaqueHashes(t *testing.T) {
+	reader := NewCompositeReader()
+	hashes := make(map[string]string)
+	tests := []struct {
+		name    string
+		content string
+	}{
+		{name: "empty", content: ""},
+		{name: "ordinary", content: "same content"},
+		{name: "changed", content: "changed content"},
+	}
+
+	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)
+			}
+
+			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,
+				},
+			}
+
+			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 +131,154 @@ func TestCompositeReaderCopiesInlineData(t *testing.T) {
 	}
 }
 
-func TestCompositeReaderHonorsCancellation(t *testing.T) {
+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}},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			cancel()
 
-	_, err := NewCompositeReader().Read(ctx, domain.ArtifactRef{
+			artifact, err := NewCompositeReader().Read(ctx, tc.ref)
-		Type: domain.ArtifactRefInline,
+			if artifact != nil || !errors.Is(err, context.Canceled) {
-		Body: "ignored",
+				t.Fatalf("artifact=%#v err=%v, want nil/context.Canceled", artifact, err)
+			}
 		})
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("expected context cancellation, got %v", err)
 	}
 }
 
-func TestFileReader_Read(t *testing.T) {
+func TestFileReaderFailuresAndMetadata(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)
-	}
-
 	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{
+		_, err := reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefFile})
-			Type: domain.ArtifactRefFile,
-			URI:  "",
-		}
-		_, err := reader.Read(ctx, ref)
 		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")
+		filePath := filepath.Join(t.TempDir(), "artifact.unknownextension")
-		if err := os.WriteFile(path, content, 0o600); err != nil {
+		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" {
+		if artifact.ContentType != "text/plain" {
-			t.Errorf("expected text/plain fallback, got %q", art.ContentType)
+			t.Fatalf("content type = %q", artifact.ContentType)
 		}
 	})
 }
+
+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 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

@@ -5,7 +5,6 @@ package backend
 import (
 	"errors"
 	"fmt"
-	"net/url"
 	"regexp"
 	"sort"
 	"strings"
@@ -19,11 +18,19 @@ const (
 	// OpenRouterID is the reserved ID of Promptkit's built-in OpenRouter
 	// backend.
 	OpenRouterID = "openrouter"
+	// RakestrawHomeID is the reserved ID of Promptkit's built-in Rakestrawhome
+	// backend.
+	RakestrawHomeID = "rakestrawhome"
 
 	openRouterEndpoint  = "https://openrouter.ai/api/v1"
 	openRouterAPIKeyEnv = "OPENROUTER_API_KEY"
 
 	openRouterConcurrencyLimit = 16
+
+	rakestrawHomeEndpoint  = "https://inference.ai.rakestrawhome.com/v1"
+	rakestrawHomeAPIKeyEnv = "RAKESTRAWHOME_INFERENCE_API_KEY"
+
+	rakestrawHomeConcurrencyLimit = 4
 	defaultQueueCapacity          = 1024
 )
 
@@ -37,20 +44,16 @@ type Registry struct {
 	backends map[string]domain.Backend
 }
 
-// NewRegistry constructs a registry containing the built-in OpenRouter
+// NewRegistry constructs a registry containing the built-in definitions
-// definition followed by the supplied additions. Every ID must be unique.
+// followed by the supplied additions. Every ID must be unique.
 func NewRegistry(additions []domain.Backend) (*Registry, error) {
+	builtIns := builtInBackends()
 	registry := &Registry{
-		backends: make(map[string]domain.Backend, len(additions)+1),
+		backends: make(map[string]domain.Backend, len(builtIns)+len(additions)),
 	}
 
-	definitions := make([]domain.Backend, 0, len(additions)+1)
+	definitions := make([]domain.Backend, 0, len(builtIns)+len(additions))
-	definitions = append(definitions, domain.Backend{
+	definitions = append(definitions, builtIns...)
-		ID:               OpenRouterID,
-		Endpoint:         openRouterEndpoint,
-		APIKeyEnv:        openRouterAPIKeyEnv,
-		ConcurrencyLimit: openRouterConcurrencyLimit,
-	})
 	definitions = append(definitions, additions...)
 
 	for _, definition := range definitions {
@@ -72,6 +75,23 @@ func NewRegistry(additions []domain.Backend) (*Registry, error) {
 	return registry, nil
 }
 
+func builtInBackends() []domain.Backend {
+	return []domain.Backend{
+		{
+			ID:               OpenRouterID,
+			Endpoint:         openRouterEndpoint,
+			APIKeyEnv:        openRouterAPIKeyEnv,
+			ConcurrencyLimit: openRouterConcurrencyLimit,
+		},
+		{
+			ID:               RakestrawHomeID,
+			Endpoint:         rakestrawHomeEndpoint,
+			APIKeyEnv:        rakestrawHomeAPIKeyEnv,
+			ConcurrencyLimit: rakestrawHomeConcurrencyLimit,
+		},
+	}
+}
+
 // GetBackend returns a defensive copy of the backend registered with id.
 func (r *Registry) GetBackend(id string) (domain.Backend, error) {
 	if r == nil {
@@ -109,10 +129,11 @@ func (r *Registry) CapacityPolicies() map[string]domain.BackendCapacityPolicy {
 }
 
 func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
-	definition.Endpoint = strings.TrimSpace(definition.Endpoint)
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(definition.Endpoint)
-	if err := validateEndpoint(definition.Endpoint); err != nil {
+	if err != nil {
 		return domain.Backend{}, fmt.Errorf("backend %q endpoint: %w", definition.ID, err)
 	}
+	definition.Endpoint = endpoint
 
 	definition.APIKeyEnv = strings.TrimSpace(definition.APIKeyEnv)
 	if definition.APIKeyEnv != "" && !environmentVariableName.MatchString(definition.APIKeyEnv) {
@@ -182,31 +203,3 @@ func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
 	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

@@ -11,32 +11,63 @@ import (
 
 const validEndpoint = "https://backend.example/v1"
 
-func TestRegistryIncludesExactOpenRouterDefinition(t *testing.T) {
+func TestRegistryIncludesExactBuiltInDefinitions(t *testing.T) {
 	registry, err := backend.NewRegistry(nil)
 	if err != nil {
 		t.Fatalf("construct registry: %v", err)
 	}
 
-	definition, err := registry.GetBackend(backend.OpenRouterID)
+	tests := []struct {
-	if err != nil {
+		name       string
-		t.Fatalf("look up OpenRouter: %v", err)
+		id         string
+		endpoint   string
+		apiKeyEnv  string
+		concurrent int
+	}{
+		{
+			name:       "OpenRouter",
+			id:         backend.OpenRouterID,
+			endpoint:   "https://openrouter.ai/api/v1",
+			apiKeyEnv:  "OPENROUTER_API_KEY",
+			concurrent: 16,
+		},
+		{
+			name:       "Rakestrawhome",
+			id:         backend.RakestrawHomeID,
+			endpoint:   "https://inference.ai.rakestrawhome.com/v1",
+			apiKeyEnv:  "RAKESTRAWHOME_INFERENCE_API_KEY",
+			concurrent: 4,
+		},
 	}
-	if definition.ID != "openrouter" ||
+	for _, tc := range tests {
-		definition.Endpoint != "https://openrouter.ai/api/v1" ||
+		t.Run(tc.name, func(t *testing.T) {
-		definition.APIKeyEnv != "OPENROUTER_API_KEY" ||
+			definition, err := registry.GetBackend(tc.id)
-		definition.ConcurrencyLimit != 16 ||
+			if err != nil {
+				t.Fatalf("look up built-in: %v", err)
+			}
+			if definition.ID != tc.id ||
+				definition.Endpoint != tc.endpoint ||
+				definition.APIKeyEnv != tc.apiKeyEnv ||
+				definition.ConcurrencyLimit != tc.concurrent ||
 				definition.QueueCapacity != 1024 ||
 				!definition.QueueCapacitySet ||
 				definition.ExtraParams != nil {
-		t.Fatalf("unexpected OpenRouter definition: %#v", definition)
+				t.Fatalf("unexpected built-in definition: %#v", definition)
 			}
+		})
+	}
+
 	policies := registry.CapacityPolicies()
-	if len(policies) != 1 ||
+	if len(policies) != 2 ||
-		policies["openrouter"] != (domain.BackendCapacityPolicy{
+		policies[backend.OpenRouterID] != (domain.BackendCapacityPolicy{
 			ConcurrencyLimit: 16,
 			QueueCapacity:    1024,
+		}) ||
+		policies[backend.RakestrawHomeID] != (domain.BackendCapacityPolicy{
+			ConcurrencyLimit: 4,
+			QueueCapacity:    1024,
 		}) {
-		t.Fatalf("unexpected OpenRouter capacity policies: %#v", policies)
+		t.Fatalf("unexpected built-in capacity policies: %#v", policies)
 	}
 }
 
@@ -109,11 +140,12 @@ func TestRegistryNormalizesUniqueAdditionsAndIsolatesMutations(t *testing.T) {
 	}
 
 	policies := registry.CapacityPolicies()
-	if len(policies) != 2 {
+	if len(policies) != 3 {
 		t.Fatalf("unexpected capacity policy count: %#v", policies)
 	}
 	policies["custom"] = domain.BackendCapacityPolicy{}
 	delete(policies, backend.OpenRouterID)
+	delete(policies, backend.RakestrawHomeID)
 	againPolicies := registry.CapacityPolicies()
 	if againPolicies["custom"] != (domain.BackendCapacityPolicy{
 		ConcurrencyLimit: 3,
@@ -122,7 +154,10 @@ func TestRegistryNormalizesUniqueAdditionsAndIsolatesMutations(t *testing.T) {
 		t.Fatalf("capacity policy map mutated registry state: %#v", againPolicies)
 	}
 	if _, ok := againPolicies[backend.OpenRouterID]; !ok {
-		t.Fatalf("capacity policy deletion mutated registry state: %#v", againPolicies)
+		t.Fatalf("OpenRouter capacity policy deletion mutated registry state: %#v", againPolicies)
+	}
+	if _, ok := againPolicies[backend.RakestrawHomeID]; !ok {
+		t.Fatalf("Rakestrawhome capacity policy deletion mutated registry state: %#v", againPolicies)
 	}
 }
 
@@ -242,11 +277,14 @@ func TestNewRegistryRejectsDuplicateIDs(t *testing.T) {
 		wantID    string
 	}{
 		{
-			name: "built-in collision after normalization",
+			name:      "OpenRouter collision after normalization",
-			additions: []domain.Backend{{
+			additions: []domain.Backend{{ID: " openrouter "}},
-				ID: " openrouter ",
+			wantID:    backend.OpenRouterID,
-			}},
+		},
-			wantID: "openrouter",
+		{
+			name:      "Rakestrawhome collision after normalization",
+			additions: []domain.Backend{{ID: " rakestrawhome "}},
+			wantID:    backend.RakestrawHomeID,
 		},
 		{
 			name: "consumer collision after normalization",

internal/defaults/defaults.go

@@ -14,21 +14,12 @@ const (
 	ContentTypeApplicationJSON = "application/json"
 	OpenAIChatCompletionsPath  = "/chat/completions"
 
-	ExecutionDefaultTemperature    = 0.0
-	ExecutionDefaultMaxTokens      = 0
-	ExecutionDefaultTopP           = 1.0
 	ExecutionDefaultTimeoutSeconds = 600
-)
-
-var (
 	LLMRequestTimeoutDefault       = 10 * time.Minute
 )
 
 func ExecutionTargetDefault() domain.ExecutionTarget {
 	return domain.ExecutionTarget{
-		Temperature:    ExecutionDefaultTemperature,
-		MaxTokens:      ExecutionDefaultMaxTokens,
-		TopP:           ExecutionDefaultTopP,
 		TimeoutSeconds: ExecutionDefaultTimeoutSeconds,
 	}
 }

internal/domain/domain.go

@@ -97,22 +97,22 @@ type RunResult struct {
 // PreparedRun contains pre-LLM execution state from the prepare/render phase.
 // It must never include resolved API key values, model output, or validation data.
 type PreparedRun struct {
-	PromptID             string                  `json:"prompt_id"`
+	PromptID             string
-	PromptVersion        string                  `json:"prompt_version,omitempty"`
+	PromptVersion        string
-	PromptHash           string                  `json:"prompt_hash,omitempty"`
+	PromptHash           string
-	SelectedProfileID    string                  `json:"selected_profile_id"`
+	SelectedProfileID    string
-	SelectedBackendID    string                  `json:"selected_backend_id,omitempty"`
+	SelectedBackendID    string
-	EffectiveModelParams ExecutionTarget         `json:"effective_model_params"`
+	EffectiveModelParams ExecutionTarget
-	TargetPresence       ExecutionTargetPresence `json:"-"`
+	TargetPresence       ExecutionTargetPresence
-	OutputContract       OutputContract          `json:"output_contract"`
+	OutputContract       OutputContract
-	StructuredOutput     *StructuredOutputSpec   `json:"structured_output,omitempty"`
+	StructuredOutput     *StructuredOutputSpec
-	InputHashes          map[string]string       `json:"input_hashes,omitempty"`
+	InputHashes          map[string]string
-	SessionID            string                  `json:"session_id,omitempty"`
+	SessionID            string
-	RenderedPromptHash   string                  `json:"rendered_prompt_hash"`
+	RenderedPromptHash   string
-	Messages             []RenderedMessage       `json:"messages"`
+	Messages             []RenderedMessage
-	StartTime            time.Time               `json:"start_time,omitempty"`
+	StartTime            time.Time
-	EndTime              time.Time               `json:"end_time,omitempty"`
+	EndTime              time.Time
-	DurationMS           int64                   `json:"duration_ms,omitempty"`
+	DurationMS           int64
 }
 
 // ArtifactRef represents a reference to an input artifact.
@@ -192,6 +192,7 @@ type BackendCapacityPolicy struct {
 // ExecutionProfile describes how and where to execute a model.
 type ExecutionProfile struct {
 	ID              string         `yaml:"id"`
+	BaseProfileID   string         `yaml:"base_profile"`
 	BackendID       string         `yaml:"backend"`
 	Endpoint        string         `yaml:"endpoint"`
 	Model           string         `yaml:"model"`

internal/domain/endpoint.go

@@ -0,0 +1,39 @@
+package domain
+
+import (
+	"errors"
+	"net/url"
+	"strings"
+)
+
+// NormalizeOpenAICompatibleBaseEndpoint trims and validates a source-neutral
+// OpenAI-compatible provider base endpoint.
+func NormalizeOpenAICompatibleBaseEndpoint(endpoint string) (string, error) {
+	endpoint = strings.TrimSpace(endpoint)
+	if endpoint == "" {
+		return "", errors.New("endpoint must not be blank")
+	}
+	if strings.Contains(endpoint, "#") {
+		return "", errors.New("endpoint must not contain a fragment")
+	}
+
+	parsed, err := url.Parse(endpoint)
+	if err != nil {
+		return "", errors.New("endpoint must be a valid URL")
+	}
+	parsed.Scheme = strings.ToLower(parsed.Scheme)
+	if parsed.Scheme != "http" && parsed.Scheme != "https" {
+		return "", errors.New("endpoint must use http or https")
+	}
+	if !parsed.IsAbs() || parsed.Hostname() == "" {
+		return "", errors.New("endpoint must be absolute and include a host")
+	}
+	if parsed.User != nil {
+		return "", errors.New("endpoint must not contain user information")
+	}
+	if parsed.RawQuery != "" || parsed.ForceQuery {
+		return "", errors.New("endpoint must not contain a query string")
+	}
+
+	return parsed.String(), nil
+}

internal/domain/endpoint_test.go

@@ -0,0 +1,47 @@
+package domain
+
+import "testing"
+
+func TestNormalizeOpenAICompatibleBaseEndpoint(t *testing.T) {
+	tests := []struct {
+		name     string
+		endpoint string
+		want     string
+		wantErr  bool
+	}{
+		{name: "http host", endpoint: "http://provider.example", want: "http://provider.example"},
+		{name: "https nested path and whitespace", endpoint: "  HTTPS://provider.example/api/openai/v1  ", want: "https://provider.example/api/openai/v1"},
+		{name: "IPv4 host and port", endpoint: "http://127.0.0.1:8080/v1", want: "http://127.0.0.1:8080/v1"},
+		{name: "IPv6 host and port", endpoint: "https://[::1]:8443/v1", want: "https://[::1]:8443/v1"},
+		{name: "repeated trailing slashes", endpoint: "https://provider.example/v1///", want: "https://provider.example/v1///"},
+		{name: "blank", endpoint: " \t\n ", wantErr: true},
+		{name: "relative path", endpoint: "/api/v1", wantErr: true},
+		{name: "scheme relative", endpoint: "//provider.example/v1", wantErr: true},
+		{name: "missing host", endpoint: "https:///v1", wantErr: true},
+		{name: "unsupported scheme", endpoint: "ftp://provider.example/v1", wantErr: true},
+		{name: "user information", endpoint: "https://user:secret@provider.example/v1", wantErr: true},
+		{name: "query", endpoint: "https://provider.example/v1?mode=chat", wantErr: true},
+		{name: "empty query", endpoint: "https://provider.example/v1?", wantErr: true},
+		{name: "fragment", endpoint: "https://provider.example/v1#chat", wantErr: true},
+		{name: "empty fragment", endpoint: "https://provider.example/v1#", wantErr: true},
+		{name: "malformed URL", endpoint: "https://provider.example/%zz", wantErr: true},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := NormalizeOpenAICompatibleBaseEndpoint(tc.endpoint)
+			if tc.wantErr {
+				if err == nil {
+					t.Fatalf("expected endpoint error, got %q", got)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("normalize endpoint: %v", err)
+			}
+			if got != tc.want {
+				t.Fatalf("normalized endpoint = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}

internal/domain/execution_settings.go

@@ -0,0 +1,34 @@
+package domain
+
+import (
+	"errors"
+	"math"
+	"time"
+)
+
+const maxExecutionTimeoutSeconds int64 = math.MaxInt64 / int64(time.Second)
+
+// ValidateExecutionTargetSettings validates source-neutral execution-setting
+// invariants on a resolved target.
+func ValidateExecutionTargetSettings(target ExecutionTarget) error {
+	if !isFinite(target.Temperature) || target.Temperature < 0 || target.Temperature > 2 {
+		return errors.New("temperature must be finite and between 0 and 2")
+	}
+	if target.MaxTokens < 0 {
+		return errors.New("max_tokens must be greater than or equal to 0")
+	}
+	if !isFinite(target.TopP) || target.TopP < 0 || target.TopP > 1 {
+		return errors.New("top_p must be finite and between 0 and 1")
+	}
+	if target.TimeoutSeconds < 0 {
+		return errors.New("timeout_seconds must be greater than or equal to 0")
+	}
+	if int64(target.TimeoutSeconds) > maxExecutionTimeoutSeconds {
+		return errors.New("timeout_seconds exceeds the maximum supported duration")
+	}
+	return nil
+}
+
+func isFinite(value float64) bool {
+	return !math.IsNaN(value) && !math.IsInf(value, 0)
+}

internal/domain/execution_settings_test.go

@@ -0,0 +1,73 @@
+package domain
+
+import (
+	"math"
+	"strconv"
+	"strings"
+	"testing"
+)
+
+func TestValidateExecutionTargetSettings(t *testing.T) {
+	valid := ExecutionTarget{
+		Temperature:    1,
+		MaxTokens:      1,
+		TopP:           0.5,
+		TimeoutSeconds: 1,
+	}
+
+	type testCase struct {
+		name    string
+		change  func(*ExecutionTarget)
+		wantErr string
+	}
+	tests := []testCase{
+		{name: "temperature lower boundary", change: func(v *ExecutionTarget) { v.Temperature = 0 }},
+		{name: "temperature finite lower neighbor", change: func(v *ExecutionTarget) { v.Temperature = math.Nextafter(0, 1) }},
+		{name: "temperature finite upper neighbor", change: func(v *ExecutionTarget) { v.Temperature = math.Nextafter(2, 0) }},
+		{name: "temperature upper boundary", change: func(v *ExecutionTarget) { v.Temperature = 2 }},
+		{name: "temperature below lower boundary", change: func(v *ExecutionTarget) { v.Temperature = math.Nextafter(0, math.Inf(-1)) }, wantErr: "temperature"},
+		{name: "temperature above upper boundary", change: func(v *ExecutionTarget) { v.Temperature = math.Nextafter(2, math.Inf(1)) }, wantErr: "temperature"},
+		{name: "temperature NaN", change: func(v *ExecutionTarget) { v.Temperature = math.NaN() }, wantErr: "temperature"},
+		{name: "temperature positive infinity", change: func(v *ExecutionTarget) { v.Temperature = math.Inf(1) }, wantErr: "temperature"},
+		{name: "temperature negative infinity", change: func(v *ExecutionTarget) { v.Temperature = math.Inf(-1) }, wantErr: "temperature"},
+		{name: "max tokens lower boundary", change: func(v *ExecutionTarget) { v.MaxTokens = 0 }},
+		{name: "max tokens finite neighbor", change: func(v *ExecutionTarget) { v.MaxTokens = 1 }},
+		{name: "max tokens below lower boundary", change: func(v *ExecutionTarget) { v.MaxTokens = -1 }, wantErr: "max_tokens"},
+		{name: "top p lower boundary", change: func(v *ExecutionTarget) { v.TopP = 0 }},
+		{name: "top p finite lower neighbor", change: func(v *ExecutionTarget) { v.TopP = math.Nextafter(0, 1) }},
+		{name: "top p finite upper neighbor", change: func(v *ExecutionTarget) { v.TopP = math.Nextafter(1, 0) }},
+		{name: "top p upper boundary", change: func(v *ExecutionTarget) { v.TopP = 1 }},
+		{name: "top p below lower boundary", change: func(v *ExecutionTarget) { v.TopP = math.Nextafter(0, math.Inf(-1)) }, wantErr: "top_p"},
+		{name: "top p above upper boundary", change: func(v *ExecutionTarget) { v.TopP = math.Nextafter(1, math.Inf(1)) }, wantErr: "top_p"},
+		{name: "top p NaN", change: func(v *ExecutionTarget) { v.TopP = math.NaN() }, wantErr: "top_p"},
+		{name: "top p positive infinity", change: func(v *ExecutionTarget) { v.TopP = math.Inf(1) }, wantErr: "top_p"},
+		{name: "top p negative infinity", change: func(v *ExecutionTarget) { v.TopP = math.Inf(-1) }, wantErr: "top_p"},
+		{name: "timeout lower boundary", change: func(v *ExecutionTarget) { v.TimeoutSeconds = 0 }},
+		{name: "timeout finite neighbor", change: func(v *ExecutionTarget) { v.TimeoutSeconds = 1 }},
+		{name: "timeout below lower boundary", change: func(v *ExecutionTarget) { v.TimeoutSeconds = -1 }, wantErr: "timeout_seconds"},
+	}
+	if strconv.IntSize == 64 {
+		durationLimit := maxExecutionTimeoutSeconds
+		tests = append(tests,
+			testCase{name: "timeout duration boundary", change: func(v *ExecutionTarget) { v.TimeoutSeconds = int(durationLimit) }},
+			testCase{name: "timeout above duration boundary", change: func(v *ExecutionTarget) { v.TimeoutSeconds = int(durationLimit) + 1 }, wantErr: "timeout_seconds"},
+		)
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			target := valid
+			tt.change(&target)
+			err := ValidateExecutionTargetSettings(target)
+			if tt.wantErr == "" {
+				if err != nil {
+					t.Fatalf("validate execution settings: %v", err)
+				}
+				return
+			}
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("error = %v, want diagnostic containing %q", err, tt.wantErr)
+			}
+		})
+	}
+}

internal/domain/output_contract.go

@@ -0,0 +1,38 @@
+package domain
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+)
+
+const maxOutputRepairAttempts = 3
+
+// ValidateOutputContract validates source-neutral output-contract invariants.
+func ValidateOutputContract(contract OutputContract) error {
+	switch contract.Format {
+	case FormatText, FormatMarkdown, FormatJSON:
+	default:
+		return fmt.Errorf("invalid output format: %q", contract.Format)
+	}
+
+	switch contract.ValidationMode {
+	case ValidationNone, ValidationBasic, ValidationJSON, ValidationJSONSchema:
+	default:
+		return fmt.Errorf("invalid validation mode: %q", contract.ValidationMode)
+	}
+
+	if contract.ValidationMode == ValidationJSONSchema && strings.TrimSpace(contract.SchemaPath) == "" {
+		return errors.New("schema_path is required when validation_mode is json_schema")
+	}
+	if contract.RepairAttempts < 0 {
+		return errors.New("repair_attempts must be greater than or equal to 0")
+	}
+	if contract.RepairAttempts > maxOutputRepairAttempts {
+		return fmt.Errorf("repair_attempts must be less than or equal to %d", maxOutputRepairAttempts)
+	}
+	if contract.ValidationMode == ValidationNone && contract.RepairAttempts > 0 {
+		return errors.New("repair_attempts requires basic, json, or json_schema validation")
+	}
+	return nil
+}

internal/domain/output_contract_test.go

@@ -0,0 +1,87 @@
+package domain
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestValidateOutputContract(t *testing.T) {
+	valid := OutputContract{
+		Format:         FormatText,
+		ValidationMode: ValidationNone,
+	}
+
+	tests := []struct {
+		name    string
+		change  func(*OutputContract)
+		wantErr string
+	}{
+		{name: "text format", change: func(c *OutputContract) { c.Format = FormatText }},
+		{name: "markdown format", change: func(c *OutputContract) { c.Format = FormatMarkdown }},
+		{name: "json format", change: func(c *OutputContract) { c.Format = FormatJSON }},
+		{name: "empty format", change: func(c *OutputContract) { c.Format = "" }, wantErr: "format"},
+		{name: "unsupported format", change: func(c *OutputContract) { c.Format = OutputFormat("binary") }, wantErr: "format"},
+		{name: "none validation", change: func(c *OutputContract) { c.ValidationMode = ValidationNone }},
+		{name: "basic validation", change: func(c *OutputContract) { c.ValidationMode = ValidationBasic }},
+		{name: "json validation", change: func(c *OutputContract) { c.ValidationMode = ValidationJSON }},
+		{name: "json schema validation", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSONSchema
+			c.SchemaPath = "schema.json"
+		}},
+		{name: "empty validation mode", change: func(c *OutputContract) { c.ValidationMode = "" }, wantErr: "validation mode"},
+		{name: "unsupported validation mode", change: func(c *OutputContract) { c.ValidationMode = ValidationMode("unknown") }, wantErr: "validation mode"},
+		{name: "negative repair attempts", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationBasic
+			c.RepairAttempts = -1
+		}, wantErr: "repair_attempts"},
+		{name: "zero repair attempts", change: func(c *OutputContract) { c.RepairAttempts = 0 }},
+		{name: "one repair attempt", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationBasic
+			c.RepairAttempts = 1
+		}},
+		{name: "maximum repair attempts", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSON
+			c.RepairAttempts = 3
+		}},
+		{name: "too many repair attempts", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSONSchema
+			c.SchemaPath = "schema.json"
+			c.RepairAttempts = 4
+		}, wantErr: "repair_attempts"},
+		{name: "none validation with repair attempts", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationNone
+			c.RepairAttempts = 1
+		}, wantErr: "repair_attempts"},
+		{name: "json schema empty path", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSONSchema
+			c.SchemaPath = ""
+		}, wantErr: "schema_path"},
+		{name: "json schema whitespace path", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSONSchema
+			c.SchemaPath = " \t "
+		}, wantErr: "schema_path"},
+		{name: "json schema nonblank path", change: func(c *OutputContract) {
+			c.ValidationMode = ValidationJSONSchema
+			c.SchemaPath = " schema.json "
+		}},
+		{name: "non-schema empty path", change: func(c *OutputContract) { c.SchemaPath = "" }},
+		{name: "non-schema populated path", change: func(c *OutputContract) { c.SchemaPath = "ignored.json" }},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			contract := valid
+			tt.change(&contract)
+			err := ValidateOutputContract(contract)
+			if tt.wantErr == "" {
+				if err != nil {
+					t.Fatalf("validate output contract: %v", err)
+				}
+				return
+			}
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("error = %v, want diagnostic containing %q", err, tt.wantErr)
+			}
+		})
+	}
+}

internal/domain/prepared_run_test.go

@@ -1,141 +0,0 @@
-package domain
-
-import (
-	"encoding/json"
-	"strings"
-	"testing"
-)
-
-func TestPreparedRunJSONDoesNotIncludeSecretValues(t *testing.T) {
-	const envName = "PROMPTKIT_TEST_API_KEY"
-	const secret = "super-secret-value"
-	t.Setenv(envName, secret)
-
-	prepared := PreparedRun{
-		PromptID:          "prompt.id",
-		PromptVersion:     "v1",
-		PromptHash:        "prompt-hash",
-		SelectedProfileID: "local-fast",
-		EffectiveModelParams: ExecutionTarget{
-			Endpoint:  "http://llm/v1",
-			Model:     "gpt-test",
-			APIKeyEnv: envName,
-			APIKey:    secret,
-		},
-		InputHashes:        map[string]string{"transcript": "hash-1"},
-		RenderedPromptHash: "rendered-hash",
-		Messages: []RenderedMessage{
-			{Role: "system", Content: "You are helpful."},
-			{Role: "user", Content: "Summarize this."},
-		},
-	}
-
-	b, err := json.Marshal(prepared)
-	if err != nil {
-		t.Fatalf("marshal failed: %v", err)
-	}
-
-	out := string(b)
-	if strings.Contains(out, secret) {
-		t.Fatalf("prepared run JSON unexpectedly contains secret value: %s", out)
-	}
-	if !strings.Contains(out, `"api_key_env":"`+envName+`"`) {
-		t.Fatalf("prepared run JSON should include api_key_env name: %s", out)
-	}
-
-	var top map[string]any
-	if err := json.Unmarshal(b, &top); err != nil {
-		t.Fatalf("unmarshal failed: %v", err)
-	}
-
-	for _, forbidden := range []string{"raw_output", "validation", "artifact"} {
-		if _, ok := top[forbidden]; ok {
-			t.Fatalf("prepared run JSON should not include %q", forbidden)
-		}
-	}
-}
-
-func TestPreparedRunJSONIncludesMessageCacheControlOnlyWhenPresent(t *testing.T) {
-	prepared := PreparedRun{
-		PromptID:          "prompt.id",
-		SelectedProfileID: "local-fast",
-		EffectiveModelParams: ExecutionTarget{
-			Endpoint: "http://llm/v1",
-			Model:    "gpt-test",
-		},
-		RenderedPromptHash: "rendered-hash",
-		Messages: []RenderedMessage{
-			{
-				Role:    "system",
-				Content: "You are helpful.",
-				CacheControl: &CacheControl{
-					Type: CacheControlEphemeral,
-					TTL:  "1h",
-				},
-			},
-			{Role: "user", Content: "Summarize this."},
-		},
-	}
-
-	b, err := json.Marshal(prepared)
-	if err != nil {
-		t.Fatalf("marshal failed: %v", err)
-	}
-
-	var decoded struct {
-		Messages []map[string]any `json:"messages"`
-	}
-	if err := json.Unmarshal(b, &decoded); err != nil {
-		t.Fatalf("unmarshal failed: %v", err)
-	}
-	if len(decoded.Messages) != 2 {
-		t.Fatalf("expected 2 messages, got %d", len(decoded.Messages))
-	}
-
-	cacheControl, ok := decoded.Messages[0]["cache_control"].(map[string]any)
-	if !ok {
-		t.Fatalf("expected cache_control on first message, got %#v", decoded.Messages[0])
-	}
-	if cacheControl["type"] != string(CacheControlEphemeral) || cacheControl["ttl"] != "1h" {
-		t.Fatalf("unexpected cache_control payload: %#v", cacheControl)
-	}
-	if _, ok := decoded.Messages[1]["cache_control"]; ok {
-		t.Fatalf("expected second message to omit cache_control, got %#v", decoded.Messages[1])
-	}
-}
-
-func TestPreparedRunJSONIncludesSessionIDOnlyWhenPresent(t *testing.T) {
-	prepared := PreparedRun{
-		PromptID:          "prompt.id",
-		SelectedProfileID: "local-fast",
-		EffectiveModelParams: ExecutionTarget{
-			Endpoint: "http://llm/v1",
-			Model:    "gpt-test",
-		},
-		SessionID:          "session-123",
-		RenderedPromptHash: "rendered-hash",
-		Messages:           []RenderedMessage{{Role: "user", Content: "Summarize this."}},
-	}
-
-	b, err := json.Marshal(prepared)
-	if err != nil {
-		t.Fatalf("marshal failed: %v", err)
-	}
-
-	var decoded map[string]any
-	if err := json.Unmarshal(b, &decoded); err != nil {
-		t.Fatalf("unmarshal failed: %v", err)
-	}
-	if decoded["session_id"] != "session-123" {
-		t.Fatalf("expected session_id in prepared run JSON, got %#v", decoded["session_id"])
-	}
-
-	prepared.SessionID = ""
-	b, err = json.Marshal(prepared)
-	if err != nil {
-		t.Fatalf("marshal failed: %v", err)
-	}
-	if strings.Contains(string(b), "session_id") {
-		t.Fatalf("expected empty session_id to be omitted, got %s", b)
-	}
-}

internal/domain/session.go

@@ -8,6 +8,9 @@ import (
 
 // NormalizeSessionID applies the shared session identifier rule.
 func NormalizeSessionID(raw string) (string, error) {
+	if !utf8.ValidString(raw) {
+		return "", fmt.Errorf("session_id must contain valid UTF-8")
+	}
 	normalized := strings.TrimSpace(raw)
 	if normalized == "" {
 		return "", nil

internal/domain/session_test.go

@@ -10,7 +10,7 @@ func TestNormalizeSessionID(t *testing.T) {
 		name            string
 		raw             string
 		want            string
-		wantErr bool
+		wantErrContains string
 	}{
 		{
 			name: "trims surrounding Unicode whitespace",
@@ -30,19 +30,22 @@ func TestNormalizeSessionID(t *testing.T) {
 		{
 			name:            "one Unicode code point over maximum is rejected",
 			raw:             strings.Repeat("界", SessionIDMaxLength+1),
-			wantErr: true,
+			wantErrContains: "exceeds maximum",
 		},
+		{name: "invalid UTF-8 before valid content", raw: string([]byte{0xff}) + "session", wantErrContains: "valid UTF-8"},
+		{name: "invalid UTF-8 within valid content", raw: "ses" + string([]byte{0xff}) + "sion", wantErrContains: "valid UTF-8"},
+		{name: "invalid UTF-8 after valid content", raw: "session" + string([]byte{0xff}), wantErrContains: "valid UTF-8"},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			got, err := NormalizeSessionID(tt.raw)
-			if tt.wantErr {
+			if tt.wantErrContains != "" {
 				if err == nil {
 					t.Fatal("expected normalization error")
 				}
-				if !strings.Contains(err.Error(), "exceeds maximum") {
+				if !strings.Contains(err.Error(), tt.wantErrContains) {
-					t.Fatalf("expected useful length diagnostic, got %v", err)
+					t.Fatalf("expected diagnostic containing %q, got %v", tt.wantErrContains, err)
 				}
 				return
 			}

internal/filecatalog/catalog.go

@@ -36,7 +36,8 @@ func FindYAMLFiles(ctx context.Context, root string) ([]string, error) {
 	return files, err
 }
 
-// FindFSYAMLFiles returns sorted paths for .yaml and .yml files under root in fsys.
+// FindFSYAMLFiles returns root itself when it names a file. For a directory
+// root, it returns sorted paths for .yaml and .yml files beneath that root.
 func FindFSYAMLFiles(ctx context.Context, fsys fs.FS, root string) ([]string, error) {
 	cleanRoot := CleanFSRoot(root)
 	var files []string
@@ -52,6 +53,10 @@ func FindFSYAMLFiles(ctx context.Context, fsys fs.FS, root string) ([]string, er
 		if d.IsDir() {
 			return nil
 		}
+		if name == cleanRoot {
+			files = append(files, name)
+			return nil
+		}
 		if !IsYAMLFile(d.Name()) {
 			return nil
 		}
@@ -71,10 +76,10 @@ func RelativePath(root string, filePath string) string {
 	return filepath.Clean(rel)
 }
 
-// CleanFSRoot normalizes a root path for use with fs.FS.
+// CleanFSRoot normalizes a root path for use with fs.FS while preserving
+// nonblank leading and trailing whitespace.
 func CleanFSRoot(root string) string {
-	root = strings.TrimSpace(root)
+	if strings.TrimSpace(root) == "" || root == "." {
-	if root == "" || root == "." {
 		return "."
 	}
 	return path.Clean(root)
@@ -97,22 +102,21 @@ func DisplayPath(root string, name string) string {
 // ResolveFSPath resolves userPath from baseDir and keeps it inside root.
 func ResolveFSPath(root string, baseDir string, userPath string) (string, string, error) {
 	cleanRoot := CleanFSRoot(root)
-	cleanBase := path.Clean(strings.TrimSpace(baseDir))
+	cleanBase := path.Clean(baseDir)
-	if cleanBase == "" {
+	if strings.TrimSpace(baseDir) == "" {
 		cleanBase = cleanRoot
 	}
 	if !containsFSPath(cleanRoot, cleanBase) {
 		return "", "", fmt.Errorf("base path %q is outside source root %q", cleanBase, cleanRoot)
 	}
 
-	cleanUserPath := strings.TrimSpace(userPath)
+	if strings.TrimSpace(userPath) == "" {
-	if cleanUserPath == "" {
 		return "", "", fmt.Errorf("path is required")
 	}
-	cleanUserPath = path.Clean(cleanUserPath)
+	if path.IsAbs(userPath) {
-	if path.IsAbs(cleanUserPath) {
 		return "", "", fmt.Errorf("path %q must be relative", userPath)
 	}
+	cleanUserPath := path.Clean(userPath)
 
 	resolved := path.Clean(path.Join(cleanBase, cleanUserPath))
 	if !containsFSPath(cleanRoot, resolved) {
@@ -130,13 +134,6 @@ func containsFSPath(root string, name string) bool {
 	return name == root || strings.HasPrefix(name, strings.TrimSuffix(root, "/")+"/")
 }
 
-// Stem strips .yaml or .yml from a file name.
-func Stem(name string) string {
-	name = strings.TrimSuffix(name, ".yaml")
-	name = strings.TrimSuffix(name, ".yml")
-	return name
-}
-
 func IsYAMLFile(name string) bool {
 	return strings.HasSuffix(name, ".yaml") || strings.HasSuffix(name, ".yml")
 }

internal/filecatalog/catalog_test.go

@@ -54,7 +54,7 @@ func TestFindFSYAMLFilesNestedSortedAndFiltered(t *testing.T) {
 		"other/ignored.yaml":        &fstest.MapFile{Data: []byte("id: ignored")},
 	}
 
-	got, err := FindFSYAMLFiles(context.Background(), fsys, " prompts ")
+	got, err := FindFSYAMLFiles(context.Background(), fsys, "prompts")
 	if err != nil {
 		t.Fatalf("expected no error, got %v", err)
 	}
@@ -98,8 +98,11 @@ func TestCleanFSRoot(t *testing.T) {
 		want string
 	}{
 		{name: "empty", root: "", want: "."},
+		{name: "whitespace only", root: " \t ", want: "."},
 		{name: "dot", root: ".", want: "."},
-		{name: "trimmed", root: " prompts/../profiles ", want: "profiles"},
+		{name: "cleaned", root: "prompts/../profiles", want: "profiles"},
+		{name: "leading whitespace preserved", root: " profiles", want: " profiles"},
+		{name: "trailing whitespace preserved", root: "profiles ", want: "profiles "},
 	}
 
 	for _, tc := range tests {
@@ -158,6 +161,22 @@ func TestResolveFSPath(t *testing.T) {
 			wantPath:    "prompts/shared/user.tmpl",
 			wantDisplay: "shared/user.tmpl",
 		},
+		{
+			name:        "leading whitespace preserved",
+			root:        "prompts",
+			baseDir:     "prompts/nested",
+			userPath:    " user.tmpl",
+			wantPath:    "prompts/nested/ user.tmpl",
+			wantDisplay: "nested/ user.tmpl",
+		},
+		{
+			name:        "trailing whitespace preserved",
+			root:        "prompts",
+			baseDir:     "prompts/nested",
+			userPath:    "user.tmpl ",
+			wantPath:    "prompts/nested/user.tmpl ",
+			wantDisplay: "nested/user.tmpl ",
+		},
 		{
 			name:     "escape rejected",
 			root:     "prompts",
@@ -218,26 +237,6 @@ func TestResolveFSPath(t *testing.T) {
 	}
 }
 
-func TestStemStripsYAMLExtensions(t *testing.T) {
-	tests := []struct {
-		name string
-		in   string
-		want string
-	}{
-		{name: "yaml", in: "prompt.yaml", want: "prompt"},
-		{name: "yml", in: "profile.yml", want: "profile"},
-		{name: "other", in: "file.txt", want: "file.txt"},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			if got := Stem(tc.in); got != tc.want {
-				t.Fatalf("expected %q, got %q", tc.want, got)
-			}
-		})
-	}
-}
-
 func TestIsYAMLFile(t *testing.T) {
 	tests := []struct {
 		name string

internal/jsonvalue/jsonvalue.go

@@ -1,5 +1,5 @@
-// Package jsonvalue validates and defensively copies JSON-compatible value
+// Package jsonvalue validates and defensively copies bounded JSON-compatible
-// trees used by configuration, request, and prepared-state boundaries.
+// value trees used by configuration, request, and prepared-state boundaries.
 package jsonvalue
 
 import (
@@ -8,29 +8,39 @@ import (
 	"math"
 	"reflect"
 	"sort"
-	"strconv"
 )
 
-const maxSafeJSONInteger = 1<<53 - 1
+const (
+	maxContainerDepth = 100
+	maxProducedNodes  = 100_000
+)
 
 type visit struct {
 	typ reflect.Type
 	ptr uintptr
 }
 
+type traversalState struct {
+	active        map[visit]struct{}
+	producedNodes int
+}
+
 // Copy validates and deeply copies a JSON-compatible value while preserving
-// compatible concrete map, slice, array, scalar, and number types.
+// compatible concrete map, slice, array, scalar, and number types. It rejects
+// cycles and values that exceed the package's traversal limits.
 func Copy(src any) (any, error) {
-	return copyValue(reflect.ValueOf(src), "value", make(map[visit]struct{}), true)
+	return copyValue(reflect.ValueOf(src), "value", newTraversalState(), true, 0)
 }
 
 // CopyMap validates and deeply copies an extra-parameter map while preserving
-// compatible concrete map, slice, array, scalar, and number types.
+// compatible concrete map, slice, array, scalar, and number types. It rejects
+// empty object keys, cycles, and values that exceed the package's traversal
+// limits.
 func CopyMap(src map[string]any) (map[string]any, error) {
 	if src == nil {
 		return nil, nil
 	}
-	copied, err := copyValue(reflect.ValueOf(src), "extra_params", make(map[visit]struct{}), false)
+	copied, err := copyValue(reflect.ValueOf(src), "extra_params", newTraversalState(), false, 0)
 	if err != nil {
 		return nil, err
 	}
@@ -44,71 +54,90 @@ func CopyMap(src map[string]any) (map[string]any, error) {
 func copyValue(
 	value reflect.Value,
 	path string,
-	seen map[visit]struct{},
+	state *traversalState,
 	allowEmptyMapKeys bool,
+	containerDepth int,
 ) (any, error) {
-	if !value.IsValid() {
+	resolved, cleanup, isNull, err := state.resolveIndirection(value, path)
+	if err != nil {
+		return nil, err
+	}
+	defer cleanup()
+	if isNull {
+		if err := state.produceNode(path); err != nil {
+			return nil, err
+		}
 		return nil, nil
 	}
-	if value.Kind() == reflect.Interface {
+	value = resolved
-		if value.IsNil() {
-			return nil, nil
-		}
-		return copyValue(value.Elem(), path, seen, allowEmptyMapKeys)
-	}
 	if !value.CanInterface() {
 		return nil, fmt.Errorf("%s: value cannot be copied", path)
 	}
 	if number, ok := value.Interface().(json.Number); ok {
-		if _, err := json.Marshal(number); err != nil {
+		if !validJSONNumber(number) {
 			return nil, fmt.Errorf("%s: invalid JSON number", path)
 		}
-		f, err := strconv.ParseFloat(number.String(), 64)
+		if err := state.produceNode(path); err != nil {
-		if err != nil || math.IsNaN(f) || math.IsInf(f, 0) {
+			return nil, err
-			return nil, fmt.Errorf("%s: invalid JSON number", path)
 		}
 		return number, nil
 	}
 
 	switch value.Kind() {
 	case reflect.Bool, reflect.String:
+		if err := state.produceNode(path); err != nil {
+			return nil, err
+		}
 		return value.Interface(), nil
 	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
-		if value.Int() < -maxSafeJSONInteger || value.Int() > maxSafeJSONInteger {
+		if err := state.produceNode(path); err != nil {
-			return nil, fmt.Errorf("%s: integer is outside the JSON-safe range", path)
+			return nil, err
 		}
 		return value.Interface(), nil
 	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
-		if value.Uint() > maxSafeJSONInteger {
+		if err := state.produceNode(path); err != nil {
-			return nil, fmt.Errorf("%s: integer is outside the JSON-safe range", path)
+			return nil, err
 		}
 		return value.Interface(), nil
 	case reflect.Float32, reflect.Float64:
-		number := value.Convert(reflect.TypeOf(float64(0))).Float()
+		number := value.Float()
 		if math.IsNaN(number) || math.IsInf(number, 0) {
 			return nil, fmt.Errorf("%s: floating-point value must be finite", path)
 		}
+		if err := state.produceNode(path); err != nil {
+			return nil, err
+		}
 		return value.Interface(), nil
-	case reflect.Pointer:
+	case reflect.Map:
 		if value.IsNil() {
+			if err := state.produceNode(path); err != nil {
+				return nil, err
+			}
 			return nil, nil
 		}
-		current := visit{typ: value.Type(), ptr: value.Pointer()}
+		nextDepth, err := state.enterContainer(path, containerDepth)
-		if _, ok := seen[current]; ok {
+		if err != nil {
-			return nil, fmt.Errorf("%s: cyclic value is not supported", path)
+			return nil, err
 		}
-		seen[current] = struct{}{}
+		return copyMapValue(value, path, state, allowEmptyMapKeys, nextDepth)
-		defer delete(seen, current)
-		return copyValue(value.Elem(), path, seen, allowEmptyMapKeys)
-	case reflect.Map:
-		return copyMapValue(value, path, seen, allowEmptyMapKeys)
 	case reflect.Slice:
 		if value.IsNil() {
+			if err := state.produceNode(path); err != nil {
+				return nil, err
+			}
 			return nil, nil
 		}
-		return copySequenceValue(value, path, seen, allowEmptyMapKeys)
+		nextDepth, err := state.enterContainer(path, containerDepth)
+		if err != nil {
+			return nil, err
+		}
+		return copySequenceValue(value, path, state, allowEmptyMapKeys, nextDepth)
 	case reflect.Array:
-		return copySequenceValue(value, path, seen, allowEmptyMapKeys)
+		nextDepth, err := state.enterContainer(path, containerDepth)
+		if err != nil {
+			return nil, err
+		}
+		return copySequenceValue(value, path, state, allowEmptyMapKeys, nextDepth)
 	default:
 		return nil, fmt.Errorf("%s: unsupported JSON value type %s", path, value.Type())
 	}
@@ -117,22 +146,26 @@ func copyValue(
 func copyMapValue(
 	value reflect.Value,
 	path string,
-	seen map[visit]struct{},
+	state *traversalState,
 	allowEmptyMapKeys bool,
+	containerDepth int,
 ) (any, error) {
-	if value.IsNil() {
-		return nil, nil
-	}
 	if value.Type().Key().Kind() != reflect.String {
 		return nil, fmt.Errorf("%s: map key type %s is not supported", path, value.Type().Key())
 	}
+	if err := state.produceNode(path); err != nil {
+		return nil, err
+	}
+	if err := state.ensureChildCapacity(path, value.Len()); err != nil {
+		return nil, err
+	}
 
 	current := visit{typ: value.Type(), ptr: value.Pointer()}
-	if _, ok := seen[current]; ok {
+	if _, ok := state.active[current]; ok {
 		return nil, fmt.Errorf("%s: cyclic value is not supported", path)
 	}
-	seen[current] = struct{}{}
+	state.active[current] = struct{}{}
-	defer delete(seen, current)
+	defer delete(state.active, current)
 
 	keys := value.MapKeys()
 	sort.Slice(keys, func(i, j int) bool {
@@ -152,7 +185,13 @@ func copyMapValue(
 		if name == "" && !allowEmptyMapKeys {
 			return nil, fmt.Errorf("%s: map key must not be empty", path)
 		}
-		copied, err := copyValue(value.MapIndex(key), path+"."+name, seen, allowEmptyMapKeys)
+		copied, err := copyValue(
+			value.MapIndex(key),
+			path+"."+name,
+			state,
+			allowEmptyMapKeys,
+			containerDepth,
+		)
 		if err != nil {
 			return nil, err
 		}
@@ -190,17 +229,25 @@ func copyMapValue(
 func copySequenceValue(
 	value reflect.Value,
 	path string,
-	seen map[visit]struct{},
+	state *traversalState,
 	allowEmptyMapKeys bool,
+	containerDepth int,
 ) (any, error) {
+	if err := state.produceNode(path); err != nil {
+		return nil, err
+	}
+	if err := state.ensureChildCapacity(path, value.Len()); err != nil {
+		return nil, err
+	}
+
 	var current visit
 	if value.Kind() == reflect.Slice {
 		current = visit{typ: value.Type(), ptr: value.Pointer()}
-		if _, ok := seen[current]; ok {
+		if _, ok := state.active[current]; ok {
 			return nil, fmt.Errorf("%s: cyclic value is not supported", path)
 		}
-		seen[current] = struct{}{}
+		state.active[current] = struct{}{}
-		defer delete(seen, current)
+		defer delete(state.active, current)
 	}
 
 	values := make([]any, value.Len())
@@ -210,8 +257,9 @@ func copySequenceValue(
 		copied, err := copyValue(
 			value.Index(i),
 			fmt.Sprintf("%s[%d]", path, i),
-			seen,
+			state,
 			allowEmptyMapKeys,
+			containerDepth,
 		)
 		if err != nil {
 			return nil, err
@@ -248,6 +296,73 @@ func copySequenceValue(
 	return out, nil
 }
 
+func validJSONNumber(number json.Number) bool {
+	var parsed json.Number
+	if err := json.Unmarshal([]byte(number.String()), &parsed); err != nil {
+		return false
+	}
+	return parsed.String() == number.String()
+}
+
+func newTraversalState() *traversalState {
+	return &traversalState{active: make(map[visit]struct{})}
+}
+
+func (state *traversalState) produceNode(path string) error {
+	if state.producedNodes >= maxProducedNodes {
+		return fmt.Errorf("%s: JSON value work limit exceeded", path)
+	}
+	state.producedNodes++
+	return nil
+}
+
+func (state *traversalState) enterContainer(path string, depth int) (int, error) {
+	depth++
+	if depth > maxContainerDepth {
+		return 0, fmt.Errorf("%s: JSON container depth limit exceeded", path)
+	}
+	return depth, nil
+}
+
+func (state *traversalState) ensureChildCapacity(path string, count int) error {
+	if count > maxProducedNodes-state.producedNodes {
+		return fmt.Errorf("%s: JSON value work limit exceeded", path)
+	}
+	return nil
+}
+
+func (state *traversalState) resolveIndirection(
+	value reflect.Value,
+	path string,
+) (reflect.Value, func(), bool, error) {
+	var visits []visit
+	cleanup := func() {
+		for _, current := range visits {
+			delete(state.active, current)
+		}
+	}
+
+	for value.IsValid() && (value.Kind() == reflect.Interface || value.Kind() == reflect.Pointer) {
+		if value.IsNil() {
+			return reflect.Value{}, cleanup, true, nil
+		}
+		if value.Kind() == reflect.Pointer {
+			current := visit{typ: value.Type(), ptr: value.Pointer()}
+			if _, ok := state.active[current]; ok {
+				cleanup()
+				return reflect.Value{}, nil, false, fmt.Errorf("%s: cyclic value is not supported", path)
+			}
+			state.active[current] = struct{}{}
+			visits = append(visits, current)
+		}
+		value = value.Elem()
+	}
+	if !value.IsValid() {
+		return reflect.Value{}, cleanup, true, nil
+	}
+	return value, cleanup, false, nil
+}
+
 func canAssignNil(typ reflect.Type) bool {
 	switch typ.Kind() {
 	case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Pointer, reflect.Slice:

internal/jsonvalue/jsonvalue_test.go

@@ -1,112 +1,332 @@
-package jsonvalue_test
+package jsonvalue
 
 import (
 	"encoding/json"
 	"math"
 	"reflect"
+	"strings"
 	"testing"
-
-	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
 )
 
-func TestCopyMapPreservesTypesAndIsolatesMutations(t *testing.T) {
+type (
-	nested := map[string]int{"limit": 2}
+	namedBool    bool
-	sequence := []string{"one", "two"}
+	namedString  string
-	input := map[string]any{
+	namedInt64   int64
-		"count":    int64(7),
+	namedUint64  uint64
-		"number":   json.Number("-1.25e+2"),
+	namedFloat32 float32
-		"nested":   nested,
+	namedFloat64 float64
-		"sequence": sequence,
+	namedKey     string
-	}
+	namedMap     map[namedKey]namedInt64
-
+	namedSlice   []namedString
-	copied, err := jsonvalue.CopyMap(input)
+	namedArray   [1]map[string]int
-	if err != nil {
+)
-		t.Fatalf("copy map: %v", err)
-	}
-	nested["limit"] = 99
-	sequence[0] = "changed"
-	input["added"] = true
-
-	if got, ok := copied["count"].(int64); !ok || got != 7 {
-		t.Fatalf("integer type or value changed: %#v", copied["count"])
-	}
-	if got, ok := copied["number"].(json.Number); !ok || got != "-1.25e+2" {
-		t.Fatalf("JSON number type or value changed: %#v", copied["number"])
-	}
-	if got := copied["nested"].(map[string]int)["limit"]; got != 2 {
-		t.Fatalf("nested map was not isolated: %d", got)
-	}
-	if got := copied["sequence"].([]string)[0]; got != "one" {
-		t.Fatalf("sequence was not isolated: %q", got)
-	}
-	if _, ok := copied["added"]; ok {
-		t.Fatalf("top-level map was not isolated: %#v", copied)
-	}
-}
-
-func TestCopyAllowsEmptyObjectKeysAndIsolatesMutations(t *testing.T) {
-	nested := map[string]any{"": []any{"original"}}
-
-	copiedValue, err := jsonvalue.Copy(nested)
-	if err != nil {
-		t.Fatalf("copy value: %v", err)
-	}
-	nested[""].([]any)[0] = "changed"
-
-	copied := copiedValue.(map[string]any)
-	if got := copied[""].([]any)[0]; got != "original" {
-		t.Fatalf("copied value was not isolated: %v", got)
-	}
-}
-
-func TestCopyMapRejectsInvalidValues(t *testing.T) {
-	cyclicMap := map[string]any{}
-	cyclicMap["self"] = cyclicMap
-	cyclicSlice := []any{nil}
-	cyclicSlice[0] = cyclicSlice
 
+func TestCopyPreservesSupportedScalarAndNumberTypes(t *testing.T) {
+	maxInt := int(^uint(0) >> 1)
+	minInt := -maxInt - 1
 	tests := []struct {
 		name  string
 		value any
 	}{
-		{name: "empty nested key", value: map[string]int{"": 1}},
+		{name: "bool", value: true},
-		{name: "non-string map key", value: map[int]string{1: "one"}},
+		{name: "named bool", value: namedBool(true)},
-		{name: "unsupported value", value: make(chan int)},
+		{name: "string", value: "value"},
-		{name: "cyclic map", value: cyclicMap},
+		{name: "named string", value: namedString("value")},
-		{name: "cyclic slice", value: cyclicSlice},
+		{name: "int", value: minInt},
-		{name: "NaN", value: math.NaN()},
+		{name: "int8", value: int8(-1 << 7)},
-		{name: "positive infinity", value: math.Inf(1)},
+		{name: "int16", value: int16(-1 << 15)},
-		{name: "unsafe signed integer", value: int64(1 << 53)},
+		{name: "int32", value: int32(-1 << 31)},
-		{name: "unsafe unsigned integer", value: uint64(1 << 53)},
+		{name: "int64", value: int64(-1 << 63)},
+		{name: "named int64", value: namedInt64(1<<63 - 1)},
+		{name: "uint", value: ^uint(0)},
+		{name: "uint8", value: ^uint8(0)},
+		{name: "uint16", value: ^uint16(0)},
+		{name: "uint32", value: ^uint32(0)},
+		{name: "uint64", value: ^uint64(0)},
+		{name: "uintptr", value: ^uintptr(0)},
+		{name: "named uint64", value: namedUint64(^uint64(0))},
+		{name: "float32", value: float32(1.25)},
+		{name: "float64", value: float64(-2.5e100)},
+		{name: "named float32", value: namedFloat32(3.5)},
+		{name: "named float64", value: namedFloat64(-4.5e200)},
+		{name: "JSON number integer", value: json.Number("18446744073709551615")},
+		{name: "JSON number fraction", value: json.Number("-1.25e+2")},
+		{name: "JSON number beyond float64", value: json.Number("1e9999")},
 	}
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			if _, err := jsonvalue.CopyMap(map[string]any{"value": tc.value}); err == nil {
+			got, err := Copy(tc.value)
+			if err != nil {
+				t.Fatalf("copy value: %v", err)
+			}
+			if !reflect.DeepEqual(got, tc.value) {
+				t.Fatalf("value or concrete type changed: got %#v (%T), want %#v (%T)", got, got, tc.value, tc.value)
+			}
+		})
+	}
+}
+
+func TestCopyRejectsInvalidNumbers(t *testing.T) {
+	tests := []struct {
+		name  string
+		value any
+	}{
+		{name: "float32 NaN", value: float32(math.NaN())},
+		{name: "float64 NaN", value: math.NaN()},
+		{name: "named float NaN", value: namedFloat64(math.NaN())},
+		{name: "positive infinity", value: math.Inf(1)},
+		{name: "negative infinity", value: math.Inf(-1)},
+		{name: "empty JSON number", value: json.Number("")},
+		{name: "leading zero JSON number", value: json.Number("01")},
+		{name: "leading plus JSON number", value: json.Number("+1")},
+		{name: "trailing decimal JSON number", value: json.Number("1.")},
+		{name: "leading decimal JSON number", value: json.Number(".1")},
+		{name: "non-number JSON number", value: json.Number("NaN")},
+		{name: "spaced JSON number", value: json.Number(" 1")},
+		{name: "quoted JSON number", value: json.Number(`"1"`)},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if _, err := Copy(tc.value); err == nil {
 				t.Fatal("expected validation error")
 			}
 		})
 	}
 }
 
-func TestCopyMapValidatesJSONNumberSyntaxAndRange(t *testing.T) {
+func TestCopyPreservesCompatibleCollectionsAndNilEmptyDistinctions(t *testing.T) {
-	for _, number := range []json.Number{"0", "-1", "1.25", "-1.25e+2"} {
+	collections := []struct {
-		t.Run("valid "+number.String(), func(t *testing.T) {
+		name  string
-			got, err := jsonvalue.CopyMap(map[string]any{"value": number})
+		value any
-			if err != nil {
+	}{
-				t.Fatalf("copy valid JSON number: %v", err)
+		{name: "unnamed map", value: map[string]int{"limit": 2}},
+		{name: "named map", value: namedMap{"limit": 2}},
+		{name: "unnamed slice", value: []string{"one", "two"}},
+		{name: "named slice", value: namedSlice{"one", "two"}},
+		{name: "unnamed array", value: [2]int{1, 2}},
+		{name: "named array", value: namedArray{{"limit": 2}}},
+		{name: "empty map", value: map[string]int{}},
+		{name: "empty named map", value: namedMap{}},
+		{name: "empty slice", value: []string{}},
+		{name: "empty named slice", value: namedSlice{}},
+		{name: "empty array", value: [0]string{}},
 	}
-			if !reflect.DeepEqual(got["value"], number) {
+
-				t.Fatalf("JSON number changed: got %#v want %#v", got["value"], number)
+	for _, tc := range collections {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := Copy(tc.value)
+			if err != nil {
+				t.Fatalf("copy collection: %v", err)
+			}
+			if !reflect.DeepEqual(got, tc.value) || reflect.TypeOf(got) != reflect.TypeOf(tc.value) {
+				t.Fatalf("collection changed: got %#v (%T), want %#v (%T)", got, got, tc.value, tc.value)
+			}
+			kind := reflect.ValueOf(got).Kind()
+			if (kind == reflect.Map || kind == reflect.Slice) && reflect.ValueOf(got).IsNil() {
+				t.Fatal("non-nil collection became nil")
 			}
 		})
 	}
 
-	for _, number := range []json.Number{"", "01", "+1", "1.", ".1", "1e9999", "not-a-number"} {
+	var nilMap map[string]int
-		t.Run("invalid "+number.String(), func(t *testing.T) {
+	var nilSlice []string
-			if _, err := jsonvalue.CopyMap(map[string]any{"value": number}); err == nil {
+	var nilPointer *namedInt64
-				t.Fatal("expected invalid JSON number error")
+	for _, value := range []any{nil, nilMap, nilSlice, nilPointer} {
+		got, err := Copy(value)
+		if err != nil {
+			t.Fatalf("copy null value: %v", err)
+		}
+		if got != nil {
+			t.Fatalf("null value became %#v (%T)", got, got)
+		}
+	}
+
+	gotNil, err := CopyMap(nil)
+	if err != nil || gotNil != nil {
+		t.Fatalf("nil CopyMap result = %#v, %v", gotNil, err)
+	}
+	gotEmpty, err := CopyMap(map[string]any{})
+	if err != nil || gotEmpty == nil || len(gotEmpty) != 0 {
+		t.Fatalf("empty CopyMap result = %#v, %v", gotEmpty, err)
+	}
+}
+
+func TestCopyHandlesIndirectionAndIsolatesNestedMutations(t *testing.T) {
+	integer := namedInt64(7)
+	nestedMap := namedMap{"limit": 2}
+	nestedSlice := namedSlice{"original"}
+	nestedArray := namedArray{{"limit": 3}}
+	shared := []any{map[string]int{"value": 4}}
+	input := map[string]any{
+		"integer": &integer,
+		"map":     nestedMap,
+		"slice":   nestedSlice,
+		"array":   nestedArray,
+		"first":   shared,
+		"second":  shared,
+	}
+
+	copiedValue, err := Copy(input)
+	if err != nil {
+		t.Fatalf("copy mixed tree: %v", err)
+	}
+	copied := copiedValue.(map[string]any)
+	nestedMap["limit"] = 20
+	nestedSlice[0] = "changed"
+	nestedArray[0]["limit"] = 30
+	shared[0].(map[string]int)["value"] = 40
+
+	if got, ok := copied["integer"].(namedInt64); !ok || got != 7 {
+		t.Fatalf("pointer target changed: %#v", copied["integer"])
+	}
+	if got := copied["map"].(namedMap)["limit"]; got != 2 {
+		t.Fatalf("nested map aliased input: %d", got)
+	}
+	if got := copied["slice"].(namedSlice)[0]; got != "original" {
+		t.Fatalf("nested slice aliased input: %q", got)
+	}
+	if got := copied["array"].(namedArray)[0]["limit"]; got != 3 {
+		t.Fatalf("nested array aliased input: %d", got)
+	}
+	first := copied["first"].([]any)
+	second := copied["second"].([]any)
+	if got := first[0].(map[string]int)["value"]; got != 4 {
+		t.Fatalf("shared child aliased input: %d", got)
+	}
+	first[0].(map[string]int)["value"] = 99
+	if got := second[0].(map[string]int)["value"]; got != 4 {
+		t.Fatalf("repeated acyclic value shared copied output: %d", got)
+	}
+}
+
+func TestCopyAndCopyMapApplyDistinctEmptyKeyRules(t *testing.T) {
+	nested := map[string]any{"": []any{"original"}}
+	copiedValue, err := Copy(nested)
+	if err != nil {
+		t.Fatalf("Copy rejected empty schema key: %v", err)
+	}
+	nested[""].([]any)[0] = "changed"
+	if got := copiedValue.(map[string]any)[""].([]any)[0]; got != "original" {
+		t.Fatalf("copied schema value was not isolated: %v", got)
+	}
+
+	_, err = CopyMap(map[string]any{"nested": map[string]any{"": true}})
+	if err == nil || !strings.Contains(err.Error(), "extra_params.nested") {
+		t.Fatalf("CopyMap empty-key error = %v", err)
+	}
+}
+
+func TestCopyRejectsUnsupportedValuesAndActiveCycles(t *testing.T) {
+	cyclicMap := map[string]any{}
+	cyclicMap["self"] = cyclicMap
+	cyclicSlice := []any{nil}
+	cyclicSlice[0] = cyclicSlice
+	var cyclicPointer any
+	cyclicPointer = &cyclicPointer
+
+	tests := []struct {
+		name     string
+		value    any
+		wantPath string
+	}{
+		{name: "non-string map key", value: map[int]string{1: "one"}, wantPath: "value"},
+		{name: "unsupported channel", value: make(chan int), wantPath: "value"},
+		{name: "deterministic map path", value: map[string]any{"z": make(chan int), "a": make(chan int)}, wantPath: "value.a"},
+		{name: "cyclic map", value: cyclicMap, wantPath: "value.self"},
+		{name: "cyclic slice", value: cyclicSlice, wantPath: "value[0]"},
+		{name: "cyclic pointer", value: cyclicPointer, wantPath: "value"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := Copy(tc.value)
+			if err == nil || !strings.Contains(err.Error(), tc.wantPath) {
+				t.Fatalf("error = %v, want structural path %q", err, tc.wantPath)
 			}
 		})
 	}
 }
+
+func TestCopyEnforcesContainerDepth(t *testing.T) {
+	tests := []struct {
+		name    string
+		depth   int
+		wantErr bool
+	}{
+		{name: "just below", depth: maxContainerDepth - 1},
+		{name: "at limit", depth: maxContainerDepth},
+		{name: "over limit", depth: maxContainerDepth + 1, wantErr: true},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := Copy(alternatingContainers(tc.depth))
+			if tc.wantErr {
+				if err == nil || !strings.HasPrefix(err.Error(), "value") || !strings.Contains(err.Error(), "container depth limit") {
+					t.Fatalf("depth error = %v", err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("copy depth %d: %v", tc.depth, err)
+			}
+		})
+	}
+}
+
+func TestCopyEnforcesProducedNodeBudgetForRepeatedAcyclicValues(t *testing.T) {
+	shared := []any{true}
+	sharedOccurrences := (maxProducedNodes - 2) / 2
+	justBelow := repeatedValues(shared, sharedOccurrences, 0)
+	atLimit := repeatedValues(shared, sharedOccurrences, 1)
+	overLimit := repeatedValues(shared, sharedOccurrences, 2)
+
+	for name, value := range map[string]any{
+		"just below": justBelow,
+		"at limit":   atLimit,
+	} {
+		t.Run(name, func(t *testing.T) {
+			if _, err := Copy(value); err != nil {
+				t.Fatalf("copy value within work budget: %v", err)
+			}
+		})
+	}
+
+	_, err := Copy(overLimit)
+	if err == nil || !strings.HasPrefix(err.Error(), "value[") || !strings.Contains(err.Error(), "value work limit") {
+		t.Fatalf("work-budget error = %v", err)
+	}
+
+	_, err = Copy(make([]any, maxProducedNodes))
+	if err == nil || !strings.HasPrefix(err.Error(), "value:") || !strings.Contains(err.Error(), "value work limit") {
+		t.Fatalf("flat work-budget error = %v", err)
+	}
+}
+
+func alternatingContainers(depth int) any {
+	var value any = true
+	for level := 0; level < depth; level++ {
+		switch level % 3 {
+		case 0:
+			value = map[string]any{"child": value}
+		case 1:
+			value = []any{value}
+		default:
+			value = [1]any{value}
+		}
+	}
+	return value
+}
+
+func repeatedValues(shared []any, occurrences, leadingScalars int) []any {
+	values := make([]any, 0, leadingScalars+occurrences)
+	for i := 0; i < leadingScalars; i++ {
+		values = append(values, false)
+	}
+	for i := 0; i < occurrences; i++ {
+		values = append(values, shared)
+	}
+	return values
+}

internal/llm/openai_compatible_client.go

@@ -25,6 +25,20 @@ var (
 	ErrMalformedResponse = errors.New("malformed llm response")
 )
 
+const maxOpenAIChatResponseBytes int64 = 16 << 20
+
+type requestFailedError struct {
+	cause error
+}
+
+func (e *requestFailedError) Error() string {
+	return ErrRequestFailed.Error()
+}
+
+func (e *requestFailedError) Unwrap() []error {
+	return []error{ErrRequestFailed, e.cause}
+}
+
 type OpenAICompatibleConfig struct {
 	BaseURL    string
 	Model      string
@@ -39,9 +53,11 @@ type OpenAICompatibleClient struct {
 }
 
 func NewOpenAICompatibleClient(cfg OpenAICompatibleConfig) (*OpenAICompatibleClient, error) {
-	baseURL := strings.TrimSpace(cfg.BaseURL)
+	baseURL := ""
-	if baseURL != "" {
+	if strings.TrimSpace(cfg.BaseURL) != "" {
-		if _, err := url.ParseRequestURI(baseURL); err != nil {
+		var err error
+		baseURL, err = domain.NormalizeOpenAICompatibleBaseEndpoint(cfg.BaseURL)
+		if err != nil {
 			return nil, fmt.Errorf("%w: invalid base URL: %v", ErrInvalidConfig, err)
 		}
 	}
@@ -63,25 +79,29 @@ func NewOpenAICompatibleClient(cfg OpenAICompatibleConfig) (*OpenAICompatibleCli
 	}
 
 	return &OpenAICompatibleClient{
-		baseURL:      strings.TrimRight(baseURL, "/"),
+		baseURL:      baseURL,
 		defaultModel: cfg.Model,
 		httpClient:   client,
 	}, nil
 }
 
 func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
-	if req.Target.TimeoutSeconds < 0 {
+	if err := domain.ValidateExecutionTargetSettings(req.Target); err != nil {
-		return nil, fmt.Errorf("%w: timeout_seconds must be greater than or equal to 0", ErrInvalidRequest)
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
 	}
 
-	endpoint := strings.TrimSpace(req.Target.Endpoint)
+	selectedEndpoint := req.Target.Endpoint
-	if endpoint == "" {
+	if strings.TrimSpace(selectedEndpoint) == "" {
-		endpoint = c.baseURL
+		selectedEndpoint = c.baseURL
 	}
-	if endpoint == "" {
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(selectedEndpoint)
-		return nil, fmt.Errorf("%w: endpoint is required", ErrInvalidRequest)
+	if err != nil {
+		return nil, fmt.Errorf("%w: invalid endpoint: %v", ErrInvalidRequest, err)
+	}
+	endpoint, err = url.JoinPath(endpoint, defaults.OpenAIChatCompletionsPath)
+	if err != nil {
+		return nil, fmt.Errorf("%w: invalid endpoint path: %v", ErrInvalidRequest, err)
 	}
-	endpoint = strings.TrimRight(endpoint, "/") + defaults.OpenAIChatCompletionsPath
 
 	wireReq, err := openAIChatRequestFromGenerateRequest(req, c.defaultModel)
 	if err != nil {
@@ -113,13 +133,18 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 		return nil, fmt.Errorf("%w: failed to create request: %v", ErrRequestFailed, err)
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
-	if apiKey := strings.TrimSpace(req.Target.APIKey); apiKey != "" {
+	apiKey := strings.TrimSpace(req.Target.APIKey)
-		httpReq.Header.Set("Authorization", "Bearer "+apiKey)
+	envName := strings.TrimSpace(req.Target.APIKeyEnv)
-	} else if envName := strings.TrimSpace(req.Target.APIKeyEnv); envName != "" {
+	if apiKey == "" && envName != "" {
-		apiKey := strings.TrimSpace(os.Getenv(envName))
+		apiKey = strings.TrimSpace(os.Getenv(envName))
-		if apiKey == "" {
+	}
+	if apiKey == "" && req.Target.APIKeyRequired {
+		if envName != "" {
 			return nil, fmt.Errorf("%w: api key environment variable %q is not set", ErrInvalidRequest, envName)
 		}
+		return nil, fmt.Errorf("%w: api key is required", ErrInvalidRequest)
+	}
+	if apiKey != "" {
 		httpReq.Header.Set("Authorization", "Bearer "+apiKey)
 	}
 
@@ -130,30 +155,36 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 
 	httpResp, err := httpClient.Do(httpReq)
 	if err != nil {
-		return nil, fmt.Errorf("%w: %v", ErrRequestFailed, err)
+		return nil, &requestFailedError{cause: err}
 	}
 	defer httpResp.Body.Close()
 
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		_, _ = io.Copy(io.Discard, io.LimitReader(httpResp.Body, 4096))
+		return nil, providerHTTPErrorFromBody(
-		return nil, fmt.Errorf("%w: status=%d", ErrUnexpectedStatus, httpResp.StatusCode)
+			httpResp.StatusCode,
+			httpResp.ContentLength,
+			httpResp.Body,
+		)
+	}
+	if httpResp.ContentLength > maxOpenAIChatResponseBytes {
+		return nil, openAIChatResponseTooLargeError()
 	}
 
-	var wireResp openAIChatResponse
+	wireResp, err := decodeOpenAIChatResponse(httpResp.Body)
-	if err := json.NewDecoder(httpResp.Body).Decode(&wireResp); err != nil {
+	if err != nil {
-		return nil, fmt.Errorf("%w: failed to decode response: %v", ErrMalformedResponse, err)
+		return nil, err
 	}
 
 	if len(wireResp.Choices) == 0 {
 		return nil, fmt.Errorf("%w: no choices returned", ErrMalformedResponse)
 	}
 	content := wireResp.Choices[0].Message.Content
-	if content == "" {
+	if content == nil {
-		return nil, fmt.Errorf("%w: first choice has empty message content", ErrMalformedResponse)
+		return nil, fmt.Errorf("%w: first choice has missing message content", ErrMalformedResponse)
 	}
 
 	return &domain.GenerateResponse{
-		Content: content,
+		Content: *content,
 		Usage: domain.TokenUsage{
 			PromptTokens:     wireResp.Usage.PromptTokens,
 			CompletionTokens: wireResp.Usage.CompletionTokens,
@@ -164,6 +195,46 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 	}, nil
 }
 
+func decodeOpenAIChatResponse(body io.Reader) (openAIChatResponse, error) {
+	limited := &io.LimitedReader{
+		R: body,
+		N: maxOpenAIChatResponseBytes + 1,
+	}
+	decoder := json.NewDecoder(limited)
+
+	var response openAIChatResponse
+	if err := decoder.Decode(&response); err != nil {
+		if limited.N == 0 {
+			return openAIChatResponse{}, openAIChatResponseTooLargeError()
+		}
+		return openAIChatResponse{}, fmt.Errorf("%w: failed to decode response", ErrMalformedResponse)
+	}
+	if limited.N == 0 {
+		return openAIChatResponse{}, openAIChatResponseTooLargeError()
+	}
+
+	var trailing any
+	if err := decoder.Decode(&trailing); !errors.Is(err, io.EOF) {
+		if limited.N == 0 {
+			return openAIChatResponse{}, openAIChatResponseTooLargeError()
+		}
+		return openAIChatResponse{}, fmt.Errorf("%w: response contains trailing data", ErrMalformedResponse)
+	}
+	if limited.N == 0 {
+		return openAIChatResponse{}, openAIChatResponseTooLargeError()
+	}
+
+	return response, nil
+}
+
+func openAIChatResponseTooLargeError() error {
+	return fmt.Errorf(
+		"%w: response exceeds %d-byte limit",
+		ErrMalformedResponse,
+		maxOpenAIChatResponseBytes,
+	)
+}
+
 func openAIChatRequestFromGenerateRequest(req domain.GenerateRequest, defaultModel string) (openAIChatRequest, error) {
 	model := strings.TrimSpace(req.Target.Model)
 	if model == "" {
@@ -309,7 +380,7 @@ type openAICacheControl struct {
 
 type openAIChatResponseMessage struct {
 	Role    string  `json:"role"`
-	Content string `json:"content"`
+	Content *string `json:"content"`
 }
 
 type openAIChatResponse struct {

internal/llm/provider_http_error.go

@@ -0,0 +1,189 @@
+package llm
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"strings"
+	"unicode"
+)
+
+const (
+	maxProviderErrorResponseBytes   int64 = 64 << 10
+	maxProviderErrorIdentifierRunes       = 256
+	maxProviderErrorMessageRunes          = 4096
+)
+
+// ProviderHTTPError describes a non-success response from an LLM provider.
+type ProviderHTTPError struct {
+	statusCode      int
+	providerCode    string
+	providerType    string
+	providerMessage string
+}
+
+func (e *ProviderHTTPError) StatusCode() int {
+	if e == nil {
+		return 0
+	}
+	return e.statusCode
+}
+
+func (e *ProviderHTTPError) ProviderCode() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerCode
+}
+
+func (e *ProviderHTTPError) ProviderType() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerType
+}
+
+func (e *ProviderHTTPError) ProviderMessage() string {
+	if e == nil {
+		return ""
+	}
+	return e.providerMessage
+}
+
+func (e *ProviderHTTPError) Error() string {
+	if e == nil || e.statusCode == 0 {
+		return ErrUnexpectedStatus.Error()
+	}
+	return fmt.Sprintf("%s: status=%d", ErrUnexpectedStatus, e.statusCode)
+}
+
+func (e *ProviderHTTPError) GoString() string {
+	return e.Error()
+}
+
+func (e *ProviderHTTPError) Unwrap() error {
+	return ErrUnexpectedStatus
+}
+
+type providerErrorDetails struct {
+	providerCode    string
+	providerType    string
+	providerMessage string
+}
+
+func newProviderHTTPError(statusCode int, details providerErrorDetails) *ProviderHTTPError {
+	return &ProviderHTTPError{
+		statusCode:      statusCode,
+		providerCode:    details.providerCode,
+		providerType:    details.providerType,
+		providerMessage: details.providerMessage,
+	}
+}
+
+func providerHTTPErrorFromBody(statusCode int, contentLength int64, body io.Reader) *ProviderHTTPError {
+	if contentLength > maxProviderErrorResponseBytes {
+		return newProviderHTTPError(statusCode, providerErrorDetails{})
+	}
+
+	limited := &io.LimitedReader{
+		R: body,
+		N: maxProviderErrorResponseBytes + 1,
+	}
+	contents, err := io.ReadAll(limited)
+	if err != nil || limited.N == 0 {
+		return newProviderHTTPError(statusCode, providerErrorDetails{})
+	}
+	return newProviderHTTPError(statusCode, parseProviderErrorEnvelope(contents))
+}
+
+func parseProviderErrorEnvelope(body []byte) providerErrorDetails {
+	decoder := json.NewDecoder(strings.NewReader(string(body)))
+	decoder.UseNumber()
+
+	var envelope map[string]json.RawMessage
+	if err := decoder.Decode(&envelope); err != nil {
+		return providerErrorDetails{}
+	}
+	var trailing any
+	if err := decoder.Decode(&trailing); !errors.Is(err, io.EOF) {
+		return providerErrorDetails{}
+	}
+
+	rawError, ok := envelope["error"]
+	if !ok {
+		return providerErrorDetails{}
+	}
+	var providerError map[string]json.RawMessage
+	if err := json.Unmarshal(rawError, &providerError); err != nil || providerError == nil {
+		return providerErrorDetails{}
+	}
+
+	var details providerErrorDetails
+	if raw, ok := providerError["message"]; ok {
+		var value string
+		if json.Unmarshal(raw, &value) == nil {
+			details.providerMessage = normalizeProviderErrorMessage(value)
+		}
+	}
+	if raw, ok := providerError["type"]; ok {
+		var value string
+		if json.Unmarshal(raw, &value) == nil {
+			details.providerType = normalizeProviderErrorIdentifier(value)
+		}
+	}
+	if raw, ok := providerError["code"]; ok {
+		var value any
+		fieldDecoder := json.NewDecoder(strings.NewReader(string(raw)))
+		fieldDecoder.UseNumber()
+		if fieldDecoder.Decode(&value) == nil {
+			switch value := value.(type) {
+			case string:
+				details.providerCode = normalizeProviderErrorIdentifier(value)
+			case json.Number:
+				details.providerCode = normalizeProviderErrorIdentifier(value.String())
+			}
+		}
+	}
+
+	return details
+}
+
+func normalizeProviderErrorIdentifier(value string) string {
+	normalized := normalizeProviderErrorText(value)
+	if len([]rune(normalized)) > maxProviderErrorIdentifierRunes {
+		return ""
+	}
+	return normalized
+}
+
+func normalizeProviderErrorMessage(value string) string {
+	normalized := normalizeProviderErrorText(value)
+	runes := []rune(normalized)
+	if len(runes) <= maxProviderErrorMessageRunes {
+		return normalized
+	}
+	return string(runes[:maxProviderErrorMessageRunes-1]) + "…"
+}
+
+func normalizeProviderErrorText(value string) string {
+	value = strings.ToValidUTF8(value, "<22>")
+
+	var result strings.Builder
+	result.Grow(len(value))
+	separatorPending := false
+	for _, r := range value {
+		if unicode.IsSpace(r) || unicode.IsControl(r) || unicode.In(r, unicode.Cf) {
+			if result.Len() > 0 {
+				separatorPending = true
+			}
+			continue
+		}
+		if separatorPending {
+			result.WriteByte(' ')
+			separatorPending = false
+		}
+		result.WriteRune(r)
+	}
+	return result.String()
+}

internal/llm/provider_http_error_test.go

@@ -0,0 +1,255 @@
+package llm
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"reflect"
+	"strings"
+	"testing"
+	"unicode/utf8"
+)
+
+type guardedReader struct {
+	reader    io.Reader
+	remaining int64
+	bytes     int64
+	violated  bool
+}
+
+func (r *guardedReader) Read(buffer []byte) (int, error) {
+	if int64(len(buffer)) > r.remaining {
+		r.violated = true
+		return 0, errors.New("reader was read past its allowed boundary")
+	}
+	n, err := r.reader.Read(buffer)
+	r.bytes += int64(n)
+	r.remaining -= int64(n)
+	return n, err
+}
+
+type failingReader struct {
+	err error
+}
+
+func (r failingReader) Read([]byte) (int, error) {
+	return 0, r.err
+}
+
+func TestProviderHTTPErrorEnvelopeParsing(t *testing.T) {
+	tests := []struct {
+		name string
+		body string
+		want providerErrorDetails
+	}{
+		{
+			name: "all supported string fields",
+			body: `{"error":{"message":"diagnostic","type":"invalid_request_error","code":"unsupported_parameter"}}`,
+			want: providerErrorDetails{providerMessage: "diagnostic", providerType: "invalid_request_error", providerCode: "unsupported_parameter"},
+		},
+		{
+			name: "integer code",
+			body: `{"error":{"code":17}}`,
+			want: providerErrorDetails{providerCode: "17"},
+		},
+		{
+			name: "fractional code",
+			body: `{"error":{"code":1.25}}`,
+			want: providerErrorDetails{providerCode: "1.25"},
+		},
+		{
+			name: "exponent code",
+			body: `{"error":{"code":6.02e+23}}`,
+			want: providerErrorDetails{providerCode: "6.02e+23"},
+		},
+		{
+			name: "invalid fields do not discard valid fields",
+			body: `{"error":{"message":null,"type":"invalid_request_error","code":false}}`,
+			want: providerErrorDetails{providerType: "invalid_request_error"},
+		},
+		{
+			name: "unknown fields are ignored",
+			body: `{"trace":"do not retain","error":{"param":"temperature","metadata":{"secret":"x"}}}`,
+			want: providerErrorDetails{},
+		},
+		{name: "missing error", body: `{}`, want: providerErrorDetails{}},
+		{name: "null error", body: `{"error":null}`, want: providerErrorDetails{}},
+		{name: "scalar error", body: `{"error":"nope"}`, want: providerErrorDetails{}},
+		{name: "empty error", body: `{"error":{}}`, want: providerErrorDetails{}},
+		{name: "malformed", body: `{"error":`, want: providerErrorDetails{}},
+		{name: "truncated", body: `{"error":{"message":"x"`, want: providerErrorDetails{}},
+		{name: "trailing garbage", body: `{"error":{"message":"x"}} garbage`, want: providerErrorDetails{}},
+		{name: "second document", body: `{"error":{"message":"x"}} {}`, want: providerErrorDetails{}},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := parseProviderErrorEnvelope([]byte(tc.body)); !reflect.DeepEqual(got, tc.want) {
+				t.Fatalf("parseProviderErrorEnvelope() = %#v, want %#v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestProviderErrorTextNormalizationAndLimits(t *testing.T) {
+	validIdentifier := strings.Repeat("界", maxProviderErrorIdentifierRunes)
+	validMessage := strings.Repeat("界", maxProviderErrorMessageRunes)
+	tests := []struct {
+		name string
+		got  string
+		want string
+	}{
+		{name: "multibyte text", got: "Grüße 世界", want: "Grüße 世界"},
+		{name: "invalid UTF-8", got: string([]byte{'a', 0xff, 'b'}), want: "a<>b"},
+		{name: "whitespace control and format runs", got: " \n\talpha\x00\u200b\u200bbeta \r ", want: "alpha beta"},
+		{name: "blank normalization", got: "\t\u200b\n", want: ""},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := normalizeProviderErrorText(tc.got); got != tc.want {
+				t.Fatalf("normalizeProviderErrorText() = %q, want %q", got, tc.want)
+			}
+		})
+	}
+
+	if got := normalizeProviderErrorIdentifier(validIdentifier); got != validIdentifier {
+		t.Fatalf("exact identifier boundary = %q, want retained value", got)
+	}
+	if got := normalizeProviderErrorIdentifier(validIdentifier + "界"); got != "" {
+		t.Fatalf("overlong identifier = %q, want empty", got)
+	}
+	if got := normalizeProviderErrorMessage(validMessage); got != validMessage {
+		t.Fatalf("exact message boundary = %q, want retained value", got)
+	}
+	wantTruncatedMessage := strings.Repeat("界", maxProviderErrorMessageRunes-1) + "…"
+	if got := normalizeProviderErrorMessage(validMessage + "界"); got != wantTruncatedMessage {
+		t.Fatalf("overlong message length = %d, want %d", utf8.RuneCountInString(got), maxProviderErrorMessageRunes)
+	}
+}
+
+func TestProviderHTTPErrorIdentityAndFormatting(t *testing.T) {
+	const marker = "provider-secret-marker"
+	err := newProviderHTTPError(429, providerErrorDetails{
+		providerCode:    marker + "-code",
+		providerType:    marker + "-type",
+		providerMessage: marker + "-message",
+	})
+
+	if err.StatusCode() != 429 || err.ProviderCode() != marker+"-code" || err.ProviderType() != marker+"-type" || err.ProviderMessage() != marker+"-message" {
+		t.Fatalf("accessors returned unexpected values: %#v", err)
+	}
+	if !errors.Is(err, ErrUnexpectedStatus) {
+		t.Fatalf("errors.Is(%v, ErrUnexpectedStatus) = false", err)
+	}
+	for _, rendered := range []string{fmt.Sprintf("%v", err), fmt.Sprintf("%+v", err), fmt.Sprintf("%#v", err)} {
+		if rendered != "llm returned non-success status: status=429" {
+			t.Fatalf("formatted error = %q", rendered)
+		}
+		if strings.Contains(rendered, marker) {
+			t.Fatalf("formatted error exposed provider marker: %q", rendered)
+		}
+	}
+
+	var nilError *ProviderHTTPError
+	if nilError.StatusCode() != 0 || nilError.ProviderCode() != "" || nilError.ProviderType() != "" || nilError.ProviderMessage() != "" {
+		t.Fatal("nil accessors returned provider values")
+	}
+	if nilError.Error() != "llm returned non-success status" || nilError.GoString() != "llm returned non-success status" || !errors.Is(nilError, ErrUnexpectedStatus) {
+		t.Fatalf("nil error behavior is not safe: %v", nilError)
+	}
+
+	zero := &ProviderHTTPError{}
+	if zero.Error() != "llm returned non-success status" || zero.GoString() != "llm returned non-success status" || !errors.Is(zero, ErrUnexpectedStatus) {
+		t.Fatalf("zero error behavior is not safe: %v", zero)
+	}
+}
+
+func TestProviderHTTPErrorBodyBounds(t *testing.T) {
+	const (
+		statusCode = 502
+		marker     = "provider-body-marker"
+	)
+
+	ordinaryBody := `{"error":{"message":"` + marker + `"}}`
+	exactLimitBody := ordinaryBody + strings.Repeat(" ", int(maxProviderErrorResponseBytes)-len(ordinaryBody))
+	overLimitBody := ordinaryBody + strings.Repeat(" ", int(maxProviderErrorResponseBytes)+1-len(ordinaryBody))
+	tests := []struct {
+		name          string
+		contentLength int64
+		reader        io.Reader
+		wantRead      int64
+		wantMessage   string
+	}{
+		{
+			name:          "recognized envelope",
+			contentLength: int64(len(ordinaryBody)),
+			reader:        strings.NewReader(ordinaryBody),
+			wantRead:      int64(len(ordinaryBody)),
+			wantMessage:   marker,
+		},
+		{
+			name:          "exact limit",
+			contentLength: maxProviderErrorResponseBytes,
+			reader:        strings.NewReader(exactLimitBody),
+			wantRead:      maxProviderErrorResponseBytes,
+			wantMessage:   marker,
+		},
+		{
+			name:          "declared oversize does not read",
+			contentLength: maxProviderErrorResponseBytes + 1,
+			reader:        strings.NewReader(ordinaryBody),
+			wantRead:      0,
+		},
+		{
+			name:          "unknown length oversize",
+			contentLength: -1,
+			reader:        strings.NewReader(overLimitBody),
+			wantRead:      maxProviderErrorResponseBytes + 1,
+		},
+		{
+			name:          "underreported oversize",
+			contentLength: maxProviderErrorResponseBytes,
+			reader:        strings.NewReader(overLimitBody),
+			wantRead:      maxProviderErrorResponseBytes + 1,
+		},
+		{
+			name:          "read failure",
+			contentLength: -1,
+			reader:        failingReader{err: errors.New("read failure")},
+			wantRead:      0,
+		},
+		{
+			name:          "empty body",
+			contentLength: 0,
+			reader:        strings.NewReader(""),
+			wantRead:      0,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			reader := &guardedReader{
+				reader:    tc.reader,
+				remaining: maxProviderErrorResponseBytes + 1,
+			}
+			err := providerHTTPErrorFromBody(statusCode, tc.contentLength, reader)
+			if err == nil || err.StatusCode() != statusCode {
+				t.Fatalf("error status = %v, want %d", err, statusCode)
+			}
+			if reader.bytes != tc.wantRead {
+				t.Fatalf("body bytes read = %d, want %d", reader.bytes, tc.wantRead)
+			}
+			if reader.violated {
+				t.Fatal("body reader was asked to read beyond the overflow probe")
+			}
+			if got := err.ProviderMessage(); got != tc.wantMessage {
+				t.Fatalf("provider message = %q, want %q", got, tc.wantMessage)
+			}
+			if tc.wantMessage == "" {
+				if err.ProviderCode() != "" || err.ProviderType() != "" || strings.Contains(err.Error(), marker) {
+					t.Fatalf("discarded details were retained: %#v", err)
+				}
+			}
+		})
+	}
+}

internal/llm/provider_http_error_transport_test.go

@@ -0,0 +1,145 @@
+package llm
+
+import (
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+)
+
+func TestOpenAICompatibleClientStructuredNonSuccessResponse(t *testing.T) {
+	body := `{"error":{"message":"  provider\nmessage\u200b","type":"invalid\ttype","code":1.5e+4}}`
+	responseBody := &countingReadCloser{reader: strings.NewReader(body)}
+	client := newNonSuccessResponseClient(t, http.StatusBadRequest, int64(len(body)), responseBody)
+
+	response, err := client.Generate(context.Background(), ordinaryGenerateRequest())
+	if response != nil {
+		t.Fatalf("response = %#v, want nil", response)
+	}
+	if !errors.Is(err, ErrUnexpectedStatus) {
+		t.Fatalf("errors.Is(%v, ErrUnexpectedStatus) = false", err)
+	}
+	var providerHTTPError *ProviderHTTPError
+	if !errors.As(err, &providerHTTPError) {
+		t.Fatalf("error = %T, want *ProviderHTTPError", err)
+	}
+	if providerHTTPError.StatusCode() != http.StatusBadRequest || providerHTTPError.ProviderCode() != "1.5e+4" || providerHTTPError.ProviderType() != "invalid type" || providerHTTPError.ProviderMessage() != "provider message" {
+		t.Fatalf("provider error = %#v", providerHTTPError)
+	}
+	if !responseBody.closed {
+		t.Fatal("non-success response body was not closed")
+	}
+}
+
+func TestOpenAICompatibleClientNonSuccessBodyOwnership(t *testing.T) {
+	const marker = "provider-body-marker"
+	normalBody := `{"error":{"message":"` + marker + `"}}`
+	overLimitBody := normalBody + strings.Repeat(" ", int(maxProviderErrorResponseBytes)+1-len(normalBody))
+	tests := []struct {
+		name          string
+		contentLength int64
+		reader        io.Reader
+		wantRead      int64
+		wantMessage   string
+	}{
+		{
+			name:          "normal",
+			contentLength: int64(len(normalBody)),
+			reader:        strings.NewReader(normalBody),
+			wantRead:      int64(len(normalBody)),
+			wantMessage:   marker,
+		},
+		{
+			name:          "declared oversize",
+			contentLength: maxProviderErrorResponseBytes + 1,
+			reader:        strings.NewReader(normalBody),
+			wantRead:      0,
+		},
+		{
+			name:          "streamed oversize",
+			contentLength: -1,
+			reader: &guardedReader{
+				reader:    strings.NewReader(overLimitBody),
+				remaining: maxProviderErrorResponseBytes + 1,
+			},
+			wantRead: maxProviderErrorResponseBytes + 1,
+		},
+		{
+			name:          "underreported oversize",
+			contentLength: maxProviderErrorResponseBytes,
+			reader: &guardedReader{
+				reader:    strings.NewReader(overLimitBody),
+				remaining: maxProviderErrorResponseBytes + 1,
+			},
+			wantRead: maxProviderErrorResponseBytes + 1,
+		},
+		{
+			name:          "malformed",
+			contentLength: 1,
+			reader:        strings.NewReader("{"),
+			wantRead:      1,
+		},
+		{
+			name:          "read failure",
+			contentLength: -1,
+			reader:        failingReader{err: errors.New("response read failed")},
+			wantRead:      0,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			body := &countingReadCloser{reader: tc.reader}
+			client := newNonSuccessResponseClient(t, http.StatusBadGateway, tc.contentLength, body)
+
+			response, err := client.Generate(context.Background(), ordinaryGenerateRequest())
+			if response != nil {
+				t.Fatalf("response = %#v, want nil", response)
+			}
+			var providerHTTPError *ProviderHTTPError
+			if !errors.As(err, &providerHTTPError) {
+				t.Fatalf("error = %T, want *ProviderHTTPError", err)
+			}
+			if !body.closed {
+				t.Fatal("response body was not closed")
+			}
+			if body.bytesRead != tc.wantRead {
+				t.Fatalf("body bytes read = %d, want %d", body.bytesRead, tc.wantRead)
+			}
+			if body.bytesRead > maxProviderErrorResponseBytes+1 {
+				t.Fatalf("body bytes read = %d, exceeds overflow probe", body.bytesRead)
+			}
+			if guarded, ok := tc.reader.(*guardedReader); ok && guarded.violated {
+				t.Fatal("body reader was asked to read beyond the overflow probe")
+			}
+			if got := providerHTTPError.ProviderMessage(); got != tc.wantMessage {
+				t.Fatalf("provider message = %q, want %q", got, tc.wantMessage)
+			}
+			if tc.wantMessage == "" && (providerHTTPError.ProviderCode() != "" || providerHTTPError.ProviderType() != "" || strings.Contains(providerHTTPError.Error(), marker)) {
+				t.Fatalf("discarded details were retained: %#v", providerHTTPError)
+			}
+		})
+	}
+}
+
+func newNonSuccessResponseClient(t *testing.T, statusCode int, contentLength int64, body io.ReadCloser) *OpenAICompatibleClient {
+	t.Helper()
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		BaseURL: "https://provider.example/v1",
+		Model:   "m",
+		HTTPClient: &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
+			return &http.Response{
+				StatusCode:    statusCode,
+				ContentLength: contentLength,
+				Body:          body,
+			}, nil
+		})},
+	})
+	if err != nil {
+		t.Fatalf("construct client: %v", err)
+	}
+	return client
+}

internal/profile/builtin/assets/google/rakestrawhome-gemma-4-31b.yml

@@ -0,0 +1,3 @@
+id: rakestrawhome-gemma-4-31b
+backend: rakestrawhome
+model: google/gemma-4-31b-it

internal/profile/builtin/repository.go

@@ -2,7 +2,6 @@ package builtin
 
 import (
 	"embed"
-	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
 )
@@ -15,17 +14,3 @@ var assets embed.FS
 func NewRepository() profile.Repository {
 	return profile.NewFSRepository(assets, assetRoot)
 }
-
-func NewRepositoryWithPrimary(primary profile.Repository) profile.Repository {
-	if primary == nil {
-		return NewRepository()
-	}
-	return profile.NewOverlayRepository(primary, NewRepository())
-}
-
-func NewRepositoryWithDirectory(dir string) profile.Repository {
-	if strings.TrimSpace(dir) == "" {
-		return NewRepository()
-	}
-	return NewRepositoryWithPrimary(profile.NewFilesystemRepository(dir))
-}

internal/profile/builtin/repository_test.go

@@ -2,14 +2,11 @@ package builtin
 
 import (
 	"context"
-	"errors"
 	"io/fs"
 	"strings"
 	"testing"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/backend"
-	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
-	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
 	"gopkg.in/yaml.v3"
 )
 
@@ -29,8 +26,8 @@ func TestBuiltInProfilesValidateThroughRepository(t *testing.T) {
 			if p.ID != id {
 				t.Fatalf("expected profile id %q, got %q", id, p.ID)
 			}
-			if p.BackendID != backend.OpenRouterID {
+			if !builtInBackendIDs[p.BackendID] {
-				t.Fatalf("expected profile %q to select %q, got %q", id, backend.OpenRouterID, p.BackendID)
+				t.Fatalf("expected profile %q to select a maintained built-in, got %q", id, p.BackendID)
 			}
 			if p.Endpoint != "" || p.APIKeyEnv != "" {
 				t.Fatalf("expected profile %q to inherit backend connection settings, got endpoint=%q api_key_env=%q", id, p.Endpoint, p.APIKeyEnv)
@@ -43,6 +40,33 @@ func TestBuiltInProfilesDoNotContainDuplicateIDsOrRawAPIKeys(t *testing.T) {
 	loadBuiltInProfileIDs(t)
 }
 
+func TestRakestrawhomeGemmaProfileUsesNativeDefaults(t *testing.T) {
+	p, err := NewRepository().GetProfile(context.Background(), "rakestrawhome-gemma-4-31b")
+	if err != nil {
+		t.Fatalf("load Rakestrawhome Gemma profile: %v", err)
+	}
+	if p.ID != "rakestrawhome-gemma-4-31b" ||
+		p.BackendID != backend.RakestrawHomeID ||
+		p.Model != "google/gemma-4-31b-it" ||
+		p.Endpoint != "" ||
+		p.Temperature != 0 ||
+		p.MaxTokens != 0 ||
+		p.TopP != 0 ||
+		p.TimeoutSeconds != 0 ||
+		p.ServiceTier != "" ||
+		p.ReasoningEffort != "" ||
+		p.APIKeyEnv != "" ||
+		p.APIKeyRequired ||
+		p.ExtraParams != nil {
+		t.Fatalf("unexpected Rakestrawhome Gemma profile: %#v", p)
+	}
+}
+
+var builtInBackendIDs = map[string]bool{
+	backend.OpenRouterID:    true,
+	backend.RakestrawHomeID: true,
+}
+
 func loadBuiltInProfileIDs(t *testing.T) map[string]string {
 	t.Helper()
 
@@ -67,8 +91,9 @@ func loadBuiltInProfileIDs(t *testing.T) map[string]string {
 		if _, ok := raw["api_key"]; ok {
 			t.Fatalf("built-in profile %s contains raw api_key", name)
 		}
-		if raw["backend"] != backend.OpenRouterID {
+		backendID, ok := raw["backend"].(string)
-			t.Fatalf("built-in profile %s does not select %q", name, backend.OpenRouterID)
+		if !ok || !builtInBackendIDs[backendID] {
+			t.Fatalf("built-in profile %s does not select a maintained built-in: %#v", name, raw["backend"])
 		}
 		if _, ok := raw["endpoint"]; ok {
 			t.Fatalf("built-in profile %s repeats endpoint", name)
@@ -91,53 +116,3 @@ func loadBuiltInProfileIDs(t *testing.T) map[string]string {
 	}
 	return ids
 }
-
-func TestRepositoryWithPrimaryUsesPrimaryBeforeBuiltIns(t *testing.T) {
-	repo := NewRepositoryWithPrimary(staticProfileRepo{
-		profiles: map[string]string{"mistral-small-3": "custom-model"},
-	})
-
-	p, err := repo.GetProfile(context.Background(), "mistral-small-3")
-	if err != nil {
-		t.Fatalf("expected profile to load, got %v", err)
-	}
-	if p.Model != "custom-model" {
-		t.Fatalf("expected primary profile to override built-in, got %+v", p)
-	}
-}
-
-func TestRepositoryWithPrimaryFallsBackToBuiltIns(t *testing.T) {
-	repo := NewRepositoryWithPrimary(staticProfileRepo{})
-
-	p, err := repo.GetProfile(context.Background(), "mistral-small-3")
-	if err != nil {
-		t.Fatalf("expected built-in profile to load, got %v", err)
-	}
-	if p.ID != "mistral-small-3" {
-		t.Fatalf("unexpected profile: %+v", p)
-	}
-}
-
-func TestRepositoryWithPrimaryDoesNotFallBackAfterPrimaryError(t *testing.T) {
-	repo := NewRepositoryWithPrimary(staticProfileRepo{err: profile.ErrInvalidProfile})
-
-	_, err := repo.GetProfile(context.Background(), "mistral-small-3")
-	if !errors.Is(err, profile.ErrInvalidProfile) {
-		t.Fatalf("expected primary error, got %v", err)
-	}
-}
-
-type staticProfileRepo struct {
-	profiles map[string]string
-	err      error
-}
-
-func (r staticProfileRepo) GetProfile(_ context.Context, id string) (*domain.ExecutionProfile, error) {
-	if r.err != nil {
-		return nil, r.err
-	}
-	if model, ok := r.profiles[id]; ok {
-		return &domain.ExecutionProfile{ID: id, Endpoint: "http://primary/v1", Model: model}, nil
-	}
-	return nil, profile.ErrProfileNotFound
-}

internal/profile/definition.go

@@ -0,0 +1,47 @@
+package profile
+
+import (
+	"errors"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+// NormalizeAndValidateDefinition normalizes and validates one source-local
+// profile definition without resolving a base profile.
+func NormalizeAndValidateDefinition(profile *domain.ExecutionProfile) error {
+	if profile == nil {
+		return errors.New("profile is required")
+	}
+
+	profile.ID = strings.TrimSpace(profile.ID)
+	profile.BaseProfileID = strings.TrimSpace(profile.BaseProfileID)
+	profile.BackendID = strings.TrimSpace(profile.BackendID)
+	profile.Endpoint = strings.TrimSpace(profile.Endpoint)
+
+	if profile.ID == "" {
+		return errors.New("id is required")
+	}
+	if profile.Endpoint != "" {
+		endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(profile.Endpoint)
+		if err != nil {
+			return err
+		}
+		profile.Endpoint = endpoint
+	}
+	if profile.BaseProfileID == "" {
+		if profile.BackendID == "" && profile.Endpoint == "" {
+			return errors.New("backend or endpoint is required")
+		}
+		if strings.TrimSpace(profile.Model) == "" {
+			return errors.New("model is required")
+		}
+	}
+
+	return domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{
+		Temperature:    profile.Temperature,
+		MaxTokens:      profile.MaxTokens,
+		TopP:           profile.TopP,
+		TimeoutSeconds: profile.TimeoutSeconds,
+	})
+}

internal/profile/filesystem_repository.go

@@ -5,13 +5,14 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
 	"os"
-	"path"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
+	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
 	"gopkg.in/yaml.v3"
 )
 
@@ -73,7 +74,8 @@ func (r *overlayRepository) GetProfile(ctx context.Context, id string) (*domain.
 }
 
 func loadProfile(ctx context.Context, fsys fs.FS, root string, id string) (*domain.ExecutionProfile, error) {
-	if strings.TrimSpace(id) == "" {
+	id = strings.TrimSpace(id)
+	if id == "" {
 		return nil, fmt.Errorf("%w: profile id is required", ErrInvalidProfile)
 	}
 	if fsys == nil {
@@ -94,42 +96,49 @@ func loadProfile(ctx context.Context, fsys fs.FS, root string, id string) (*doma
 		}
 
 		relPath := filecatalog.DisplayPath(root, fullPath)
-		fileMatch := filecatalog.Stem(path.Base(fullPath)) == id
 		data, err := fs.ReadFile(fsys, fullPath)
 		if err != nil {
 			return nil, fmt.Errorf("failed to read profile file %s: %w", relPath, err)
 		}
-		metadata := readProfileFileMetadata(data)
+		metadata, metadataErr := readProfileFileMetadata(data)
-		idMatch := fileMatch || metadata.id == id
+		idMatch := metadata.matchesID(id)
+		if metadataErr != nil {
+			if idMatch {
+				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, metadataErr)
+			}
+			continue
+		}
 		if metadata.hasRawAPIKey {
 			if idMatch {
 				return nil, fmt.Errorf("%w: %s", ErrRawAPIKeyNotAllowed, relPath)
 			}
 			continue
 		}
-
+		if !idMatch {
-		var prof domain.ExecutionProfile
-		decoder := yaml.NewDecoder(bytes.NewReader(data))
-		decoder.KnownFields(true)
-		if err := decoder.Decode(&prof); err != nil {
-			if idMatch {
-				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
-			}
 			continue
 		}
 
+		prof, err := decodeProfile(data)
+		if err != nil {
+			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
+		}
+
+		prof.ID = strings.TrimSpace(prof.ID)
 		if prof.ID != id {
 			continue
 		}
-		prof.BackendID = strings.TrimSpace(prof.BackendID)
+		prof.ExtraParams, err = jsonvalue.CopyMap(prof.ExtraParams)
-		if err := validateProfile(&prof); err != nil {
+		if err != nil {
+			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidProfile, relPath, err)
+		}
+		if err := NormalizeAndValidateDefinition(prof); err != nil {
 			if errors.Is(err, ErrRawAPIKeyNotAllowed) {
 				return nil, fmt.Errorf("%w: %s", err, relPath)
 			}
 			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidProfile, relPath, err)
 		}
 		matches = append(matches, profileMatch{
-			profile: &prof,
+			profile: prof,
 			path:    relPath,
 		})
 	}
@@ -155,15 +164,36 @@ type profileMatch struct {
 }
 
 type profileFileMetadata struct {
-	id           string
+	ids          []string
 	hasRawAPIKey bool
 }
 
-func readProfileFileMetadata(data []byte) profileFileMetadata {
+func readProfileFileMetadata(data []byte) (profileFileMetadata, error) {
+	decoder := yaml.NewDecoder(bytes.NewReader(data))
 	var node yaml.Node
-	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&node); err != nil {
+	if err := decoder.Decode(&node); err != nil {
-		return profileFileMetadata{}
+		return profileFileMetadata{}, err
 	}
+	metadata := profileMetadataFromNode(&node)
+	documentCount := 1
+	for {
+		var trailing yaml.Node
+		err := decoder.Decode(&trailing)
+		if errors.Is(err, io.EOF) {
+			if documentCount == 1 {
+				return metadata, nil
+			}
+			return metadata, errors.New("profile file must contain exactly one YAML document")
+		}
+		if err != nil {
+			return metadata, err
+		}
+		documentCount++
+		metadata.merge(profileMetadataFromNode(&trailing))
+	}
+}
+
+func profileMetadataFromNode(node *yaml.Node) profileFileMetadata {
 	if node.Kind != yaml.DocumentNode || len(node.Content) == 0 {
 		return profileFileMetadata{}
 	}
@@ -178,7 +208,7 @@ func readProfileFileMetadata(data []byte) profileFileMetadata {
 		value := mapping.Content[i+1]
 		switch key.Value {
 		case "id":
-			metadata.id = strings.TrimSpace(value.Value)
+			metadata.ids = append(metadata.ids, strings.TrimSpace(value.Value))
 		case "api_key":
 			metadata.hasRawAPIKey = true
 		}
@@ -186,29 +216,41 @@ func readProfileFileMetadata(data []byte) profileFileMetadata {
 	return metadata
 }
 
-func validateProfile(p *domain.ExecutionProfile) error {
+func (m profileFileMetadata) matchesID(id string) bool {
-	if strings.TrimSpace(p.ID) == "" {
+	for _, candidate := range m.ids {
-		return errors.New("id is required")
+		if candidate == id {
+			return true
 		}
-	if strings.TrimSpace(p.BackendID) == "" && strings.TrimSpace(p.Endpoint) == "" {
-		return errors.New("backend or endpoint is required")
 	}
-	if strings.TrimSpace(p.Model) == "" {
+	return false
-		return errors.New("model is required")
+}
-	}
+
-
+func (m *profileFileMetadata) merge(other profileFileMetadata) {
-	if p.Temperature < 0 || p.Temperature > 2 {
+	m.ids = append(m.ids, other.ids...)
-		return errors.New("temperature must be between 0 and 2")
+	m.hasRawAPIKey = m.hasRawAPIKey || other.hasRawAPIKey
-	}
+}
-	if p.MaxTokens < 0 {
+
-		return errors.New("max_tokens must be greater than or equal to 0")
+func decodeProfile(data []byte) (*domain.ExecutionProfile, error) {
-	}
+	var prof domain.ExecutionProfile
-	if p.TopP < 0 || p.TopP > 1 {
+	decoder := yaml.NewDecoder(bytes.NewReader(data))
-		return errors.New("top_p must be between 0 and 1")
+	decoder.KnownFields(true)
-	}
+	if err := decoder.Decode(&prof); err != nil {
-	if p.TimeoutSeconds < 0 {
+		return nil, err
-		return errors.New("timeout_seconds must be greater than or equal to 0")
+	}
-	}
+	if err := requireYAMLStreamEnd(decoder); err != nil {
-
+		return nil, err
-	return nil
+	}
+	return &prof, nil
+}
+
+func requireYAMLStreamEnd(decoder *yaml.Decoder) error {
+	var trailing yaml.Node
+	err := decoder.Decode(&trailing)
+	if errors.Is(err, io.EOF) {
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	return errors.New("profile file must contain exactly one YAML document")
 }

internal/profile/repository_benchmark_test.go

@@ -0,0 +1,58 @@
+package profile
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"testing/fstest"
+)
+
+func BenchmarkProfileRepositoryLookup(b *testing.B) {
+	for _, size := range []int{10, 1000} {
+		b.Run(fmt.Sprintf("catalog-%d", size), func(b *testing.B) {
+			files := fstest.MapFS{
+				"target.yaml": profileMapFile(`
+id: target
+endpoint: http://localhost:8000/v1
+model: target-model
+extra_params:
+  selected: true
+`),
+			}
+			metadataNames := []string{"target.yaml"}
+			for i := 1; i < size; i++ {
+				name := fmt.Sprintf("profile-%04d.yaml", i)
+				files[name] = profileMapFile(fmt.Sprintf(`
+id: profile-%04d
+endpoint: http://localhost:8000/v1
+model: unrelated-model
+temperature: 0.5
+max_tokens: 500
+extra_params:
+  provider:
+    order:
+      - first
+      - second
+`, i))
+				metadataNames = append(metadataNames, name)
+			}
+
+			fsys := &recordingProfileFS{FS: files}
+			repo := NewFSRepository(fsys, ".")
+			b.ReportAllocs()
+			b.ResetTimer()
+			for i := 0; i < b.N; i++ {
+				if _, err := repo.GetProfile(context.Background(), "target"); err != nil {
+					b.Fatal(err)
+				}
+			}
+			b.StopTimer()
+
+			for _, name := range metadataNames {
+				if got := fsys.openCount(name); got != b.N {
+					b.Fatalf("metadata %q opens = %d, want %d", name, got, b.N)
+				}
+			}
+		})
+	}
+}

internal/profile/repository_test.go

@@ -2,11 +2,13 @@ package profile
 
 import (
 	"context"
-	"encoding/json"
 	"errors"
+	"fmt"
+	"io/fs"
 	"os"
 	"path/filepath"
 	"strings"
+	"sync"
 	"testing"
 	"testing/fstest"
 
@@ -61,7 +63,7 @@ func TestFilesystemRepository_GetProfile(t *testing.T) {
 			wantErr      bool
 		}{
 			{name: "backend only", connection: "backend: '  openrouter  '", wantBackend: "openrouter"},
-			{name: "endpoint only", connection: "endpoint: http://localhost:8000/v1", wantEndpoint: "http://localhost:8000/v1"},
+			{name: "endpoint only", connection: "endpoint: '  https://localhost:8000/nested/v1  '", wantEndpoint: "https://localhost:8000/nested/v1"},
 			{name: "both", connection: "backend: openrouter\nendpoint: http://localhost:8000/v1", wantBackend: "openrouter", wantEndpoint: "http://localhost:8000/v1"},
 			{name: "neither", wantErr: true},
 			{name: "blank backend", connection: "backend: '   '", wantErr: true},
@@ -126,63 +128,6 @@ temperature: 0.1
 		}
 	})
 
-	t.Run("valid profile with JSON-compatible extra params", func(t *testing.T) {
-		writeProfileTestFile(t, filepath.Join(tmpDir, "json-extra-params.yaml"), `
-id: json-extra-params
-endpoint: http://localhost:8000/v1
-model: nested-model
-extra_params:
-  string_value: enabled
-  number_value: 42
-  boolean_value: true
-  object_value:
-    nested: value
-    count: 2
-  array_value:
-    - first
-    - 3
-    - false
-`)
-
-		p, err := repo.GetProfile(ctx, "json-extra-params")
-		if err != nil {
-			t.Fatalf("expected no error, got %v", err)
-		}
-
-		var got map[string]any
-		encoded, err := json.Marshal(p.ExtraParams)
-		if err != nil {
-			t.Fatalf("expected extra_params to marshal as JSON, got %v", err)
-		}
-		if err := json.Unmarshal(encoded, &got); err != nil {
-			t.Fatalf("expected extra_params JSON to decode, got %v", err)
-		}
-
-		if got["string_value"] != "enabled" {
-			t.Fatalf("unexpected string extra param: %#v", got["string_value"])
-		}
-		if got["number_value"] != float64(42) {
-			t.Fatalf("unexpected number extra param: %#v", got["number_value"])
-		}
-		if got["boolean_value"] != true {
-			t.Fatalf("unexpected boolean extra param: %#v", got["boolean_value"])
-		}
-		objectValue, ok := got["object_value"].(map[string]any)
-		if !ok {
-			t.Fatalf("expected object extra param, got %#v", got["object_value"])
-		}
-		if objectValue["nested"] != "value" || objectValue["count"] != float64(2) {
-			t.Fatalf("unexpected object extra param: %#v", objectValue)
-		}
-		arrayValue, ok := got["array_value"].([]any)
-		if !ok {
-			t.Fatalf("expected array extra param, got %#v", got["array_value"])
-		}
-		if len(arrayValue) != 3 || arrayValue[0] != "first" || arrayValue[1] != float64(3) || arrayValue[2] != false {
-			t.Fatalf("unexpected array extra param: %#v", arrayValue)
-		}
-	})
-
 	t.Run("duplicate profile IDs fail as ambiguous", func(t *testing.T) {
 		writeProfileTestFile(t, filepath.Join(tmpDir, "duplicate-profile-a.yaml"), `
 id: duplicate-profile
@@ -245,10 +190,10 @@ api_key: secret
 		}
 	})
 
-	t.Run("invalid yaml", func(t *testing.T) {
+	t.Run("unidentifiable invalid yaml is unrelated", func(t *testing.T) {
 		_, err := repo.GetProfile(ctx, "invalid_yaml")
-		if !errors.Is(err, ErrInvalidYAML) {
+		if !errors.Is(err, ErrProfileNotFound) {
-			t.Fatalf("expected ErrInvalidYAML, got %v", err)
+			t.Fatalf("expected ErrProfileNotFound, got %v", err)
 		}
 	})
 
@@ -274,14 +219,14 @@ api_key: secret
 	})
 
 	t.Run("unknown field", func(t *testing.T) {
-		_, err := repo.GetProfile(ctx, "unknown_field")
+		_, err := repo.GetProfile(ctx, "unknown-field")
 		if !errors.Is(err, ErrInvalidYAML) {
 			t.Fatalf("expected ErrInvalidYAML for strict decode unknown field, got %v", err)
 		}
 	})
 
 	t.Run("raw api_key rejected", func(t *testing.T) {
-		_, err := repo.GetProfile(ctx, "raw_api_key")
+		_, err := repo.GetProfile(ctx, "raw-api-key")
 		if !errors.Is(err, ErrRawAPIKeyNotAllowed) {
 			t.Fatalf("expected ErrRawAPIKeyNotAllowed, got %v", err)
 		}
@@ -406,6 +351,614 @@ model: second
 	})
 }
 
+func TestProfileRepositoriesRejectInvalidEndpoints(t *testing.T) {
+	tests := []struct {
+		name        string
+		endpoint    string
+		withBackend bool
+	}{
+		{name: "relative", endpoint: "/v1"},
+		{name: "missing host", endpoint: "https:///v1"},
+		{name: "unsupported scheme", endpoint: "ftp://provider.example/v1"},
+		{name: "user information", endpoint: "https://user@provider.example/v1"},
+		{name: "query", endpoint: "https://provider.example/v1?mode=chat"},
+		{name: "fragment", endpoint: "https://provider.example/v1#chat"},
+		{name: "backend with invalid override", endpoint: "/v1", withBackend: true},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			backend := ""
+			if tc.withBackend {
+				backend = "backend: openrouter\n"
+			}
+			repo := NewFSRepository(fstest.MapFS{
+				"profiles/invalid.yaml": profileMapFile(fmt.Sprintf(
+					"id: invalid-endpoint\nmodel: model\n%sendpoint: %q\n",
+					backend,
+					tc.endpoint,
+				)),
+			}, "profiles")
+
+			_, err := repo.GetProfile(context.Background(), "invalid-endpoint")
+			if !errors.Is(err, ErrInvalidProfile) {
+				t.Fatalf("expected ErrInvalidProfile, got %v", err)
+			}
+		})
+	}
+}
+
+func TestProfileRepositoriesValidateExtraParams(t *testing.T) {
+	const validProfile = `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: model
+extra_params:
+  string_value: enabled
+  object_value:
+    nested: true
+  array_value:
+    - first
+    - 3
+`
+	tests := []struct {
+		name        string
+		definition  string
+		wantErr     bool
+		diagnostics []string
+	}{
+		{name: "valid nested values", definition: validProfile},
+		{
+			name: "empty key",
+			definition: `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: model
+extra_params:
+  "": value
+`,
+			wantErr:     true,
+			diagnostics: []string{"extra_params", "key must not be empty"},
+		},
+		{
+			name: "non-finite value",
+			definition: `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: model
+extra_params:
+  invalid: .nan
+`,
+			wantErr:     true,
+			diagnostics: []string{"extra_params.invalid", "must be finite"},
+		},
+		{
+			name: "nested non-finite value",
+			definition: `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: model
+extra_params:
+  outer:
+    invalid: .inf
+`,
+			wantErr:     true,
+			diagnostics: []string{"extra_params.outer.invalid", "must be finite"},
+		},
+		{
+			name: "unsupported decoded value",
+			definition: `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: model
+extra_params:
+  timestamp: 2026-08-11T12:34:56Z
+`,
+			wantErr:     true,
+			diagnostics: []string{"extra_params.timestamp", "unsupported JSON value type"},
+		},
+		{
+			name:        "excessive nesting",
+			definition:  deeplyNestedExtraParamsProfile(101),
+			wantErr:     true,
+			diagnostics: []string{"extra_params", "JSON container depth limit exceeded"},
+		},
+	}
+
+	for _, source := range profileRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, map[string]string{"selected.yaml": tc.definition})
+				got, err := repo.GetProfile(context.Background(), "selected-profile")
+				if tc.wantErr {
+					if !errors.Is(err, ErrInvalidProfile) {
+						t.Fatalf("expected ErrInvalidProfile, got %v", err)
+					}
+					if !strings.Contains(err.Error(), "selected.yaml") {
+						t.Fatalf("expected source path in error, got %v", err)
+					}
+					for _, diagnostic := range tc.diagnostics {
+						if !strings.Contains(err.Error(), diagnostic) {
+							t.Fatalf("expected error to contain %q, got %v", diagnostic, err)
+						}
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load valid profile: %v", err)
+				}
+				if got.ExtraParams["string_value"] != "enabled" {
+					t.Fatalf("unexpected copied extra params: %#v", got.ExtraParams)
+				}
+				objectValue, objectOK := got.ExtraParams["object_value"].(map[string]any)
+				arrayValue, arrayOK := got.ExtraParams["array_value"].([]any)
+				if !objectOK || objectValue["nested"] != true ||
+					!arrayOK || len(arrayValue) != 2 || arrayValue[0] != "first" || arrayValue[1] != 3 {
+					t.Fatalf("unexpected copied nested extra params: %#v", got.ExtraParams)
+				}
+			})
+		}
+	}
+}
+
+func TestProfileRepositoriesSelectCanonicalYAMLID(t *testing.T) {
+	const validProfile = `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected-model
+`
+	tests := []struct {
+		name        string
+		files       map[string]string
+		wantErr     error
+		diagnostics []string
+	}{
+		{
+			name: "same stem unknown field with different id is unrelated",
+			files: map[string]string{
+				"selected-profile.yaml": `
+id: unrelated-profile
+endpoint: http://localhost:8000/v1
+model: unrelated
+unknown: true
+`,
+				"valid.yaml": validProfile,
+			},
+		},
+		{
+			name: "same stem unidentifiable yaml is unrelated",
+			files: map[string]string{
+				"selected-profile.yaml": "id: [",
+				"valid.yaml":            validProfile,
+			},
+		},
+		{
+			name: "same stem raw key with different id is unrelated",
+			files: map[string]string{
+				"selected-profile.yaml": `
+id: unrelated-profile
+endpoint: http://localhost:8000/v1
+model: unrelated
+api_key: secret
+`,
+				"valid.yaml": validProfile,
+			},
+		},
+		{
+			name: "leading and trailing whitespace is normalized",
+			files: map[string]string{
+				"padded.yaml": `
+id: "  selected-profile  "
+endpoint: http://localhost:8000/v1
+model: selected-model
+`,
+			},
+		},
+		{
+			name: "blank id is unrelated",
+			files: map[string]string{
+				"selected-profile.yaml": `
+id: "   "
+endpoint: http://localhost:8000/v1
+model: unrelated
+`,
+			},
+			wantErr: ErrProfileNotFound,
+		},
+		{
+			name: "normalized duplicates are ambiguous",
+			files: map[string]string{
+				"first.yaml": validProfile,
+				"nested/second.yaml": `
+id: " selected-profile "
+endpoint: http://localhost:8000/v1
+model: duplicate
+`,
+			},
+			wantErr:     ErrInvalidProfile,
+			diagnostics: []string{"duplicate execution profile id", "first.yaml", "nested/second.yaml"},
+		},
+		{
+			name: "selected unknown field is authoritative",
+			files: map[string]string{
+				"malformed.yaml": `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected-model
+unknown: true
+`,
+			},
+			wantErr:     ErrInvalidYAML,
+			diagnostics: []string{"malformed.yaml"},
+		},
+		{
+			name: "selected raw key is authoritative",
+			files: map[string]string{
+				"insecure.yaml": `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected-model
+api_key: secret
+`,
+			},
+			wantErr:     ErrRawAPIKeyNotAllowed,
+			diagnostics: []string{"insecure.yaml"},
+		},
+		{
+			name: "selected identity in an additional document is authoritative",
+			files: map[string]string{
+				"additional-document.yaml": `
+---
+---
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected-model
+`,
+			},
+			wantErr:     ErrInvalidYAML,
+			diagnostics: []string{"additional-document.yaml", "exactly one YAML document"},
+		},
+	}
+
+	for _, source := range profileRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, tc.files)
+				got, err := repo.GetProfile(context.Background(), " selected-profile ")
+				if tc.wantErr != nil {
+					if !errors.Is(err, tc.wantErr) {
+						t.Fatalf("expected %v, got %v", tc.wantErr, err)
+					}
+					for _, diagnostic := range tc.diagnostics {
+						if !strings.Contains(err.Error(), diagnostic) {
+							t.Fatalf("expected error to contain %q, got %v", diagnostic, err)
+						}
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load selected profile: %v", err)
+				}
+				if got.ID != "selected-profile" || got.Model != "selected-model" {
+					t.Fatalf("unexpected selected profile: %+v", got)
+				}
+			})
+		}
+	}
+}
+
+func TestProfileRepositoriesRequireOneYAMLDocument(t *testing.T) {
+	const profile = `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected-model
+`
+	tests := []struct {
+		name    string
+		suffix  string
+		wantErr bool
+	}{
+		{name: "comments and trailing whitespace", suffix: "\n# trailing comment\n\n"},
+		{name: "second populated document", suffix: "\n---\nid: another\n", wantErr: true},
+		{name: "second empty document", suffix: "\n---\n", wantErr: true},
+		{name: "malformed trailing yaml", suffix: "\n---\n[", wantErr: true},
+		{name: "raw key in trailing document", suffix: "\n---\napi_key: secret\n", wantErr: true},
+	}
+
+	for _, source := range profileRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, map[string]string{"definition.yaml": profile + tc.suffix})
+				got, err := repo.GetProfile(context.Background(), "selected-profile")
+				if tc.wantErr {
+					if !errors.Is(err, ErrInvalidYAML) {
+						t.Fatalf("expected ErrInvalidYAML, got %v", err)
+					}
+					if !strings.Contains(err.Error(), "definition.yaml") {
+						t.Fatalf("expected source path in error, got %v", err)
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load one-document profile: %v", err)
+				}
+				if got.ID != "selected-profile" {
+					t.Fatalf("unexpected profile: %+v", got)
+				}
+			})
+		}
+	}
+}
+
+func TestProfileRepositoriesPreserveOverlayFallbackRules(t *testing.T) {
+	fallback := staticProfileRepo{profiles: map[string]*domain.ExecutionProfile{
+		"selected-profile": {ID: "selected-profile", Endpoint: "http://fallback", Model: "fallback-model"},
+	}}
+	tests := []struct {
+		name      string
+		files     map[string]string
+		wantModel string
+		wantErr   error
+	}{
+		{
+			name: "same stem malformed different id falls back",
+			files: map[string]string{
+				"selected-profile.yaml": `
+id: unrelated-profile
+endpoint: http://localhost:8000/v1
+model: unrelated
+unknown: true
+`,
+			},
+			wantModel: "fallback-model",
+		},
+		{
+			name: "blank id falls back",
+			files: map[string]string{
+				"selected-profile.yaml": `
+id: "   "
+endpoint: http://localhost:8000/v1
+model: unrelated
+`,
+			},
+			wantModel: "fallback-model",
+		},
+		{
+			name: "selected malformed profile stops fallback",
+			files: map[string]string{
+				"other-name.yaml": `
+id: selected-profile
+endpoint: http://localhost:8000/v1
+model: selected
+unknown: true
+`,
+			},
+			wantErr: ErrInvalidYAML,
+		},
+	}
+
+	for _, source := range profileRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				primary := source.newRepository(t, tc.files)
+				got, err := NewOverlayRepository(primary, fallback).GetProfile(context.Background(), "selected-profile")
+				if tc.wantErr != nil {
+					if !errors.Is(err, tc.wantErr) {
+						t.Fatalf("expected %v, got %v", tc.wantErr, err)
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load fallback profile: %v", err)
+				}
+				if got.Model != tc.wantModel {
+					t.Fatalf("model = %q, want %q", got.Model, tc.wantModel)
+				}
+			})
+		}
+	}
+}
+
+func TestProfileRepositoryReadsSourcesFreshOnEveryLookup(t *testing.T) {
+	newSource := func() (*recordingProfileFS, Repository) {
+		fsys := &recordingProfileFS{FS: fstest.MapFS{
+			"target.yaml": profileMapFile(`
+id: target
+endpoint: http://localhost:8000/v1
+model: target-model
+`),
+			"unrelated.yaml": profileMapFile(`
+id: unrelated
+endpoint: http://localhost:8000/v1
+model: unrelated-model
+`),
+		}}
+		return fsys, NewFSRepository(fsys, ".")
+	}
+
+	t.Run("selected source", func(t *testing.T) {
+		fsys, repo := newSource()
+		for lookup := 1; lookup <= 2; lookup++ {
+			got, err := repo.GetProfile(context.Background(), "target")
+			if err != nil {
+				t.Fatalf("lookup %d: %v", lookup, err)
+			}
+			if got.Model != "target-model" {
+				t.Fatalf("lookup %d model = %q", lookup, got.Model)
+			}
+			for _, name := range []string{"target.yaml", "unrelated.yaml"} {
+				if count := fsys.openCount(name); count != lookup {
+					t.Fatalf("%s opens after lookup %d = %d, want %d", name, lookup, count, lookup)
+				}
+			}
+		}
+	})
+
+	t.Run("overlay fallthrough", func(t *testing.T) {
+		primaryFS := &recordingProfileFS{FS: fstest.MapFS{
+			"unrelated.yaml": profileMapFile(`
+id: unrelated
+endpoint: http://localhost:8000/v1
+model: unrelated-model
+`),
+		}}
+		fallbackFS, fallback := newSource()
+		repo := NewOverlayRepository(NewFSRepository(primaryFS, "."), fallback)
+
+		for lookup := 1; lookup <= 2; lookup++ {
+			got, err := repo.GetProfile(context.Background(), "target")
+			if err != nil {
+				t.Fatalf("lookup %d: %v", lookup, err)
+			}
+			if got.Model != "target-model" {
+				t.Fatalf("lookup %d model = %q", lookup, got.Model)
+			}
+			if count := primaryFS.openCount("unrelated.yaml"); count != lookup {
+				t.Fatalf("primary opens after lookup %d = %d, want %d", lookup, count, lookup)
+			}
+			if count := fallbackFS.openCount("target.yaml"); count != lookup {
+				t.Fatalf("fallback opens after lookup %d = %d, want %d", lookup, count, lookup)
+			}
+		}
+	})
+}
+
+func TestProfileRepositoriesRejectInvalidExecutionSettings(t *testing.T) {
+	ctx := context.Background()
+
+	t.Run("operating-system filesystem", func(t *testing.T) {
+		dir := t.TempDir()
+		writeProfileTestFile(t, filepath.Join(dir, "invalid.yaml"), `
+id: invalid
+endpoint: http://localhost:8000/v1
+model: model
+temperature: .nan
+`)
+
+		_, err := NewFilesystemRepository(dir).GetProfile(ctx, "invalid")
+		if !errors.Is(err, ErrInvalidProfile) {
+			t.Fatalf("expected ErrInvalidProfile, got %v", err)
+		}
+	})
+
+	t.Run("fs.FS", func(t *testing.T) {
+		repo := NewFSRepository(fstest.MapFS{
+			"profiles/invalid.yaml": profileMapFile(`
+id: invalid
+endpoint: http://localhost:8000/v1
+model: model
+top_p: .inf
+`),
+		}, "profiles")
+
+		_, err := repo.GetProfile(ctx, "invalid")
+		if !errors.Is(err, ErrInvalidProfile) {
+			t.Fatalf("expected ErrInvalidProfile, got %v", err)
+		}
+	})
+}
+
+func TestProfileRepositoriesValidateDerivedDefinitions(t *testing.T) {
+	tests := []struct {
+		name        string
+		files       map[string]string
+		wantError   error
+		wantBaseID  string
+		wantProfile bool
+	}{
+		{
+			name: "alias is locally valid and normalizes base id",
+			files: map[string]string{"alias.yaml": `
+id: selected-profile
+base_profile: "  base-profile  "
+`},
+			wantBaseID:  "base-profile",
+			wantProfile: true,
+		},
+		{
+			name: "derived endpoint remains valid",
+			files: map[string]string{"invalid.yaml": `
+id: selected-profile
+base_profile: base-profile
+endpoint: /v1
+`},
+			wantError: ErrInvalidProfile,
+		},
+		{
+			name: "derived settings remain valid",
+			files: map[string]string{"invalid.yaml": `
+id: selected-profile
+base_profile: base-profile
+top_p: 1.1
+`},
+			wantError: ErrInvalidProfile,
+		},
+		{
+			name: "derived extra params remain valid",
+			files: map[string]string{"invalid.yaml": `
+id: selected-profile
+base_profile: base-profile
+extra_params:
+  timestamp: 2026-08-11T12:34:56Z
+`},
+			wantError: ErrInvalidProfile,
+		},
+		{
+			name: "derived raw key remains prohibited",
+			files: map[string]string{"invalid.yaml": `
+id: selected-profile
+base_profile: base-profile
+api_key: secret
+`},
+			wantError: ErrRawAPIKeyNotAllowed,
+		},
+		{
+			name: "derived duplicate id remains invalid",
+			files: map[string]string{
+				"first.yaml":  "id: selected-profile\nbase_profile: first-base\n",
+				"second.yaml": "id: selected-profile\nbase_profile: second-base\n",
+			},
+			wantError: ErrInvalidProfile,
+		},
+		{
+			name: "derived extra document remains invalid",
+			files: map[string]string{"invalid.yaml": `
+id: selected-profile
+base_profile: base-profile
+---
+id: other
+`},
+			wantError: ErrInvalidYAML,
+		},
+		{
+			name:      "standalone profile remains complete",
+			files:     map[string]string{"invalid.yaml": "id: selected-profile\n"},
+			wantError: ErrInvalidProfile,
+		},
+	}
+
+	for _, source := range profileRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, tc.files)
+				got, err := repo.GetProfile(context.Background(), "selected-profile")
+				if tc.wantError != nil {
+					if !errors.Is(err, tc.wantError) {
+						t.Fatalf("error = %v, want %v", err, tc.wantError)
+					}
+					return
+				}
+				if err != nil || !tc.wantProfile {
+					t.Fatalf("profile = %+v, error = %v, want valid derived definition", got, err)
+				}
+				if got.BaseProfileID != tc.wantBaseID {
+					t.Fatalf("BaseProfileID = %q, want %q", got.BaseProfileID, tc.wantBaseID)
+				}
+			})
+		}
+	}
+}
+
 func TestOverlayRepository(t *testing.T) {
 	ctx := context.Background()
 	primaryProfile := &domain.ExecutionProfile{ID: "shared", Endpoint: "http://primary", Model: "primary"}
@@ -495,6 +1048,77 @@ func TestOverlayRepository(t *testing.T) {
 	})
 }
 
+type profileRepositorySource struct {
+	name          string
+	newRepository func(t *testing.T, files map[string]string) Repository
+}
+
+type recordingProfileFS struct {
+	fs.FS
+	mu     sync.Mutex
+	opened []string
+}
+
+func (f *recordingProfileFS) Open(name string) (fs.File, error) {
+	f.mu.Lock()
+	f.opened = append(f.opened, name)
+	f.mu.Unlock()
+	return f.FS.Open(name)
+}
+
+func (f *recordingProfileFS) openCount(name string) int {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	count := 0
+	for _, opened := range f.opened {
+		if opened == name {
+			count++
+		}
+	}
+	return count
+}
+
+func profileRepositorySources() []profileRepositorySource {
+	return []profileRepositorySource{
+		{
+			name: "operating system",
+			newRepository: func(t *testing.T, files map[string]string) Repository {
+				t.Helper()
+				root := t.TempDir()
+				for name, content := range files {
+					filePath := filepath.Join(root, filepath.FromSlash(name))
+					if err := os.MkdirAll(filepath.Dir(filePath), 0o755); err != nil {
+						t.Fatalf("create profile directory: %v", err)
+					}
+					writeProfileTestFile(t, filePath, content)
+				}
+				return NewFilesystemRepository(root)
+			},
+		},
+		{
+			name: "filesystem",
+			newRepository: func(t *testing.T, files map[string]string) Repository {
+				t.Helper()
+				fsys := make(fstest.MapFS, len(files))
+				for name, content := range files {
+					fsys[name] = profileMapFile(content)
+				}
+				return NewFSRepository(fsys, ".")
+			},
+		},
+	}
+}
+
+func deeplyNestedExtraParamsProfile(depth int) string {
+	var definition strings.Builder
+	definition.WriteString("id: selected-profile\nendpoint: http://localhost:8000/v1\nmodel: model\nextra_params:\n")
+	for level := 0; level < depth; level++ {
+		fmt.Fprintf(&definition, "%slevel_%d:\n", strings.Repeat("  ", level+1), level)
+	}
+	fmt.Fprintf(&definition, "%svalue: true\n", strings.Repeat("  ", depth+1))
+	return definition.String()
+}
+
 func profileMapFile(content string) *fstest.MapFile {
 	return &fstest.MapFile{Data: []byte(strings.TrimLeft(content, "\n"))}
 }

internal/profile/resolving_repository.go

@@ -0,0 +1,160 @@
+package profile
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
+)
+
+const maximumProfileChainLength = 32
+
+type resolvingRepository struct {
+	source Repository
+}
+
+// NewResolvingRepository resolves inherited profile definitions from source.
+func NewResolvingRepository(source Repository) Repository {
+	return &resolvingRepository{source: source}
+}
+
+func (r *resolvingRepository) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	if r == nil || r.source == nil {
+		return nil, fmt.Errorf("%w: profile repository is required", ErrInvalidProfile)
+	}
+	requestedID := strings.TrimSpace(id)
+	if requestedID == "" {
+		return nil, fmt.Errorf("%w: profile id is required", ErrInvalidProfile)
+	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+
+	profile, err := r.getRawProfile(ctx, requestedID)
+	if err != nil {
+		return nil, err
+	}
+	if profile == nil {
+		return nil, fmt.Errorf("%w: selected profile %q is nil", ErrInvalidProfile, requestedID)
+	}
+
+	chain := []*domain.ExecutionProfile{profile}
+	chainIDs := []string{requestedID}
+	visited := map[string]struct{}{requestedID: {}}
+	current := profile
+
+	for {
+		baseID := strings.TrimSpace(current.BaseProfileID)
+		if baseID == "" {
+			break
+		}
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		if _, seen := visited[baseID]; seen {
+			return nil, fmt.Errorf("%w: profile inheritance cycle %s", ErrInvalidProfile, joinProfileChain(chainIDs, baseID))
+		}
+		if len(chain) >= maximumProfileChainLength {
+			return nil, fmt.Errorf("%w: profile inheritance chain exceeds %d profiles: %s", ErrInvalidProfile, maximumProfileChainLength, joinProfileChain(chainIDs, baseID))
+		}
+
+		base, err := r.getRawProfile(ctx, baseID)
+		if err != nil {
+			if errors.Is(err, ErrProfileNotFound) {
+				return nil, fmt.Errorf("%w: base profile %q is missing in chain %s", ErrInvalidProfile, baseID, joinProfileChain(chainIDs, baseID))
+			}
+			return nil, fmt.Errorf("%w: failed to load base profile %q in chain %s: %w", ErrInvalidProfile, baseID, joinProfileChain(chainIDs, baseID), err)
+		}
+		if base == nil {
+			return nil, fmt.Errorf("%w: base profile %q is nil in chain %s", ErrInvalidProfile, baseID, joinProfileChain(chainIDs, baseID))
+		}
+
+		chain = append(chain, base)
+		chainIDs = append(chainIDs, baseID)
+		visited[baseID] = struct{}{}
+		current = base
+	}
+
+	resolved, err := mergeProfileChain(chain)
+	if err != nil {
+		return nil, fmt.Errorf("%w: resolved profile chain %s: %w", ErrInvalidProfile, strings.Join(chainIDs, " -> "), err)
+	}
+	if err := validateResolvedProfile(resolved); err != nil {
+		return nil, fmt.Errorf("%w: resolved profile chain %s: %w", ErrInvalidProfile, strings.Join(chainIDs, " -> "), err)
+	}
+	return resolved, nil
+}
+
+func (r *resolvingRepository) getRawProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	profile, err := r.source.GetProfile(ctx, id)
+	if err != nil {
+		return nil, err
+	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	return profile, nil
+}
+
+func joinProfileChain(chain []string, next string) string {
+	return strings.Join(append(append([]string(nil), chain...), next), " -> ")
+}
+
+func mergeProfileChain(chain []*domain.ExecutionProfile) (*domain.ExecutionProfile, error) {
+	resolved := &domain.ExecutionProfile{ID: chain[0].ID}
+	for index := len(chain) - 1; index >= 0; index-- {
+		definition := chain[index]
+		if strings.TrimSpace(definition.BackendID) != "" {
+			resolved.BackendID = definition.BackendID
+		}
+		if strings.TrimSpace(definition.Endpoint) != "" {
+			resolved.Endpoint = definition.Endpoint
+		}
+		if strings.TrimSpace(definition.Model) != "" {
+			resolved.Model = definition.Model
+		}
+		if definition.Temperature != 0 {
+			resolved.Temperature = definition.Temperature
+		}
+		if definition.MaxTokens != 0 {
+			resolved.MaxTokens = definition.MaxTokens
+		}
+		if definition.TopP != 0 {
+			resolved.TopP = definition.TopP
+		}
+		if definition.TimeoutSeconds != 0 {
+			resolved.TimeoutSeconds = definition.TimeoutSeconds
+		}
+		if strings.TrimSpace(definition.ServiceTier) != "" {
+			resolved.ServiceTier = definition.ServiceTier
+		}
+		if strings.TrimSpace(definition.ReasoningEffort) != "" {
+			resolved.ReasoningEffort = definition.ReasoningEffort
+		}
+		if strings.TrimSpace(definition.APIKeyEnv) != "" {
+			resolved.APIKeyEnv = definition.APIKeyEnv
+		}
+		resolved.APIKeyRequired = resolved.APIKeyRequired || definition.APIKeyRequired
+		if len(definition.ExtraParams) != 0 {
+			extraParams, err := jsonvalue.CopyMap(definition.ExtraParams)
+			if err != nil {
+				return nil, err
+			}
+			resolved.ExtraParams = extraParams
+		}
+	}
+	resolved.ID = chain[0].ID
+	resolved.BaseProfileID = ""
+	return resolved, nil
+}
+
+func validateResolvedProfile(profile *domain.ExecutionProfile) error {
+	if profile == nil {
+		return errors.New("resolved profile is required")
+	}
+	profile.BaseProfileID = ""
+	return NormalizeAndValidateDefinition(profile)
+}

internal/profile/resolving_repository_test.go

@@ -0,0 +1,418 @@
+package profile
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io/fs"
+	"reflect"
+	"strings"
+	"sync"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestResolvingRepositoryMergesProfileChain(t *testing.T) {
+	repo := &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+		"leaf": {
+			ID:              "leaf",
+			BaseProfileID:   "middle",
+			BackendID:       "leaf-backend",
+			TopP:            0.8,
+			TimeoutSeconds:  45,
+			ReasoningEffort: "high",
+		},
+		"middle": {
+			ID:             "middle",
+			BaseProfileID:  "root",
+			Endpoint:       "https://middle.example/v1",
+			Model:          "middle-model",
+			MaxTokens:      256,
+			APIKeyEnv:      "MIDDLE_API_KEY",
+			APIKeyRequired: true,
+			ExtraParams:    map[string]any{"middle": map[string]any{"value": "middle"}},
+		},
+		"root": {
+			ID:          "root",
+			BackendID:   "root-backend",
+			Endpoint:    "https://root.example/v1",
+			Model:       "root-model",
+			Temperature: 0.3,
+			ServiceTier: "priority",
+			ExtraParams: map[string]any{"root": "value"},
+		},
+	}}
+
+	got, err := NewResolvingRepository(repo).GetProfile(context.Background(), "leaf")
+	if err != nil {
+		t.Fatalf("resolve profile: %v", err)
+	}
+	want := &domain.ExecutionProfile{
+		ID:              "leaf",
+		BackendID:       "leaf-backend",
+		Endpoint:        "https://middle.example/v1",
+		Model:           "middle-model",
+		Temperature:     0.3,
+		MaxTokens:       256,
+		TopP:            0.8,
+		TimeoutSeconds:  45,
+		ServiceTier:     "priority",
+		ReasoningEffort: "high",
+		APIKeyEnv:       "MIDDLE_API_KEY",
+		APIKeyRequired:  true,
+		ExtraParams:     map[string]any{"middle": map[string]any{"value": "middle"}},
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("resolved profile:\n got %#v\nwant %#v", got, want)
+	}
+}
+
+func TestResolvingRepositoryRejectsMissingSourceAndProfileID(t *testing.T) {
+	if _, err := NewResolvingRepository(nil).GetProfile(context.Background(), "profile"); !errors.Is(err, ErrInvalidProfile) {
+		t.Fatalf("nil source error = %v, want ErrInvalidProfile", err)
+	}
+
+	repo := &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{}}
+	if _, err := NewResolvingRepository(repo).GetProfile(context.Background(), " \t "); !errors.Is(err, ErrInvalidProfile) {
+		t.Fatalf("blank id error = %v, want ErrInvalidProfile", err)
+	}
+	if got := repo.callCount(" "); got != 0 {
+		t.Fatalf("blank id looked up source %d times", got)
+	}
+}
+
+func TestResolvingRepositoryCopiesExtraParams(t *testing.T) {
+	baseParams := map[string]any{"nested": map[string]any{"value": "base"}}
+	repo := &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+		"child": {ID: "child", BaseProfileID: "base"},
+		"base": {
+			ID:          "base",
+			Endpoint:    "https://base.example/v1",
+			Model:       "model",
+			ExtraParams: baseParams,
+		},
+	}}
+	resolver := NewResolvingRepository(repo)
+
+	first, err := resolver.GetProfile(context.Background(), "child")
+	if err != nil {
+		t.Fatalf("resolve inherited map: %v", err)
+	}
+	first.ExtraParams["nested"].(map[string]any)["value"] = "mutated"
+	second, err := resolver.GetProfile(context.Background(), "child")
+	if err != nil {
+		t.Fatalf("resolve inherited map again: %v", err)
+	}
+	if got := second.ExtraParams["nested"].(map[string]any)["value"]; got != "base" {
+		t.Fatalf("later result retained mutation: %v", got)
+	}
+	if got := baseParams["nested"].(map[string]any)["value"]; got != "base" {
+		t.Fatalf("source map retained mutation: %v", got)
+	}
+
+	repo.set("child", &domain.ExecutionProfile{
+		ID:            "child",
+		BaseProfileID: "base",
+		ExtraParams:   map[string]any{"child": "replacement"},
+	})
+	replaced, err := resolver.GetProfile(context.Background(), "child")
+	if err != nil {
+		t.Fatalf("resolve replacement map: %v", err)
+	}
+	if !reflect.DeepEqual(replaced.ExtraParams, map[string]any{"child": "replacement"}) {
+		t.Fatalf("extra params = %#v, want complete child replacement", replaced.ExtraParams)
+	}
+}
+
+func TestResolvingRepositoryUsesRawOverlayForEachLookup(t *testing.T) {
+	leafSource := NewFSRepository(profileTestFS(map[string]string{
+		"leaf.yaml": "id: leaf\nbase_profile: base\n",
+	}), ".")
+	fallback := NewFSRepository(profileTestFS(map[string]string{
+		"base.yaml": "id: base\nendpoint: https://fallback.example/v1\nmodel: fallback-model\n",
+	}), ".")
+	overlay := NewOverlayRepository(leafSource, fallback)
+	resolver := NewResolvingRepository(overlay)
+
+	got, err := resolver.GetProfile(context.Background(), "leaf")
+	if err != nil {
+		t.Fatalf("resolve fallback base: %v", err)
+	}
+	if got.Model != "fallback-model" {
+		t.Fatalf("fallback base model = %q", got.Model)
+	}
+
+	shadowing := NewOverlayRepository(NewFSRepository(profileTestFS(map[string]string{
+		"leaf.yaml": "id: leaf\nbase_profile: base\n",
+		"base.yaml": "id: base\nendpoint: https://primary.example/v1\nmodel: primary-model\n",
+	}), "."), fallback)
+	got, err = NewResolvingRepository(shadowing).GetProfile(context.Background(), "leaf")
+	if err != nil {
+		t.Fatalf("resolve shadowed base: %v", err)
+	}
+	if got.Model != "primary-model" || got.Endpoint != "https://primary.example/v1" {
+		t.Fatalf("shadowed base = %+v", got)
+	}
+}
+
+func TestResolvingRepositoryReportsSafetyAndSourceErrors(t *testing.T) {
+	sourceErr := errors.New("source failure")
+	tests := []struct {
+		name     string
+		repo     *resolvingTestRepository
+		id       string
+		want     []error
+		wantNot  error
+		contains []string
+	}{
+		{
+			name:    "missing selected profile preserves not found",
+			repo:    &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{}},
+			id:      "missing",
+			want:    []error{ErrProfileNotFound},
+			wantNot: ErrInvalidProfile,
+		},
+		{
+			name: "missing base is invalid but not not found",
+			repo: &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+				"leaf": {ID: "leaf", BaseProfileID: "missing"},
+			}},
+			id:       "leaf",
+			want:     []error{ErrInvalidProfile},
+			wantNot:  ErrProfileNotFound,
+			contains: []string{"missing", "leaf -> missing"},
+		},
+		{
+			name: "direct cycle",
+			repo: &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+				"a": {ID: "a", BaseProfileID: "a"},
+			}},
+			id:       "a",
+			want:     []error{ErrInvalidProfile},
+			contains: []string{"a -> a"},
+		},
+		{
+			name: "indirect cycle",
+			repo: &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+				"a": {ID: "a", BaseProfileID: "b"},
+				"b": {ID: "b", BaseProfileID: "c"},
+				"c": {ID: "c", BaseProfileID: "a"},
+			}},
+			id:       "a",
+			want:     []error{ErrInvalidProfile},
+			contains: []string{"a -> b -> c -> a"},
+		},
+		{
+			name: "nil result",
+			repo: &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+				"leaf": nil,
+			}},
+			id:   "leaf",
+			want: []error{ErrInvalidProfile},
+		},
+		{
+			name: "incomplete resolved profile",
+			repo: &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+				"leaf": {ID: "leaf", BaseProfileID: "base"},
+				"base": {ID: "base", Model: "model"},
+			}},
+			id:   "leaf",
+			want: []error{ErrInvalidProfile},
+		},
+		{
+			name: "base source error is retained",
+			repo: &resolvingTestRepository{
+				profiles: map[string]*domain.ExecutionProfile{"leaf": {ID: "leaf", BaseProfileID: "base"}},
+				errors:   map[string]error{"base": sourceErr},
+			},
+			id:   "leaf",
+			want: []error{ErrInvalidProfile, sourceErr},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := NewResolvingRepository(tc.repo).GetProfile(context.Background(), tc.id)
+			for _, want := range tc.want {
+				if !errors.Is(err, want) {
+					t.Fatalf("error = %v, want %v", err, want)
+				}
+			}
+			if tc.wantNot != nil && errors.Is(err, tc.wantNot) {
+				t.Fatalf("error = %v, must not match %v", err, tc.wantNot)
+			}
+			for _, fragment := range tc.contains {
+				if !strings.Contains(err.Error(), fragment) {
+					t.Fatalf("error = %v, want %q", err, fragment)
+				}
+			}
+		})
+	}
+}
+
+func TestResolvingRepositoryEnforcesChainLength(t *testing.T) {
+	for _, count := range []int{maximumProfileChainLength, maximumProfileChainLength + 1} {
+		t.Run(fmt.Sprintf("%d profiles", count), func(t *testing.T) {
+			profiles := make(map[string]*domain.ExecutionProfile, count)
+			for index := 1; index <= count; index++ {
+				id := fmt.Sprintf("profile-%d", index)
+				definition := &domain.ExecutionProfile{ID: id}
+				if index == count {
+					definition.Endpoint = "https://root.example/v1"
+					definition.Model = "model"
+				} else {
+					definition.BaseProfileID = fmt.Sprintf("profile-%d", index+1)
+				}
+				profiles[id] = definition
+			}
+
+			got, err := NewResolvingRepository(&resolvingTestRepository{profiles: profiles}).GetProfile(context.Background(), "profile-1")
+			if count == maximumProfileChainLength {
+				if err != nil || got == nil {
+					t.Fatalf("profile = %+v, error = %v, want accepted chain", got, err)
+				}
+				return
+			}
+			if !errors.Is(err, ErrInvalidProfile) {
+				t.Fatalf("error = %v, want ErrInvalidProfile", err)
+			}
+		})
+	}
+}
+
+func TestResolvingRepositoryIsFreshAndCancellationAware(t *testing.T) {
+	repo := &resolvingTestRepository{profiles: map[string]*domain.ExecutionProfile{
+		"leaf": {ID: "leaf", BaseProfileID: "base"},
+		"base": {ID: "base", Endpoint: "https://base.example/v1", Model: "first", ExtraParams: map[string]any{"nested": map[string]any{"value": "first"}}},
+	}}
+	resolver := NewResolvingRepository(repo)
+
+	first, err := resolver.GetProfile(context.Background(), "leaf")
+	if err != nil || first.Model != "first" {
+		t.Fatalf("first result=(%+v, %v)", first, err)
+	}
+	repo.set("base", &domain.ExecutionProfile{ID: "base", Endpoint: "https://base.example/v1", Model: "second", ExtraParams: map[string]any{"nested": map[string]any{"value": "second"}}})
+	second, err := resolver.GetProfile(context.Background(), "leaf")
+	if err != nil || second.Model != "second" {
+		t.Fatalf("second result=(%+v, %v)", second, err)
+	}
+
+	canceled, cancel := context.WithCancel(context.Background())
+	cancel()
+	if _, err := resolver.GetProfile(canceled, "leaf"); !errors.Is(err, context.Canceled) {
+		t.Fatalf("canceled lookup error = %v", err)
+	}
+	if got := repo.callCount("leaf"); got != 2 {
+		t.Fatalf("calls after canceled lookup = %d, want 2", got)
+	}
+
+	duringTraversal, cancelDuringTraversal := context.WithCancel(context.Background())
+	repo.afterGet = func(id string) {
+		if id == "leaf" {
+			cancelDuringTraversal()
+		}
+	}
+	if _, err := resolver.GetProfile(duringTraversal, "leaf"); !errors.Is(err, context.Canceled) {
+		t.Fatalf("during traversal error = %v", err)
+	}
+	if got := repo.callCount("base"); got != 2 {
+		t.Fatalf("base calls after cancellation = %d, want 2", got)
+	}
+
+	terminalLookup, cancelTerminalLookup := context.WithCancel(context.Background())
+	repo.afterGet = func(id string) {
+		if id == "base" {
+			cancelTerminalLookup()
+		}
+	}
+	if _, err := resolver.GetProfile(terminalLookup, "leaf"); !errors.Is(err, context.Canceled) {
+		t.Fatalf("terminal lookup cancellation error = %v", err)
+	}
+	if got := repo.callCount("base"); got != 3 {
+		t.Fatalf("base calls after terminal cancellation = %d, want 3", got)
+	}
+
+	repo.afterGet = nil
+	var wg sync.WaitGroup
+	errors := make(chan error, 8)
+	for index := 0; index < cap(errors); index++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			resolved, err := resolver.GetProfile(context.Background(), "leaf")
+			if err != nil {
+				errors <- err
+				return
+			}
+			resolved.ExtraParams["nested"].(map[string]any)["value"] = "mutated"
+		}()
+	}
+	wg.Wait()
+	close(errors)
+	for err := range errors {
+		t.Errorf("concurrent resolution: %v", err)
+	}
+	latest, err := resolver.GetProfile(context.Background(), "leaf")
+	if err != nil || latest.ExtraParams["nested"].(map[string]any)["value"] != "second" {
+		t.Fatalf("latest result=(%+v, %v)", latest, err)
+	}
+}
+
+type resolvingTestRepository struct {
+	mu       sync.Mutex
+	profiles map[string]*domain.ExecutionProfile
+	errors   map[string]error
+	calls    map[string]int
+	afterGet func(string)
+}
+
+func (r *resolvingTestRepository) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	r.mu.Lock()
+	if r.calls == nil {
+		r.calls = make(map[string]int)
+	}
+	r.calls[id]++
+	err := r.errors[id]
+	profile := r.profiles[id]
+	afterGet := r.afterGet
+	r.mu.Unlock()
+	if afterGet != nil {
+		afterGet(id)
+	}
+	if err != nil {
+		return nil, err
+	}
+	if profile == nil {
+		if _, exists := r.profiles[id]; exists {
+			return nil, nil
+		}
+		return nil, ErrProfileNotFound
+	}
+	copy := *profile
+	return &copy, nil
+}
+
+func (r *resolvingTestRepository) set(id string, profile *domain.ExecutionProfile) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	r.profiles[id] = profile
+}
+
+func (r *resolvingTestRepository) callCount(id string) int {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	return r.calls[id]
+}
+
+func profileTestFS(files map[string]string) fs.FS {
+	fsys := make(fstest.MapFS, len(files))
+	for name, content := range files {
+		fsys[name] = profileMapFile(content)
+	}
+	return fsys
+}

internal/prompt/go_renderer.go

@@ -5,6 +5,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"strings"
 	"text/template"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
@@ -18,6 +19,8 @@ var (
 	ErrInvalidMessageRole   = errors.New("invalid or empty message role")
 )
 
+const artifactTextChunkSize = 64 * 1024
+
 type goRenderer struct{}
 
 func NewGoRenderer() Renderer {
@@ -25,11 +28,13 @@ func NewGoRenderer() Renderer {
 }
 
 func (r *goRenderer) Render(ctx context.Context, definition *domain.PromptDefinition, inputs map[string]*domain.Artifact, vars map[string]string) (*domain.RenderedPrompt, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
 	if definition == nil {
 		return nil, fmt.Errorf("%w: nil prompt definition", ErrRenderFailure)
 	}
 
-	// 1. Verify required inputs
 	for _, in := range definition.Inputs {
 		if !in.Required {
 			continue
@@ -40,44 +45,54 @@ func (r *goRenderer) Render(ctx context.Context, definition *domain.PromptDefini
 		}
 	}
 
-	// 2. Setup template functions
+	resolver := newArtifactTextResolver(ctx, inputs)
 	funcs := template.FuncMap{
-		"input": func(name string) (string, error) {
+		"input": resolver.resolve,
-			art, ok := inputs[name]
-			if !ok || art == nil {
-				return "", fmt.Errorf("%w: %s", ErrUnknownInput, name)
-			}
-			return string(art.Body), nil
-		},
 	}
 
-	sessionID, err := renderSessionID(definition.SessionID, funcs, vars)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	sessionID, err := renderSessionID(ctx, definition.SessionID, funcs, vars)
 	if err != nil {
 		return nil, err
 	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
 
-	var renderedMessages []domain.RenderedMessage
+	renderedMessages := make([]domain.RenderedMessage, 0, len(definition.Templates))
 
 	for i, tmplMsg := range definition.Templates {
-		select {
+		if err := ctx.Err(); err != nil {
-		case <-ctx.Done():
+			return nil, err
-			return nil, ctx.Err()
-		default:
 		}
 
 		if tmplMsg.Role == "" {
 			return nil, fmt.Errorf("%w: message %d", ErrInvalidMessageRole, i)
 		}
 
-		// Parse and execute template
+		if err := ctx.Err(); err != nil {
-		tmpl, err := template.New(fmt.Sprintf("msg_%d", i)).Funcs(funcs).Option("missingkey=error").Parse(tmplMsg.Content)
+			return nil, err
-		if err != nil {
+		}
-			return nil, fmt.Errorf("%w: message %d: %v", ErrInvalidTemplate, i, err)
+		tmpl, parseErr := template.New(fmt.Sprintf("msg_%d", i)).Funcs(funcs).Option("missingkey=error").Parse(tmplMsg.Content)
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		if parseErr != nil {
+			return nil, fmt.Errorf("%w: message %d: %v", ErrInvalidTemplate, i, parseErr)
 		}
 
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
 		var buf bytes.Buffer
-		if err := tmpl.Execute(&buf, vars); err != nil {
+		executeErr := tmpl.Execute(&buf, vars)
-			return nil, fmt.Errorf("%w: message %d: %w", ErrRenderFailure, i, err)
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		if executeErr != nil {
+			return nil, fmt.Errorf("%w: message %d: %w", ErrRenderFailure, i, executeErr)
 		}
 
 		renderedMessages = append(renderedMessages, domain.RenderedMessage{
@@ -85,6 +100,13 @@ func (r *goRenderer) Render(ctx context.Context, definition *domain.PromptDefini
 			Content:      buf.String(),
 			CacheControl: cloneCacheControl(tmplMsg.CacheControl),
 		})
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+	}
+
+	if err := ctx.Err(); err != nil {
+		return nil, err
 	}
 
 	return &domain.RenderedPrompt{
@@ -93,15 +115,75 @@ func (r *goRenderer) Render(ctx context.Context, definition *domain.PromptDefini
 	}, nil
 }
 
-func renderSessionID(raw string, funcs template.FuncMap, vars map[string]string) (string, error) {
+type artifactTextResolver struct {
-	tmpl, err := template.New("session_id").Funcs(funcs).Option("missingkey=error").Parse(raw)
+	ctx        context.Context
-	if err != nil {
+	inputs     map[string]*domain.Artifact
-		return "", fmt.Errorf("%w: session_id: %v", ErrInvalidTemplate, err)
+	textByName map[string]string
+}
+
+func newArtifactTextResolver(ctx context.Context, inputs map[string]*domain.Artifact) *artifactTextResolver {
+	return &artifactTextResolver{
+		ctx:        ctx,
+		inputs:     inputs,
+		textByName: make(map[string]string),
+	}
+}
+
+func (r *artifactTextResolver) resolve(name string) (string, error) {
+	if err := r.ctx.Err(); err != nil {
+		return "", err
+	}
+	artifact, ok := r.inputs[name]
+	if !ok || artifact == nil {
+		return "", fmt.Errorf("%w: %s", ErrUnknownInput, name)
+	}
+	if text, ok := r.textByName[name]; ok {
+		return text, nil
 	}
 
+	var builder strings.Builder
+	builder.Grow(len(artifact.Body))
+	for start := 0; start < len(artifact.Body); start += artifactTextChunkSize {
+		if err := r.ctx.Err(); err != nil {
+			return "", err
+		}
+		end := min(start+artifactTextChunkSize, len(artifact.Body))
+		_, _ = builder.Write(artifact.Body[start:end])
+		if err := r.ctx.Err(); err != nil {
+			return "", err
+		}
+	}
+	if err := r.ctx.Err(); err != nil {
+		return "", err
+	}
+
+	text := builder.String()
+	r.textByName[name] = text
+	return text, nil
+}
+
+func renderSessionID(ctx context.Context, raw string, funcs template.FuncMap, vars map[string]string) (string, error) {
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
+	tmpl, parseErr := template.New("session_id").Funcs(funcs).Option("missingkey=error").Parse(raw)
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
+	if parseErr != nil {
+		return "", fmt.Errorf("%w: session_id: %v", ErrInvalidTemplate, parseErr)
+	}
+
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	var buf bytes.Buffer
-	if err := tmpl.Execute(&buf, vars); err != nil {
+	executeErr := tmpl.Execute(&buf, vars)
-		return "", fmt.Errorf("%w: session_id: %w", ErrRenderFailure, err)
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
+	if executeErr != nil {
+		return "", fmt.Errorf("%w: session_id: %w", ErrRenderFailure, executeErr)
 	}
 
 	sessionID, err := domain.NormalizeSessionID(buf.String())

internal/prompt/renderer_test.go

@@ -1,6 +1,7 @@
 package prompt
 
 import (
+	"bytes"
 	"context"
 	"errors"
 	"strings"
@@ -228,6 +229,24 @@ func TestGoRenderer_Render(t *testing.T) {
 		}
 	})
 
+	t.Run("malformed rendered session id fails rendering", func(t *testing.T) {
+		def := &domain.PromptDefinition{
+			SessionID: "{{ .session_id }}",
+			Inputs:    []domain.PromptInput{{Name: "transcript", Required: true}},
+			Templates: []domain.PromptMessageTemplate{
+				{Role: "system", Content: "Speak in a {{.tone}} tone."},
+			},
+		}
+
+		_, err := renderer.Render(ctx, def, inputs, map[string]string{
+			"tone":       "concise",
+			"session_id": "session" + string([]byte{0xff}),
+		})
+		if !errors.Is(err, ErrRenderFailure) {
+			t.Fatalf("expected ErrRenderFailure, got %v", err)
+		}
+	})
+
 	t.Run("inserting required input artifact", func(t *testing.T) {
 		def := &domain.PromptDefinition{
 			Inputs: []domain.PromptInput{{Name: "transcript", Required: true}},
@@ -343,3 +362,186 @@ func TestGoRenderer_Render(t *testing.T) {
 		}
 	})
 }
+
+func TestGoRendererCancellation(t *testing.T) {
+	t.Run("before session parsing", func(t *testing.T) {
+		definition := &domain.PromptDefinition{
+			SessionID: "{{ malformed",
+			Templates: []domain.PromptMessageTemplate{
+				{Role: "user", Content: "not rendered"},
+			},
+		}
+		ctx, cancel := context.WithCancel(context.Background())
+		cancel()
+
+		result, err := NewGoRenderer().Render(ctx, definition, nil, nil)
+		if result != nil || !errors.Is(err, context.Canceled) {
+			t.Fatalf("result=%#v err=%v, want nil/context.Canceled", result, err)
+		}
+		if errors.Is(err, ErrInvalidTemplate) {
+			t.Fatalf("pre-canceled render parsed the malformed session: %v", err)
+		}
+
+		result, err = NewGoRenderer().Render(context.Background(), definition, nil, nil)
+		if result != nil || !errors.Is(err, ErrInvalidTemplate) {
+			t.Fatalf("active render result=%#v err=%v, want nil/ErrInvalidTemplate", result, err)
+		}
+	})
+
+	t.Run("during artifact text conversion", func(t *testing.T) {
+		ctx := newCancelOnCheckContext(3)
+		body := bytes.Repeat([]byte("x"), artifactTextChunkSize*2)
+		original := append([]byte(nil), body...)
+		resolver := newArtifactTextResolver(ctx, map[string]*domain.Artifact{
+			"document": {Body: body},
+		})
+
+		text, err := resolver.resolve("document")
+		if text != "" || !errors.Is(err, context.Canceled) {
+			t.Fatalf("text length=%d err=%v, want empty/context.Canceled", len(text), err)
+		}
+		if _, published := resolver.textByName["document"]; published {
+			t.Fatal("canceled conversion published partial artifact text")
+		}
+		if !bytes.Equal(body, original) {
+			t.Fatal("resolver mutated the artifact body")
+		}
+	})
+
+	t.Run("after final message execution", func(t *testing.T) {
+		definition := &domain.PromptDefinition{
+			Templates: []domain.PromptMessageTemplate{
+				{Role: "user", Content: "fully rendered"},
+			},
+		}
+		counter := &checkCountingContext{Context: context.Background()}
+		if _, err := NewGoRenderer().Render(counter, definition, nil, nil); err != nil {
+			t.Fatalf("count render checkpoints: %v", err)
+		}
+
+		// The final three checks occur after template execution, after the
+		// message is assembled, and immediately before publication.
+		ctx := newCancelOnCheckContext(counter.checks - 2)
+		result, err := NewGoRenderer().Render(ctx, definition, nil, nil)
+		if result != nil || !errors.Is(err, context.Canceled) {
+			t.Fatalf("result=%#v err=%v, want nil/context.Canceled", result, err)
+		}
+	})
+}
+
+func TestGoRendererArtifactTextLifecycle(t *testing.T) {
+	body := []byte{'a', 0xff, 'b', 0xfe}
+	original := append([]byte(nil), body...)
+	artifact := &domain.Artifact{Body: body}
+	inputs := map[string]*domain.Artifact{"document": artifact}
+	definition := &domain.PromptDefinition{
+		Templates: []domain.PromptMessageTemplate{
+			{Role: "user", Content: "{{input \"document\"}}|{{input \"document\"}}"},
+		},
+	}
+
+	first, err := NewGoRenderer().Render(context.Background(), definition, inputs, nil)
+	if err != nil {
+		t.Fatalf("first render: %v", err)
+	}
+	wantFirst := append(append(append([]byte(nil), body...), '|'), body...)
+	if !bytes.Equal([]byte(first.Messages[0].Content), wantFirst) {
+		t.Fatalf("rendered bytes=%v, want %v", []byte(first.Messages[0].Content), wantFirst)
+	}
+	if !bytes.Equal(body, original) {
+		t.Fatalf("renderer mutated artifact body: got %v want %v", body, original)
+	}
+
+	body[0] = 'z'
+	if bytes.Equal([]byte(first.Messages[0].Content), append(append(append([]byte(nil), body...), '|'), body...)) {
+		t.Fatal("completed render aliases the artifact body")
+	}
+	second, err := NewGoRenderer().Render(context.Background(), definition, inputs, nil)
+	if err != nil {
+		t.Fatalf("second render: %v", err)
+	}
+	wantSecond := append(append(append([]byte(nil), body...), '|'), body...)
+	if !bytes.Equal([]byte(second.Messages[0].Content), wantSecond) {
+		t.Fatalf("second render reused text from another call: got %v want %v", []byte(second.Messages[0].Content), wantSecond)
+	}
+
+	nilInputs := map[string]*domain.Artifact{"document": nil}
+	result, err := NewGoRenderer().Render(context.Background(), definition, nilInputs, nil)
+	if result != nil || !errors.Is(err, ErrUnknownInput) || !errors.Is(err, ErrRenderFailure) {
+		t.Fatalf("nil input result=%#v err=%v, want ErrUnknownInput and ErrRenderFailure", result, err)
+	}
+}
+
+func BenchmarkGoRendererArtifactReferences(b *testing.B) {
+	body := bytes.Repeat([]byte("document content "), (artifactTextChunkSize*4)/len("document content "))
+	inputs := map[string]*domain.Artifact{
+		"document": {Body: body},
+	}
+	tests := []struct {
+		name       string
+		definition *domain.PromptDefinition
+	}{
+		{
+			name: "one reference",
+			definition: &domain.PromptDefinition{
+				Templates: []domain.PromptMessageTemplate{
+					{Role: "user", Content: "{{input \"document\"}}"},
+				},
+			},
+		},
+		{
+			name: "repeated across session and messages",
+			definition: &domain.PromptDefinition{
+				SessionID: "document-{{len (input \"document\")}}",
+				Templates: []domain.PromptMessageTemplate{
+					{Role: "system", Content: "{{input \"document\"}}"},
+					{Role: "user", Content: "{{input \"document\"}} {{input \"document\"}}"},
+				},
+			},
+		},
+	}
+
+	for _, tc := range tests {
+		b.Run(tc.name, func(b *testing.B) {
+			renderer := NewGoRenderer()
+			b.ReportAllocs()
+			b.SetBytes(int64(len(body)))
+			for range b.N {
+				if _, err := renderer.Render(context.Background(), tc.definition, inputs, nil); err != nil {
+					b.Fatal(err)
+				}
+			}
+		})
+	}
+}
+
+type checkCountingContext struct {
+	context.Context
+	checks int
+}
+
+func (c *checkCountingContext) Err() error {
+	c.checks++
+	return c.Context.Err()
+}
+
+type cancelOnCheckContext struct {
+	context.Context
+	cancel    context.CancelFunc
+	remaining int
+}
+
+func newCancelOnCheckContext(checks int) *cancelOnCheckContext {
+	ctx, cancel := context.WithCancel(context.Background())
+	return &cancelOnCheckContext{Context: ctx, cancel: cancel, remaining: checks}
+}
+
+func (c *cancelOnCheckContext) Err() error {
+	if c.Context.Err() == nil {
+		c.remaining--
+		if c.remaining == 0 {
+			c.cancel()
+		}
+	}
+	return c.Context.Err()
+}

internal/promptdef/content_source.go

@@ -0,0 +1,98 @@
+package promptdef
+
+import (
+	"fmt"
+	"io/fs"
+	"os"
+	"path"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
+)
+
+type contentSourceRoot interface {
+	readContentFile(sourcePath string, contentFile string) (string, string, error)
+}
+
+type osContentSourceRoot struct {
+	root                string
+	sourcePathsRelative bool
+}
+
+func (r osContentSourceRoot) readContentFile(sourcePath string, contentFile string) (string, string, error) {
+	if strings.TrimSpace(contentFile) == "" {
+		return "", "", fmt.Errorf("path is required")
+	}
+	if filepath.IsAbs(contentFile) {
+		return "", "", fmt.Errorf("path %q must be relative", contentFile)
+	}
+
+	root, err := filepath.Abs(r.root)
+	if err != nil {
+		return "", "", fmt.Errorf("resolve source root %q: %w", r.root, err)
+	}
+	canonicalRoot, err := filepath.EvalSymlinks(root)
+	if err != nil {
+		return "", "", fmt.Errorf("resolve source root %q: %w", r.root, err)
+	}
+
+	promptPath := sourcePath
+	if r.sourcePathsRelative && !filepath.IsAbs(promptPath) {
+		promptPath = filepath.Join(root, filepath.FromSlash(promptPath))
+	} else {
+		promptPath, err = filepath.Abs(promptPath)
+		if err != nil {
+			return "", "", fmt.Errorf("resolve prompt source %q: %w", sourcePath, err)
+		}
+	}
+	resolvedPath := filepath.Clean(filepath.Join(filepath.Dir(promptPath), contentFile))
+	if !containsOSPath(root, resolvedPath) {
+		return "", "", fmt.Errorf("path %q escapes source root %q", contentFile, r.root)
+	}
+
+	canonicalPath, err := filepath.EvalSymlinks(resolvedPath)
+	if err != nil {
+		return "", "", err
+	}
+	if !containsOSPath(canonicalRoot, canonicalPath) {
+		return "", "", fmt.Errorf("path %q escapes source root %q", contentFile, r.root)
+	}
+
+	body, err := os.ReadFile(canonicalPath)
+	if err != nil {
+		return "", "", err
+	}
+	return string(body), resolvedPath, nil
+}
+
+type fsContentSourceRoot struct {
+	fsys fs.FS
+	root string
+}
+
+func (r fsContentSourceRoot) readContentFile(sourcePath string, contentFile string) (string, string, error) {
+	root := filecatalog.CleanFSRoot(r.root)
+	cleanSourcePath := path.Clean(sourcePath)
+	if cleanSourcePath == root {
+		root = path.Dir(root)
+	}
+
+	resolvedPath, _, err := filecatalog.ResolveFSPath(root, path.Dir(cleanSourcePath), contentFile)
+	if err != nil {
+		return "", "", err
+	}
+	body, err := fs.ReadFile(r.fsys, resolvedPath)
+	if err != nil {
+		return "", "", err
+	}
+	return string(body), resolvedPath, nil
+}
+
+func containsOSPath(root string, name string) bool {
+	relative, err := filepath.Rel(root, name)
+	if err != nil || filepath.IsAbs(relative) {
+		return false
+	}
+	return relative != ".." && !strings.HasPrefix(relative, ".."+string(filepath.Separator))
+}

internal/promptdef/filesystem_repository.go

@@ -5,14 +5,11 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
-	"os"
-	"path"
-	"path/filepath"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
-	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
 	"gopkg.in/yaml.v3"
 )
 
@@ -22,13 +19,8 @@ var (
 	ErrInvalidPromptDefinition  = errors.New("invalid prompt definition configuration")
 )
 
-type filesystemRepository struct {
+type sourceRepository struct {
-	dir string
+	source promptDefinitionSource
-}
-
-type fsRepository struct {
-	fsys fs.FS
-	root string
 }
 
 type promptDefinitionFile struct {
@@ -69,19 +61,47 @@ type promptOutputContractFile struct {
 }
 
 func NewFilesystemRepository(dir string) Repository {
-	return &filesystemRepository{dir: dir}
+	return &sourceRepository{
+		source: osPromptSource{
+			root:        dir,
+			contentRoot: osContentSourceRoot{root: dir},
+		},
+	}
 }
 
 func NewFSRepository(fsys fs.FS, root string) Repository {
-	return &fsRepository{fsys: fsys, root: root}
+	return &sourceRepository{
+		source: fsPromptSource{
+			fsys:        fsys,
+			root:        root,
+			contentRoot: fsContentSourceRoot{fsys: fsys, root: root},
+		},
+	}
 }
 
-func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id string, version string) (*domain.PromptDefinition, error) {
+// NewFileRepository constructs a repository for one operating-system prompt file.
+func NewFileRepository(fsys fs.FS, file string, sourceDir string) Repository {
+	return &sourceRepository{
+		source: fsPromptSource{
+			fsys: fsys,
+			root: file,
+			contentRoot: osContentSourceRoot{
+				root:                sourceDir,
+				sourcePathsRelative: true,
+			},
+		},
+	}
+}
+
+func (r *sourceRepository) GetPromptDefinition(ctx context.Context, id string, version string) (*domain.PromptDefinition, error) {
 	if strings.TrimSpace(id) == "" {
 		return nil, fmt.Errorf("%w: prompt id is required", ErrInvalidPromptDefinition)
 	}
+	if r == nil || r.source == nil {
+		return nil, errors.New("failed to read prompt definition directory: source is nil")
+	}
 
-	files, err := filecatalog.FindYAMLFiles(ctx, r.dir)
+	files, err := r.source.findYAMLFiles(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read prompt definition directory: %w", err)
 	}
@@ -94,33 +114,24 @@ func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id strin
 		default:
 		}
 
-		relPath := filecatalog.RelativePath(r.dir, fullPath)
+		relPath := r.source.displayPath(fullPath)
-		fileMatch := filecatalog.Stem(filepath.Base(fullPath)) == id
+		data, err := r.source.readDefinition(fullPath)
-
-		raw, err := loadPromptDefinitionFile(fullPath)
 		if err != nil {
-			if fileMatch || promptDefinitionFileHasID(fullPath, id) {
+			return nil, fmt.Errorf("failed to read prompt definition file %s: %w", relPath, err)
+		}
+		raw, err := decodePromptDefinition(data)
+		if err != nil {
+			if promptDefinitionDataMatches(data, id, version) {
 				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
 			}
 			continue
 		}
-
+		if !promptDefinitionMatches(raw, id, version) {
-		def, err := normalizePromptDefinition(raw, fullPath)
-		if err != nil {
-			if fileMatch || strings.TrimSpace(raw.ID) == id {
-				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidPromptDefinition, relPath, err)
-			}
-			continue
-		}
-
-		if def.ID != id {
-			continue
-		}
-		if version != "" && def.Version != version {
 			continue
 		}
 		matches = append(matches, promptDefinitionMatch{
-			def:  def,
+			raw:        raw,
+			sourcePath: fullPath,
 			path:       relPath,
 		})
 	}
@@ -137,132 +148,23 @@ func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id strin
 	}
 
 	if len(matches) == 1 {
-		return matches[0].def, nil
+		match := matches[0]
+		def, err := normalizePromptDefinition(match.raw, r.source, match.sourcePath)
+		if err != nil {
+			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidPromptDefinition, match.path, err)
+		}
+		return def, nil
 	}
 
 	return nil, ErrPromptDefinitionNotFound
 }
 
-func (r *fsRepository) GetPromptDefinition(ctx context.Context, id string, version string) (*domain.PromptDefinition, error) {
-	return loadPromptDefinition(ctx, r.fsys, r.root, id, version)
-}
-
 type promptDefinitionMatch struct {
-	def  *domain.PromptDefinition
+	raw        *promptDefinitionFile
+	sourcePath string
 	path       string
 }
 
-func loadPromptDefinitionFile(path string) (*promptDefinitionFile, error) {
-	data, err := os.ReadFile(path)
-	if err != nil {
-		return nil, fmt.Errorf("failed to read prompt definition file: %w", err)
-	}
-
-	var raw promptDefinitionFile
-	decoder := yaml.NewDecoder(bytes.NewReader(data))
-	decoder.KnownFields(true)
-	if err := decoder.Decode(&raw); err != nil {
-		return nil, err
-	}
-	return &raw, nil
-}
-
-func promptDefinitionFileHasID(path string, id string) bool {
-	data, err := os.ReadFile(path)
-	if err != nil {
-		return false
-	}
-	var raw struct {
-		ID string `yaml:"id"`
-	}
-	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&raw); err != nil {
-		return false
-	}
-	return strings.TrimSpace(raw.ID) == id
-}
-
-func loadPromptDefinition(ctx context.Context, fsys fs.FS, root string, id string, version string) (*domain.PromptDefinition, error) {
-	if strings.TrimSpace(id) == "" {
-		return nil, fmt.Errorf("%w: prompt id is required", ErrInvalidPromptDefinition)
-	}
-	if fsys == nil {
-		return nil, fmt.Errorf("failed to read prompt definition directory: filesystem is nil")
-	}
-
-	files, err := filecatalog.FindFSYAMLFiles(ctx, fsys, root)
-	if err != nil {
-		return nil, fmt.Errorf("failed to read prompt definition directory: %w", err)
-	}
-	cleanRoot := filecatalog.CleanFSRoot(root)
-	rootInfo, err := fs.Stat(fsys, cleanRoot)
-	if err != nil {
-		return nil, fmt.Errorf("failed to read prompt definition directory: %w", err)
-	}
-
-	var matches []promptDefinitionMatch
-	for _, fullPath := range files {
-		select {
-		case <-ctx.Done():
-			return nil, ctx.Err()
-		default:
-		}
-
-		relPath := filecatalog.DisplayPath(root, fullPath)
-		fileMatch := filecatalog.Stem(path.Base(fullPath)) == id
-		data, err := fs.ReadFile(fsys, fullPath)
-		if err != nil {
-			if fileMatch {
-				return nil, fmt.Errorf("%w: %s: failed to read prompt definition file: %v", ErrInvalidYAML, relPath, err)
-			}
-			continue
-		}
-
-		raw, err := decodePromptDefinition(data)
-		if err != nil {
-			if fileMatch || promptDefinitionDataHasID(data, id) {
-				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
-			}
-			continue
-		}
-
-		def, err := normalizePromptDefinitionFromFS(raw, fsys, root, fullPath, rootInfo.IsDir())
-		if err != nil {
-			if fileMatch || strings.TrimSpace(raw.ID) == id {
-				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidPromptDefinition, relPath, err)
-			}
-			continue
-		}
-
-		if def.ID != id {
-			continue
-		}
-		if version != "" && def.Version != version {
-			continue
-		}
-		matches = append(matches, promptDefinitionMatch{
-			def:  def,
-			path: relPath,
-		})
-	}
-
-	if len(matches) > 1 {
-		paths := make([]string, 0, len(matches))
-		for _, match := range matches {
-			paths = append(paths, match.path)
-		}
-		if version != "" {
-			return nil, fmt.Errorf("%w: duplicate prompt definition id %q version %q found in: %s", ErrInvalidPromptDefinition, id, version, strings.Join(paths, ", "))
-		}
-		return nil, fmt.Errorf("%w: duplicate prompt definition id %q found in: %s", ErrInvalidPromptDefinition, id, strings.Join(paths, ", "))
-	}
-
-	if len(matches) == 1 {
-		return matches[0].def, nil
-	}
-
-	return nil, ErrPromptDefinitionNotFound
-}
-
 func decodePromptDefinition(data []byte) (*promptDefinitionFile, error) {
 	var raw promptDefinitionFile
 	decoder := yaml.NewDecoder(bytes.NewReader(data))
@@ -270,59 +172,44 @@ func decodePromptDefinition(data []byte) (*promptDefinitionFile, error) {
 	if err := decoder.Decode(&raw); err != nil {
 		return nil, err
 	}
+	var additional yaml.Node
+	if err := decoder.Decode(&additional); err != io.EOF {
+		if err != nil {
+			return nil, err
+		}
+		return nil, errors.New("prompt definition file must contain exactly one YAML document")
+	}
 	return &raw, nil
 }
 
-func promptDefinitionDataHasID(data []byte, id string) bool {
+func promptDefinitionDataMatches(data []byte, id string, version string) bool {
 	var raw struct {
 		ID      string `yaml:"id"`
+		Version string `yaml:"version"`
 	}
 	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&raw); err != nil {
 		return false
 	}
-	return strings.TrimSpace(raw.ID) == id
+	return promptSelectorMatches(raw.ID, raw.Version, id, version)
 }
 
-func normalizePromptDefinition(raw *promptDefinitionFile, sourcePath string) (*domain.PromptDefinition, error) {
+func promptDefinitionMatches(raw *promptDefinitionFile, id string, version string) bool {
-	promptDir := filepath.Dir(sourcePath)
+	if raw == nil {
-	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
+		return false
-		resolvedPath := strings.TrimSpace(contentFile)
-		if !filepath.IsAbs(resolvedPath) {
-			resolvedPath = filepath.Join(promptDir, resolvedPath)
 	}
-		resolvedPath = filepath.Clean(resolvedPath)
+	return promptSelectorMatches(raw.ID, raw.Version, id, version)
-
-		body, err := os.ReadFile(resolvedPath)
-		if err != nil {
-			return "", "", err
-		}
-		return string(body), resolvedPath, nil
-	})
 }
 
-func normalizePromptDefinitionFromFS(raw *promptDefinitionFile, fsys fs.FS, root string, sourcePath string, rootIsDir bool) (*domain.PromptDefinition, error) {
+func promptSelectorMatches(rawID string, rawVersion string, id string, version string) bool {
-	promptDir := path.Dir(sourcePath)
+	if strings.TrimSpace(rawID) != id {
-	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
+		return false
-		var resolvedPath string
-		if rootIsDir {
-			var err error
-			resolvedPath, _, err = filecatalog.ResolveFSPath(root, promptDir, contentFile)
-			if err != nil {
-				return "", "", err
-			}
-		} else {
-			resolvedPath = strings.TrimSpace(contentFile)
-			if !path.IsAbs(resolvedPath) {
-				resolvedPath = path.Join(promptDir, resolvedPath)
-			}
-			resolvedPath = strings.TrimPrefix(path.Clean(resolvedPath), "/")
 	}
+	return version == "" || strings.TrimSpace(rawVersion) == version
+}
 
-		body, err := fs.ReadFile(fsys, resolvedPath)
+func normalizePromptDefinition(raw *promptDefinitionFile, sourceRoot contentSourceRoot, sourcePath string) (*domain.PromptDefinition, error) {
-		if err != nil {
+	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
-			return "", "", err
+		return sourceRoot.readContentFile(sourcePath, contentFile)
-		}
-		return string(body), resolvedPath, nil
 	})
 }
 
@@ -402,17 +289,14 @@ func normalizePromptDefinitionWithContent(raw *promptDefinitionFile, readContent
 		})
 	}
 
-	if !isValidOutputFormat(raw.Output.Format) {
+	outputContract := domain.OutputContract{
-		return nil, fmt.Errorf("invalid output format: %q", raw.Output.Format)
+		Format:         raw.Output.Format,
+		ValidationMode: raw.Output.ValidationMode,
+		SchemaPath:     strings.TrimSpace(raw.Output.SchemaPath),
+		RepairAttempts: raw.Output.RepairAttempts,
 	}
-	if !isValidValidationMode(raw.Output.ValidationMode) {
+	if err := domain.ValidateOutputContract(outputContract); err != nil {
-		return nil, fmt.Errorf("invalid validation mode: %q", raw.Output.ValidationMode)
+		return nil, fmt.Errorf("output: %w", err)
-	}
-	if raw.Output.ValidationMode == domain.ValidationJSONSchema && strings.TrimSpace(raw.Output.SchemaPath) == "" {
-		return nil, errors.New("output.schema_path is required when output.validation_mode is json_schema")
-	}
-	if raw.Output.RepairAttempts < 0 {
-		return nil, errors.New("output.repair_attempts must be greater than or equal to 0")
 	}
 
 	defaultProfile := ""
@@ -432,12 +316,7 @@ func normalizePromptDefinitionWithContent(raw *promptDefinitionFile, readContent
 		Inputs:         inputs,
 		Templates:      templates,
 		OutputFormat:   raw.Output.Format,
-		Validation: domain.OutputContract{
+		Validation:     outputContract,
-			Format:         raw.Output.Format,
-			ValidationMode: raw.Output.ValidationMode,
-			SchemaPath:     strings.TrimSpace(raw.Output.SchemaPath),
-			RepairAttempts: raw.Output.RepairAttempts,
-		},
 	}, nil
 }
 
@@ -464,21 +343,3 @@ func normalizeCacheControl(raw *cacheControlFile) (*domain.CacheControl, error)
 		TTL:  ttl,
 	}, nil
 }
-
-func isValidOutputFormat(f domain.OutputFormat) bool {
-	switch f {
-	case domain.FormatText, domain.FormatMarkdown, domain.FormatJSON:
-		return true
-	default:
-		return false
-	}
-}
-
-func isValidValidationMode(m domain.ValidationMode) bool {
-	switch m {
-	case domain.ValidationNone, domain.ValidationBasic, domain.ValidationJSON, domain.ValidationJSONSchema:
-		return true
-	default:
-		return false
-	}
-}

internal/promptdef/repository_test.go

@@ -3,23 +3,51 @@ package promptdef
 import (
 	"context"
 	"errors"
+	"fmt"
 	"io/fs"
 	"os"
 	"path/filepath"
 	"strings"
+	"sync"
 	"testing"
 	"testing/fstest"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 )
 
-func TestFilesystemRepository_GetPromptDefinition(t *testing.T) {
+func TestPromptRepositoryDefinitionFixtures(t *testing.T) {
+	sources := []struct {
+		name                string
+		newRepository       func(string) Repository
+		contentPathsAreFull bool
+	}{
+		{
+			name:                "operating system",
+			newRepository:       NewFilesystemRepository,
+			contentPathsAreFull: true,
+		},
+		{
+			name: "filesystem",
+			newRepository: func(root string) Repository {
+				return NewFSRepository(os.DirFS(root), ".")
+			},
+		},
+	}
+
+	for _, source := range sources {
+		t.Run(source.name, func(t *testing.T) {
+			testPromptRepositoryDefinitionFixtures(t, source.newRepository, source.contentPathsAreFull)
+		})
+	}
+}
+
+func testPromptRepositoryDefinitionFixtures(t *testing.T, newRepository func(string) Repository, contentPathsAreFull bool) {
 	tmpDir := t.TempDir()
 	if err := copyTree("testdata", tmpDir); err != nil {
 		t.Fatalf("failed to copy testdata: %v", err)
 	}
 
-	repo := NewFilesystemRepository(tmpDir)
+	repo := newRepository(tmpDir)
 	ctx := context.Background()
 
 	t.Run("valid inline prompt", func(t *testing.T) {
@@ -64,8 +92,8 @@ func TestFilesystemRepository_GetPromptDefinition(t *testing.T) {
 		if p.Templates[1].ContentFile == "" {
 			t.Fatal("expected ContentFile source metadata to be preserved")
 		}
-		if !filepath.IsAbs(p.Templates[1].ContentFile) {
+		if filepath.IsAbs(p.Templates[1].ContentFile) != contentPathsAreFull {
-			t.Fatalf("expected resolved content_file path to be absolute, got %q", p.Templates[1].ContentFile)
+			t.Fatalf("unexpected content_file path representation: %q", p.Templates[1].ContentFile)
 		}
 	})
 
@@ -287,20 +315,20 @@ output:
 		targetErr  error
 		errSubstrs []string
 	}{
-		{name: "invalid YAML", id: "invalid_yaml", targetErr: ErrInvalidYAML},
+		{name: "unidentifiable invalid YAML is unrelated", id: "invalid-yaml", targetErr: ErrPromptDefinitionNotFound},
-		{name: "missing id", id: "missing_id", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"id is required"}},
+		{name: "missing id is not selected by filename", id: "missing_id", targetErr: ErrPromptDefinitionNotFound},
-		{name: "no messages", id: "no_messages", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"at least one message is required"}},
+		{name: "no messages", id: "no-messages", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"at least one message is required"}},
-		{name: "both content and content_file", id: "both_content_and_content_file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"exactly one"}},
+		{name: "both content and content_file", id: "both-content-and-content-file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"exactly one"}},
-		{name: "neither content nor content_file", id: "neither_content_nor_content_file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"exactly one"}},
+		{name: "neither content nor content_file", id: "neither-content-nor-content-file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"exactly one"}},
-		{name: "missing content_file", id: "missing_content_file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"failed to read content_file"}},
+		{name: "missing content_file", id: "missing-content-file", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"failed to read content_file"}},
-		{name: "duplicate input names", id: "duplicate_input_names", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"duplicate input name"}},
+		{name: "duplicate input names", id: "duplicate-input-names", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"duplicate input name"}},
-		{name: "invalid validation mode", id: "invalid_validation_mode", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"invalid validation mode"}},
+		{name: "invalid validation mode", id: "invalid-validation-mode", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"invalid validation mode"}},
-		{name: "json_schema without schema_path", id: "json_schema_without_schema_path", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"schema_path"}},
+		{name: "json_schema without schema_path", id: "json-schema-without-schema-path", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"schema_path"}},
-		{name: "unknown input field", id: "unknown_input_field", targetErr: ErrInvalidYAML, errSubstrs: []string{"field unknown_input_setting not found"}},
+		{name: "unknown input field", id: "unknown-input-field", targetErr: ErrInvalidYAML, errSubstrs: []string{"field unknown_input_setting not found"}},
-		{name: "empty cache control type", id: "empty_cache_control_type", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "type is required"}},
+		{name: "empty cache control type", id: "empty-cache-control-type", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "type is required"}},
-		{name: "unsupported cache control type", id: "unsupported_cache_control_type", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "unsupported type"}},
+		{name: "unsupported cache control type", id: "unsupported-cache-control-type", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "unsupported type"}},
-		{name: "unsupported cache control ttl", id: "unsupported_cache_control_ttl", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "unsupported ttl"}},
+		{name: "unsupported cache control ttl", id: "unsupported-cache-control-ttl", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"cache_control", "unsupported ttl"}},
-		{name: "unknown cache control field", id: "unknown_cache_control_field", targetErr: ErrInvalidYAML, errSubstrs: []string{"field unexpected not found"}},
+		{name: "unknown cache control field", id: "unknown-cache-control-field", targetErr: ErrInvalidYAML, errSubstrs: []string{"field unexpected not found"}},
 	}
 
 	for _, tc := range cases {
@@ -396,7 +424,7 @@ output:
 
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
-			repo := NewFSRepository(fstest.MapFS{
+			fsys := &recordingFS{FS: fstest.MapFS{
 				"prompts/prompt.yaml": &fstest.MapFile{Data: []byte(`
 id: fs-escaped-prompt
 version: "1.0.0"
@@ -409,7 +437,8 @@ output:
   repair_attempts: 0
 `)},
 				"outside.tmpl": &fstest.MapFile{Data: []byte(`Outside root.`)},
-			}, "prompts")
+			}}
+			repo := NewFSRepository(fsys, "prompts")
 
 			_, err := repo.GetPromptDefinition(context.Background(), "fs-escaped-prompt", "")
 			if !errors.Is(err, ErrInvalidPromptDefinition) {
@@ -418,65 +447,644 @@ output:
 			if !strings.Contains(err.Error(), tc.wantErr) {
 				t.Fatalf("expected error to contain %q, got %v", tc.wantErr, err)
 			}
+			if fsys.wasOpened("outside.tmpl") {
+				t.Fatal("rejected content path opened the outside file")
+			}
 		})
 	}
 }
 
-func TestFSRepositoryRejectsDuplicatePromptIDs(t *testing.T) {
+type recordingFS struct {
-	repo := NewFSRepository(fstest.MapFS{
+	fs.FS
-		"one.yaml": &fstest.MapFile{Data: []byte(`
+	mu     sync.Mutex
-id: duplicate-fs-prompt
+	opened []string
-version: "1.0.0"
+}
-messages:
-  - role: user
-    content: First.
-output:
-  format: text
-  validation_mode: none
-  repair_attempts: 0
-`)},
-		"nested/two.yaml": &fstest.MapFile{Data: []byte(`
-id: duplicate-fs-prompt
-version: "1.0.0"
-messages:
-  - role: user
-    content: Second.
-output:
-  format: text
-  validation_mode: none
-  repair_attempts: 0
-`)},
-	}, ".")
 
-	_, err := repo.GetPromptDefinition(context.Background(), "duplicate-fs-prompt", "")
+func TestPromptRepositoryReturnsDefinitionReadFailures(t *testing.T) {
-	if !errors.Is(err, ErrInvalidPromptDefinition) {
+	readErr := errors.New("definition read failed")
-		t.Fatalf("expected ErrInvalidPromptDefinition, got %v", err)
+	fsys := &definitionReadFailureFS{
+		FS: fstest.MapFS{
+			"prompts/target.yaml": &fstest.MapFile{Data: []byte("unread")},
+		},
+		target: "prompts/target.yaml",
+		err:    readErr,
 	}
-	if !strings.Contains(err.Error(), "one.yaml") || !strings.Contains(err.Error(), "nested/two.yaml") {
+	repo := NewFSRepository(fsys, "prompts")
-		t.Fatalf("expected duplicate paths in error, got %v", err)
+
+	definition, err := repo.GetPromptDefinition(context.Background(), "target", "1")
+	if definition != nil || !errors.Is(err, readErr) {
+		t.Fatalf("GetPromptDefinition() = (%#v, %v), want nil and definition read error", definition, err)
+	}
+	if errors.Is(err, ErrPromptDefinitionNotFound) {
+		t.Fatalf("definition read error was classified as absence: %v", err)
+	}
+	if !strings.Contains(err.Error(), "target.yaml") {
+		t.Fatalf("definition read error lacks source context: %v", err)
 	}
 }
 
-func TestFSRepositoryRejectsUnknownYAMLFields(t *testing.T) {
+type definitionReadFailureFS struct {
-	repo := NewFSRepository(fstest.MapFS{
+	fs.FS
-		"not_named_like_id.yaml": &fstest.MapFile{Data: []byte(`
+	target string
-id: strict-fs-prompt
+	err    error
-version: "1.0.0"
+}
-unknown: true
+
+func (f *definitionReadFailureFS) Open(name string) (fs.File, error) {
+	if name == f.target {
+		return nil, f.err
+	}
+	return f.FS.Open(name)
+}
+
+func (f *recordingFS) Open(name string) (fs.File, error) {
+	f.mu.Lock()
+	f.opened = append(f.opened, name)
+	f.mu.Unlock()
+	return f.FS.Open(name)
+}
+
+func (f *recordingFS) wasOpened(name string) bool {
+	return f.openCount(name) > 0
+}
+
+func (f *recordingFS) openCount(name string) int {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	count := 0
+	for _, opened := range f.opened {
+		if opened == name {
+			count++
+		}
+	}
+	return count
+}
+
+func TestPromptRepositorySelectionUsesYAMLMetadata(t *testing.T) {
+	const validDefinition = `
+id: selected-prompt
+version: "1"
 messages:
   - role: user
-    content: Invalid.
+    content: selected
 output:
   format: text
   validation_mode: none
-  repair_attempts: 0
+`
-`)},
+	tests := []struct {
-	}, ".")
+		name        string
+		files       map[string]string
+		wantErr     error
+		diagnostics []string
+		wantContent string
+	}{
+		{
+			name: "same-stem strict error with different YAML ID is unrelated",
+			files: map[string]string{
+				"selected-prompt.yaml": `
+id: another-prompt
+version: "1"
+unknown: true
+`,
+				"valid.yaml": validDefinition,
+			},
+		},
+		{
+			name: "unidentifiable same-stem YAML is unrelated",
+			files: map[string]string{
+				"selected-prompt.yaml": "id: [",
+				"valid.yaml":           validDefinition,
+			},
+		},
+		{
+			name: "same ID invalid different version is unrelated",
+			files: map[string]string{
+				"invalid-version.yaml": `
+id: selected-prompt
+version: "2"
+output:
+  format: text
+  validation_mode: none
+`,
+				"valid.yaml": validDefinition,
+			},
+		},
+		{
+			name: "selected content is resolved relative to its definition",
+			files: map[string]string{
+				"nested/selected.yaml": `
+id: selected-prompt
+version: "1"
+messages:
+  - role: user
+    content_file: ./content/selected.tmpl
+output:
+  format: text
+  validation_mode: none
+`,
+				"nested/content/selected.tmpl": "selected from file",
+			},
+			wantContent: "selected from file",
+		},
+		{
+			name: "duplicate selected definitions are ambiguous",
+			files: map[string]string{
+				"selected-a.yaml": validDefinition,
+				"nested/selected-b.yaml": `
+id: selected-prompt
+version: "1"
+messages:
+  - role: user
+    content: duplicate
+output:
+  format: text
+  validation_mode: none
+`,
+			},
+			wantErr:     ErrInvalidPromptDefinition,
+			diagnostics: []string{"duplicate prompt definition id", "selected-a.yaml", "nested/selected-b.yaml"},
+		},
+		{
+			name: "selected strict error is authoritative",
+			files: map[string]string{
+				"selected-strict.yaml": `
+id: selected-prompt
+version: "1"
+unknown: true
+`,
+			},
+			wantErr:     ErrInvalidYAML,
+			diagnostics: []string{"selected-strict.yaml"},
+		},
+		{
+			name: "selected semantic error is authoritative",
+			files: map[string]string{
+				"selected-invalid.yaml": `
+id: selected-prompt
+version: "1"
+output:
+  format: text
+  validation_mode: none
+`,
+			},
+			wantErr:     ErrInvalidPromptDefinition,
+			diagnostics: []string{"selected-invalid.yaml", "at least one message"},
+		},
+		{
+			name: "selected content error includes definition context",
+			files: map[string]string{
+				"selected-missing-content.yaml": `
+id: selected-prompt
+version: "1"
+messages:
+  - role: user
+    content_file: missing.tmpl
+output:
+  format: text
+  validation_mode: none
+`,
+			},
+			wantErr:     ErrInvalidPromptDefinition,
+			diagnostics: []string{"selected-missing-content.yaml", "failed to read content_file", "missing.tmpl"},
+		},
+	}
 
-	_, err := repo.GetPromptDefinition(context.Background(), "strict-fs-prompt", "")
+	for _, source := range promptRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, tc.files)
+				got, err := repo.GetPromptDefinition(context.Background(), "selected-prompt", "1")
+				if tc.wantErr != nil {
+					if !errors.Is(err, tc.wantErr) {
+						t.Fatalf("expected %v, got %v", tc.wantErr, err)
+					}
+					for _, diagnostic := range tc.diagnostics {
+						if !strings.Contains(err.Error(), diagnostic) {
+							t.Fatalf("expected error to contain %q, got %v", diagnostic, err)
+						}
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load selected prompt: %v", err)
+				}
+				wantContent := tc.wantContent
+				if wantContent == "" {
+					wantContent = "selected"
+				}
+				if got.ID != "selected-prompt" || got.Version != "1" || got.Templates[0].Content != wantContent {
+					t.Fatalf("unexpected selected definition: %+v", got)
+				}
+			})
+		}
+	}
+}
+
+func TestPromptRepositoryHonorsCancellation(t *testing.T) {
+	for _, source := range promptRepositorySources() {
+		t.Run(source.name, func(t *testing.T) {
+			repo := source.newRepository(t, map[string]string{
+				"definition.yaml": `
+id: cancelled-prompt
+version: "1"
+messages:
+  - role: user
+    content: selected
+output:
+  format: text
+  validation_mode: none
+`,
+			})
+			ctx, cancel := context.WithCancel(context.Background())
+			cancel()
+
+			_, err := repo.GetPromptDefinition(ctx, "cancelled-prompt", "1")
+			if !errors.Is(err, context.Canceled) {
+				t.Fatalf("expected context cancellation, got %v", err)
+			}
+		})
+	}
+}
+
+func TestPromptRepositoryRequiresOneYAMLDocument(t *testing.T) {
+	const definition = `
+id: one-document
+version: "1"
+messages:
+  - role: user
+    content: selected
+output:
+  format: text
+  validation_mode: none
+`
+	tests := []struct {
+		name    string
+		suffix  string
+		wantErr bool
+	}{
+		{name: "comments and trailing whitespace", suffix: "\n# trailing comment\n\n"},
+		{name: "second populated document", suffix: "\n---\nid: another\n", wantErr: true},
+		{name: "second empty document", suffix: "\n---\n", wantErr: true},
+		{name: "malformed trailing YAML", suffix: "\n---\n[", wantErr: true},
+	}
+
+	for _, source := range promptRepositorySources() {
+		for _, tc := range tests {
+			t.Run(source.name+"/"+tc.name, func(t *testing.T) {
+				repo := source.newRepository(t, map[string]string{"definition.yaml": definition + tc.suffix})
+				_, err := repo.GetPromptDefinition(context.Background(), "one-document", "1")
+				if tc.wantErr {
 					if !errors.Is(err, ErrInvalidYAML) {
 						t.Fatalf("expected ErrInvalidYAML, got %v", err)
 					}
+					if !strings.Contains(err.Error(), "definition.yaml") {
+						t.Fatalf("expected source path in error, got %v", err)
+					}
+					return
+				}
+				if err != nil {
+					t.Fatalf("load one document: %v", err)
+				}
+			})
+		}
+	}
+}
+
+func TestPromptRepositoryReadsOnlySelectedContent(t *testing.T) {
+	fsys := &recordingFS{FS: fstest.MapFS{
+		"target.yaml": &fstest.MapFile{Data: []byte(`
+id: target
+version: "1"
+messages:
+  - role: user
+    content_file: target.tmpl
+output:
+  format: text
+  validation_mode: none
+`)},
+		"target.tmpl": &fstest.MapFile{Data: []byte("selected")},
+		"unrelated.yaml": &fstest.MapFile{Data: []byte(`
+id: unrelated
+version: "1"
+messages:
+  - role: user
+    content_file: unrelated.tmpl
+output:
+  format: text
+  validation_mode: none
+`)},
+		"unrelated.tmpl": &fstest.MapFile{Data: []byte("unrelated")},
+		"other-version.yaml": &fstest.MapFile{Data: []byte(`
+id: target
+version: "2"
+messages:
+  - role: user
+    content_file: other-version.tmpl
+output:
+  format: text
+  validation_mode: none
+`)},
+		"other-version.tmpl": &fstest.MapFile{Data: []byte("other version")},
+	}}
+	repo := NewFSRepository(fsys, ".")
+
+	for lookup := 1; lookup <= 2; lookup++ {
+		got, err := repo.GetPromptDefinition(context.Background(), "target", "1")
+		if err != nil {
+			t.Fatalf("lookup %d: %v", lookup, err)
+		}
+		if got.Templates[0].Content != "selected" {
+			t.Fatalf("lookup %d content = %q", lookup, got.Templates[0].Content)
+		}
+		if count := fsys.openCount("target.tmpl"); count != lookup {
+			t.Fatalf("selected content opens after lookup %d = %d, want %d", lookup, count, lookup)
+		}
+		for _, name := range []string{"unrelated.tmpl", "other-version.tmpl"} {
+			if count := fsys.openCount(name); count != 0 {
+				t.Fatalf("unselected content %q opened %d times", name, count)
+			}
+		}
+		for _, name := range []string{"target.yaml", "unrelated.yaml", "other-version.yaml"} {
+			if count := fsys.openCount(name); count != lookup {
+				t.Fatalf("metadata %q opens after lookup %d = %d, want %d", name, lookup, count, lookup)
+			}
+		}
+	}
+}
+
+func TestPromptDefinitionNormalizationRules(t *testing.T) {
+	tests := []struct {
+		name           string
+		definition     string
+		wantErr        bool
+		wantDiagnostic string
+		wantSchemaPath string
+	}{
+		{
+			name: "missing version",
+			definition: `
+id: normalization-rule
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: none
+`,
+			wantErr:        true,
+			wantDiagnostic: "version",
+		},
+		{
+			name: "blank input name",
+			definition: `
+id: normalization-rule
+version: "1"
+inputs:
+  - name: "  "
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: none
+`,
+			wantErr:        true,
+			wantDiagnostic: "input 0",
+		},
+		{
+			name: "blank message role",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: "  "
+    content: test
+output:
+  format: text
+  validation_mode: none
+`,
+			wantErr:        true,
+			wantDiagnostic: "role",
+		},
+		{
+			name: "invalid output format",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: user
+    content: test
+output:
+  format: binary
+  validation_mode: none
+`,
+			wantErr:        true,
+			wantDiagnostic: "format",
+		},
+		{
+			name: "negative repair attempts",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: none
+  repair_attempts: -1
+`,
+			wantErr:        true,
+			wantDiagnostic: "repair_attempts",
+		},
+		{
+			name: "repair attempts above maximum",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: basic
+  repair_attempts: 4
+`,
+			wantErr:        true,
+			wantDiagnostic: "repair_attempts",
+		},
+		{
+			name: "none validation with repair attempts",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: none
+  repair_attempts: 1
+`,
+			wantErr:        true,
+			wantDiagnostic: "repair_attempts",
+		},
+		{
+			name: "explicit blank default profile",
+			definition: `
+id: normalization-rule
+version: "1"
+default_profile: "  "
+messages:
+  - role: user
+    content: test
+output:
+  format: text
+  validation_mode: none
+`,
+			wantErr:        true,
+			wantDiagnostic: "default_profile",
+		},
+		{
+			name: "schema path normalization",
+			definition: `
+id: normalization-rule
+version: "1"
+messages:
+  - role: user
+    content: test
+output:
+  format: json
+  validation_mode: json_schema
+  schema_path: ' schema.json '
+`,
+			wantSchemaPath: "schema.json",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			repo := NewFSRepository(fstest.MapFS{
+				"definition.yaml": &fstest.MapFile{Data: []byte(tt.definition)},
+			}, ".")
+
+			got, err := repo.GetPromptDefinition(context.Background(), "normalization-rule", "")
+			if tt.wantErr {
+				if !errors.Is(err, ErrInvalidPromptDefinition) {
+					t.Fatalf("expected ErrInvalidPromptDefinition, got %v", err)
+				}
+				if !strings.Contains(err.Error(), tt.wantDiagnostic) {
+					t.Fatalf("expected error containing %q, got %v", tt.wantDiagnostic, err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("load prompt definition: %v", err)
+			}
+			if got.Validation.SchemaPath != tt.wantSchemaPath {
+				t.Fatalf("schema path = %q, want %q", got.Validation.SchemaPath, tt.wantSchemaPath)
+			}
+		})
+	}
+}
+
+type promptRepositorySource struct {
+	name          string
+	newRepository func(t *testing.T, files map[string]string) Repository
+}
+
+func promptRepositorySources() []promptRepositorySource {
+	return []promptRepositorySource{
+		{
+			name: "operating system",
+			newRepository: func(t *testing.T, files map[string]string) Repository {
+				t.Helper()
+				root := t.TempDir()
+				for name, content := range files {
+					filePath := filepath.Join(root, filepath.FromSlash(name))
+					if err := os.MkdirAll(filepath.Dir(filePath), 0o755); err != nil {
+						t.Fatalf("create prompt directory: %v", err)
+					}
+					writePromptTestFile(t, filePath, content)
+				}
+				return NewFilesystemRepository(root)
+			},
+		},
+		{
+			name: "filesystem",
+			newRepository: func(t *testing.T, files map[string]string) Repository {
+				t.Helper()
+				fsys := make(fstest.MapFS, len(files))
+				for name, content := range files {
+					fsys[name] = &fstest.MapFile{Data: []byte(strings.TrimLeft(content, "\n"))}
+				}
+				return NewFSRepository(fsys, ".")
+			},
+		},
+	}
+}
+
+func BenchmarkPromptRepositoryLookup(b *testing.B) {
+	for _, size := range []int{10, 1000} {
+		b.Run(fmt.Sprintf("catalog-%d", size), func(b *testing.B) {
+			files := fstest.MapFS{
+				"target.yaml": &fstest.MapFile{Data: []byte(`
+id: target
+version: "1"
+messages:
+  - role: user
+    content_file: target.tmpl
+output:
+  format: text
+  validation_mode: none
+`)},
+				"target.tmpl": &fstest.MapFile{Data: []byte("selected")},
+			}
+			metadataNames := []string{"target.yaml"}
+			contentNames := make([]string, 0, size-1)
+			for i := 1; i < size; i++ {
+				definitionName := fmt.Sprintf("prompt-%04d.yaml", i)
+				contentName := fmt.Sprintf("prompt-%04d.tmpl", i)
+				files[definitionName] = &fstest.MapFile{Data: []byte(fmt.Sprintf(`
+id: prompt-%04d
+version: "1"
+messages:
+  - role: user
+    content_file: %s
+output:
+  format: text
+  validation_mode: none
+`, i, contentName))}
+				files[contentName] = &fstest.MapFile{Data: []byte("unrelated")}
+				metadataNames = append(metadataNames, definitionName)
+				contentNames = append(contentNames, contentName)
+			}
+
+			fsys := &recordingFS{FS: files}
+			repo := NewFSRepository(fsys, ".")
+			b.ReportAllocs()
+			b.ResetTimer()
+			for i := 0; i < b.N; i++ {
+				if _, err := repo.GetPromptDefinition(context.Background(), "target", "1"); err != nil {
+					b.Fatal(err)
+				}
+			}
+			b.StopTimer()
+
+			if got := fsys.openCount("target.tmpl"); got != b.N {
+				b.Fatalf("selected content opens = %d, want %d", got, b.N)
+			}
+			for _, name := range contentNames {
+				if got := fsys.openCount(name); got != 0 {
+					b.Fatalf("unrelated content %q opened %d times", name, got)
+				}
+			}
+			for _, name := range metadataNames {
+				if got := fsys.openCount(name); got != b.N {
+					b.Fatalf("metadata %q opens = %d, want %d", name, got, b.N)
+				}
+			}
+		})
+	}
 }
 
 func assertCacheControl(t *testing.T, got *domain.CacheControl, wantType domain.CacheControlType, wantTTL string) {

internal/promptdef/source.go

@@ -0,0 +1,63 @@
+package promptdef
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"os"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/filecatalog"
+)
+
+type promptDefinitionSource interface {
+	contentSourceRoot
+	findYAMLFiles(context.Context) ([]string, error)
+	readDefinition(string) ([]byte, error)
+	displayPath(string) string
+}
+
+type osPromptSource struct {
+	root        string
+	contentRoot osContentSourceRoot
+}
+
+func (s osPromptSource) findYAMLFiles(ctx context.Context) ([]string, error) {
+	return filecatalog.FindYAMLFiles(ctx, s.root)
+}
+
+func (s osPromptSource) readDefinition(name string) ([]byte, error) {
+	return os.ReadFile(name)
+}
+
+func (s osPromptSource) displayPath(name string) string {
+	return filecatalog.RelativePath(s.root, name)
+}
+
+func (s osPromptSource) readContentFile(sourcePath string, contentFile string) (string, string, error) {
+	return s.contentRoot.readContentFile(sourcePath, contentFile)
+}
+
+type fsPromptSource struct {
+	fsys        fs.FS
+	root        string
+	contentRoot contentSourceRoot
+}
+
+func (s fsPromptSource) findYAMLFiles(ctx context.Context) ([]string, error) {
+	if s.fsys == nil {
+		return nil, errors.New("filesystem is nil")
+	}
+	return filecatalog.FindFSYAMLFiles(ctx, s.fsys, s.root)
+}
+
+func (s fsPromptSource) readDefinition(name string) ([]byte, error) {
+	return fs.ReadFile(s.fsys, name)
+}
+
+func (s fsPromptSource) displayPath(name string) string {
+	return filecatalog.DisplayPath(s.root, name)
+}
+
+func (s fsPromptSource) readContentFile(sourcePath string, contentFile string) (string, string, error) {
+	return s.contentRoot.readContentFile(sourcePath, contentFile)
+}

internal/usecase/execution_settings_test.go

@@ -0,0 +1,83 @@
+package usecase
+
+import (
+	"context"
+	"errors"
+	"math"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestRunnerPrepareExecutionRejectsInvalidExecutionSettings(t *testing.T) {
+	runner := NewRunner(
+		&fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)},
+		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+		nil,
+		defaultArtifactReader(),
+		defaultRenderer(),
+		&fakeLLM{forbid: true},
+		nil,
+		nil,
+	)
+
+	_, err := runner.PrepareExecution(context.Background(), domain.RunRequest{
+		PromptID:  "p",
+		ProfileID: "exec",
+		Inputs:    singleInputRef(),
+		Execution: &domain.ExecutionTargetOverride{TopP: float64Ptr(math.Inf(-1))},
+	})
+	if !errors.Is(err, ErrInvalidRequest) {
+		t.Fatalf("expected ErrInvalidRequest, got %v", err)
+	}
+}
+
+func TestRunnerPrepareExecutionValidatesAndNormalizesRequestEndpoints(t *testing.T) {
+	newRunner := func() *Runner {
+		return NewRunner(
+			&fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)},
+			&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+			nil,
+			defaultArtifactReader(),
+			defaultRenderer(),
+			&fakeLLM{forbid: true},
+			nil,
+			nil,
+		)
+	}
+
+	invalidEndpoints := []string{
+		"/v1",
+		"https:///v1",
+		"ftp://provider.example/v1",
+		"https://user@provider.example/v1",
+		"https://provider.example/v1?mode=chat",
+		"https://provider.example/v1#chat",
+	}
+	for _, endpoint := range invalidEndpoints {
+		t.Run(endpoint, func(t *testing.T) {
+			_, err := newRunner().PrepareExecution(context.Background(), domain.RunRequest{
+				PromptID:  "p",
+				ProfileID: "exec",
+				Inputs:    singleInputRef(),
+				Execution: &domain.ExecutionTargetOverride{Endpoint: endpoint},
+			})
+			if !errors.Is(err, ErrInvalidRequest) {
+				t.Fatalf("expected ErrInvalidRequest, got %v", err)
+			}
+		})
+	}
+
+	prepared, err := newRunner().PrepareExecution(context.Background(), domain.RunRequest{
+		PromptID:  "p",
+		ProfileID: "exec",
+		Inputs:    singleInputRef(),
+		Execution: &domain.ExecutionTargetOverride{Endpoint: "  https://provider.example/nested/v1  "},
+	})
+	if err != nil {
+		t.Fatalf("prepare normalized endpoint: %v", err)
+	}
+	if got := prepared.Details().EffectiveModelParams.Endpoint; got != "https://provider.example/nested/v1" {
+		t.Fatalf("effective endpoint = %q", got)
+	}
+}

internal/usecase/generation_request.go

@@ -0,0 +1,19 @@
+package usecase
+
+import "gitea.maximumdirect.net/eric/promptkit/internal/domain"
+
+func newGenerationRequest(
+	prompt domain.RenderedPrompt,
+	sessionID string,
+	target domain.ExecutionTarget,
+	targetPresence domain.ExecutionTargetPresence,
+	structuredOutput *domain.StructuredOutputSpec,
+) domain.GenerateRequest {
+	prompt.SessionID = sessionID
+	return domain.GenerateRequest{
+		Prompt:           prompt,
+		Target:           target,
+		TargetPresence:   targetPresence,
+		StructuredOutput: structuredOutput,
+	}
+}

internal/usecase/output_contract_test.go

@@ -0,0 +1,187 @@
+package usecase
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+type outputContractTestCollaborators struct {
+	artifacts *fakeArtifactReader
+	renderer  *fakeRenderer
+	llm       *fakeLLM
+	validator *recordingValidationPreparer
+	admitter  *fakeRunAdmitter
+}
+
+func newOutputContractTestRunner() (*Runner, outputContractTestCollaborators) {
+	collaborators := outputContractTestCollaborators{
+		artifacts: defaultArtifactReader(),
+		renderer:  defaultRenderer(),
+		llm:       &fakeLLM{forbid: true},
+		validator: &recordingValidationPreparer{plan: &recordingPreparedValidation{}},
+		admitter:  &fakeRunAdmitter{},
+	}
+	return NewRunner(
+		&fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)},
+		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+		nil,
+		collaborators.artifacts,
+		collaborators.renderer,
+		collaborators.llm,
+		collaborators.validator,
+		collaborators.admitter,
+	), collaborators
+}
+
+func TestRunnerPreparationNormalizesOutputContractConsistently(t *testing.T) {
+	tests := []struct {
+		name     string
+		override domain.OutputContract
+		want     domain.OutputContract
+	}{
+		{
+			name:     "empty replacement format defaults to text",
+			override: domain.OutputContract{ValidationMode: domain.ValidationNone},
+			want:     domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationNone},
+		},
+		{
+			name:     "markdown basic replacement",
+			override: domain.OutputContract{Format: domain.FormatMarkdown, ValidationMode: domain.ValidationBasic},
+			want:     domain.OutputContract{Format: domain.FormatMarkdown, ValidationMode: domain.ValidationBasic},
+		},
+		{
+			name: "json replacement preserves non-schema fields",
+			override: domain.OutputContract{
+				Format:         domain.FormatJSON,
+				ValidationMode: domain.ValidationJSON,
+				SchemaPath:     "ignored.json",
+				RepairAttempts: 2,
+			},
+			want: domain.OutputContract{
+				Format:         domain.FormatJSON,
+				ValidationMode: domain.ValidationJSON,
+				SchemaPath:     "ignored.json",
+				RepairAttempts: 2,
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			runner, _ := newOutputContractTestRunner()
+			req := domain.RunRequest{
+				PromptID:   "p",
+				ProfileID:  "exec",
+				Inputs:     singleInputRef(),
+				Validation: &tt.override,
+			}
+
+			prepared, err := runner.Prepare(context.Background(), req)
+			if err != nil {
+				t.Fatalf("prepare: %v", err)
+			}
+			preparedExecution, err := runner.PrepareExecution(context.Background(), req)
+			if err != nil {
+				t.Fatalf("prepare execution: %v", err)
+			}
+			details := preparedExecution.Details()
+			if details == nil {
+				t.Fatal("prepared execution returned nil details")
+			}
+			if prepared.OutputContract != tt.want || details.OutputContract != tt.want {
+				t.Fatalf("output contracts = (%+v, %+v), want %+v", prepared.OutputContract, details.OutputContract, tt.want)
+			}
+		})
+	}
+}
+
+func TestRunnerPreparationRejectsInvalidOutputContractsBeforeCompletion(t *testing.T) {
+	tests := []struct {
+		name     string
+		override domain.OutputContract
+	}{
+		{name: "unsupported format", override: domain.OutputContract{Format: "binary", ValidationMode: domain.ValidationNone}},
+		{name: "empty validation mode", override: domain.OutputContract{Format: domain.FormatText}},
+		{name: "negative repair attempts", override: domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationNone, RepairAttempts: -1}},
+		{name: "repair attempts above maximum", override: domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationBasic, RepairAttempts: 4}},
+		{name: "none validation with repair attempts", override: domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationNone, RepairAttempts: 1}},
+		{name: "json schema without path", override: domain.OutputContract{Format: domain.FormatJSON, ValidationMode: domain.ValidationJSONSchema}},
+	}
+
+	for _, tt := range tests {
+		for _, operation := range []string{"Prepare", "PrepareExecution"} {
+			t.Run(tt.name+"/"+operation, func(t *testing.T) {
+				runner, collaborators := newOutputContractTestRunner()
+				req := domain.RunRequest{
+					PromptID:   "p",
+					ProfileID:  "exec",
+					Inputs:     singleInputRef(),
+					Validation: &tt.override,
+				}
+
+				var err error
+				switch operation {
+				case "Prepare":
+					var prepared *domain.PreparedRun
+					prepared, err = runner.Prepare(context.Background(), req)
+					if prepared != nil {
+						t.Fatalf("expected no partial prepared run, got %+v", prepared)
+					}
+				case "PrepareExecution":
+					var prepared *PreparedExecution
+					prepared, err = runner.PrepareExecution(context.Background(), req)
+					if prepared != nil {
+						t.Fatalf("expected no partial prepared execution, got %+v", prepared)
+					}
+				default:
+					t.Fatalf("unknown operation %q", operation)
+				}
+
+				if !errors.Is(err, ErrInvalidRequest) {
+					t.Fatalf("expected ErrInvalidRequest, got %v", err)
+				}
+				assertOutputContractCompletionSkipped(t, collaborators)
+			})
+		}
+	}
+}
+
+func TestRunnerRunRejectsInvalidOutputContractBeforeAdmission(t *testing.T) {
+	runner, collaborators := newOutputContractTestRunner()
+	result, err := runner.Run(context.Background(), domain.RunRequest{
+		PromptID:  "p",
+		ProfileID: "exec",
+		Inputs:    singleInputRef(),
+		Validation: &domain.OutputContract{
+			Format:         domain.OutputFormat("binary"),
+			ValidationMode: domain.ValidationNone,
+		},
+	})
+	if result != nil {
+		t.Fatalf("expected no partial result, got %+v", result)
+	}
+	if !errors.Is(err, ErrInvalidRequest) {
+		t.Fatalf("expected ErrInvalidRequest, got %v", err)
+	}
+	assertOutputContractCompletionSkipped(t, collaborators)
+}
+
+func assertOutputContractCompletionSkipped(t *testing.T, collaborators outputContractTestCollaborators) {
+	t.Helper()
+	if collaborators.artifacts.calls != 0 || collaborators.renderer.calls != 0 ||
+		collaborators.validator.prepareCalls != 0 || collaborators.validator.directValidateCalls != 0 ||
+		len(collaborators.admitter.backendIDs) != 0 || collaborators.llm.calls != 0 {
+		t.Fatalf(
+			"invalid output contract reached downstream work: artifacts=%d renderer=%d prepare_validation=%d validation=%d admissions=%d generation=%d",
+			collaborators.artifacts.calls,
+			collaborators.renderer.calls,
+			collaborators.validator.prepareCalls,
+			collaborators.validator.directValidateCalls,
+			len(collaborators.admitter.backendIDs),
+			collaborators.llm.calls,
+		)
+	}
+}

internal/usecase/prepared_execution.go

@@ -42,25 +42,12 @@ func (r *Runner) PrepareExecution(ctx context.Context, req domain.RunRequest) (*
 		return nil, err
 	}
 
-	validationPlan, err := r.prepareValidation(ctx, state.effectiveContract)
+	operation, err := r.completePreparation(ctx, req, state)
 	if err != nil {
 		return nil, err
 	}
 
-	structuredOutput, err := r.structuredOutputFromValidationPlan(
+	executionSnapshot, err := clonePreparedRun(operation.run)
-		state.definition,
-		state.effectiveContract,
-		validationPlan,
-	)
-	if err != nil {
-		return nil, err
-	}
-	prepared, err := r.completePreparationWithStructuredOutput(ctx, req, state, structuredOutput)
-	if err != nil {
-		return nil, err
-	}
-
-	executionSnapshot, err := clonePreparedRun(prepared)
 	if err != nil {
 		return nil, fmt.Errorf("%w: failed to copy prepared execution: %v", ErrInvalidRequest, err)
 	}
@@ -76,52 +63,12 @@ func (r *Runner) PrepareExecution(ctx context.Context, req domain.RunRequest) (*
 		details: details,
 		payload: &preparedExecutionPayload{
 			prepared:   executionSnapshot,
-			validation: validationPlan,
+			validation: operation.validation,
 			directKey:  state.effectiveModel.APIKey,
 		},
 	}, nil
 }
 
-func (r *Runner) prepareValidation(
-	ctx context.Context,
-	contract domain.OutputContract,
-) (validate.PreparedValidation, error) {
-	if r.validator == nil {
-		return noOpPreparedValidation{contract: contract}, nil
-	}
-	preparer, ok := r.validator.(validate.ValidationPreparer)
-	if !ok {
-		return nil, fmt.Errorf("%w: validator does not support prepared validation", ErrValidation)
-	}
-	plan, err := preparer.PrepareValidation(ctx, contract)
-	if err != nil {
-		return nil, fmt.Errorf("%w: %w", ErrValidation, err)
-	}
-	if plan == nil {
-		return nil, fmt.Errorf("%w: validator returned nil prepared validation", ErrValidation)
-	}
-	return plan, nil
-}
-
-func (r *Runner) structuredOutputFromValidationPlan(
-	def *domain.PromptDefinition,
-	contract domain.OutputContract,
-	plan validate.PreparedValidation,
-) (*domain.StructuredOutputSpec, error) {
-	if contract.ValidationMode != domain.ValidationJSONSchema {
-		return nil, nil
-	}
-
-	schemaDocument := plan.SchemaDocument()
-	if schemaDocument == nil {
-		if r.validator == nil {
-			return nil, nil
-		}
-		return nil, fmt.Errorf("%w: prepared json_schema validation has no schema document", ErrValidation)
-	}
-	return structuredOutputSpec(def, schemaDocument), nil
-}
-
 // Details returns a fresh credential-redacted copy of the prepared run.
 func (p *PreparedExecution) Details() *domain.PreparedRun {
 	if p == nil {

internal/usecase/prepared_execution_test.go

@@ -3,11 +3,12 @@ package usecase
 import (
 	"context"
 	"errors"
-	"os"
+	"fmt"
 	"reflect"
 	"testing"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 	"gitea.maximumdirect.net/eric/promptkit/internal/validate"
 )
 
@@ -52,6 +53,12 @@ type recordingValidationPreparer struct {
 	directValidateCalls int
 }
 
+type validationOnly struct{}
+
+func (validationOnly) Validate(context.Context, *domain.Artifact, domain.OutputContract) (domain.ValidationResult, error) {
+	return domain.ValidationResult{}, nil
+}
+
 func (v *recordingValidationPreparer) Validate(
 	context.Context,
 	*domain.Artifact,
@@ -164,15 +171,107 @@ func TestRunnerPrepareExecutionCompletesWithoutAdmissionOrGeneration(t *testing.
 	}
 }
 
-func TestRunnerRunPreparedRechecksEnvironmentCredentialBeforeAdmission(t *testing.T) {
+func TestRunnerPreparationRejectsExcessivelyDeepPreparedSchema(t *testing.T) {
+	operations := []struct {
+		name string
+		run  func(*Runner, domain.RunRequest) error
+	}{
+		{
+			name: "Prepare",
+			run: func(runner *Runner, request domain.RunRequest) error {
+				_, err := runner.Prepare(context.Background(), request)
+				return err
+			},
+		},
+		{
+			name: "Run",
+			run: func(runner *Runner, request domain.RunRequest) error {
+				_, err := runner.Run(context.Background(), request)
+				return err
+			},
+		},
+		{
+			name: "PrepareExecution",
+			run: func(runner *Runner, request domain.RunRequest) error {
+				_, err := runner.PrepareExecution(context.Background(), request)
+				return err
+			},
+		},
+	}
+
+	for _, operation := range operations {
+		t.Run(operation.name, func(t *testing.T) {
+			def := promptDef(domain.FormatJSON, domain.ValidationJSONSchema, 0)
+			def.Validation.SchemaPath = "schema.json"
+			llmClient := &fakeLLM{forbid: true}
+			validator := &recordingValidationPreparer{
+				plan: &recordingPreparedValidation{schemaDocument: excessivelyDeepPreparedJSONValue()},
+			}
+			runner := NewRunner(
+				&fakePromptRepo{def: def},
+				&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+				nil,
+				defaultArtifactReader(),
+				defaultRenderer(),
+				llmClient,
+				validator,
+				nil,
+			)
+
+			err := operation.run(runner, domain.RunRequest{
+				PromptID:  "p",
+				ProfileID: "exec",
+				Inputs:    singleInputRef(),
+			})
+			if !errors.Is(err, ErrValidation) {
+				t.Fatalf("expected ErrValidation, got %v", err)
+			}
+			if llmClient.calls != 0 {
+				t.Fatalf("invalid prepared schema reached generation: %d calls", llmClient.calls)
+			}
+		})
+	}
+}
+
+func excessivelyDeepPreparedJSONValue() any {
+	const clearlyUnsafeContainerDepth = 1_000
+	var value any = true
+	for level := 0; level < clearlyUnsafeContainerDepth; level++ {
+		value = map[string]any{"child": value}
+	}
+	return value
+}
+
+func TestRunnerRunPreparedCredentialAvailabilityBeforeAdmission(t *testing.T) {
 	const environmentName = "PROMPTKIT_PREPARED_EXECUTION_TEST_KEY"
+	tests := []struct {
+		name           string
+		apiKeyRequired bool
+		profileEnv     bool
+		overrideEnv    bool
+		wantFailure    bool
+	}{
+		{name: "optional environment becomes unavailable", profileEnv: true},
+		{
+			name:           "required request environment becomes unavailable",
+			apiKeyRequired: true,
+			overrideEnv:    true,
+			wantFailure:    true,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
 			t.Setenv(environmentName, "available-during-preparation")
 
 			profile := defaultExecutionProfile()
+			profile.APIKeyRequired = tc.apiKeyRequired
+			if tc.profileEnv {
 				profile.APIKeyEnv = environmentName
+			}
 			validator := &recordingValidationPreparer{plan: &recordingPreparedValidation{}}
 			admitter := &fakeRunAdmitter{}
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: "unexpected"}}
+			llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: "ok"}}
 			runner := NewRunner(
 				&fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)},
 				&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": profile}},
@@ -183,30 +282,36 @@ func TestRunnerRunPreparedRechecksEnvironmentCredentialBeforeAdmission(t *testin
 				validator,
 				admitter,
 			)
-	prepared, err := runner.PrepareExecution(context.Background(), domain.RunRequest{
+			request := domain.RunRequest{PromptID: "p", ProfileID: "exec", Inputs: singleInputRef()}
-		PromptID:  "p",
+			if tc.overrideEnv {
-		ProfileID: "exec",
+				request.Execution = &domain.ExecutionTargetOverride{APIKeyEnv: environmentName}
-		Inputs:    singleInputRef(),
+			}
-	})
+			prepared, err := runner.PrepareExecution(context.Background(), request)
 			if err != nil {
 				t.Fatalf("prepare execution: %v", err)
 			}
-	if err := os.Unsetenv(environmentName); err != nil {
+			t.Setenv(environmentName, "")
-		t.Fatalf("unset credential environment: %v", err)
-	}
 
 			result, err := runner.RunPrepared(context.Background(), prepared)
-	if result != nil {
+			if tc.wantFailure {
-		t.Fatalf("credential failure returned partial result: %+v", result)
+				if result != nil || !errors.Is(err, ErrInvalidRequest) || !errors.Is(err, ErrAPIKeyEnvMissing) {
-	}
+					t.Fatalf("required credential result = (%+v, %v)", result, err)
-	if !errors.Is(err, ErrInvalidRequest) || !errors.Is(err, ErrAPIKeyEnvMissing) {
-		t.Fatalf("credential error identities are missing: %v", err)
 				}
 				if len(admitter.backendIDs) != 0 || llmClient.calls != 0 {
-		t.Fatalf("credential failure reached admission or generation: admission=%v generation=%d", admitter.backendIDs, llmClient.calls)
+					t.Fatalf("required credential reached admission or generation: admission=%v generation=%d", admitter.backendIDs, llmClient.calls)
+				}
+			} else {
+				if result == nil || err != nil {
+					t.Fatalf("optional credential result = (%+v, %v), want success", result, err)
+				}
+				if len(admitter.backendIDs) != 1 || llmClient.calls != 1 {
+					t.Fatalf("optional credential admission=%v generation=%d, want one each", admitter.backendIDs, llmClient.calls)
+				}
 			}
 			if _, err := runner.RunPrepared(context.Background(), prepared); !errors.Is(err, ErrInvalidRequest) {
-		t.Fatalf("credential failure did not consume execution: %v", err)
+				t.Fatalf("execution outcome did not consume handle: %v", err)
+			}
+		})
 	}
 }
 
@@ -271,18 +376,31 @@ func TestRunnerRunPreparedUsesFrozenValidationForInitialAndRepairOutputs(t *test
 		},
 	}
 	repairer := &fakeRepairer{
-		responses: []*domain.GenerateResponse{{Content: `{"repaired":true}`}},
+		responses: []*domain.GenerateResponse{{
+			Content: `{"repaired":true}`,
+			Usage: domain.TokenUsage{
+				PromptTokens: 2, CompletionTokens: 3, TotalTokens: 5,
+				CachedTokens: 7, CacheWriteTokens: 11,
+			},
+		}},
 	}
 	admitter := &fakeRunAdmitter{}
 	reader := defaultArtifactReader()
 	renderer := defaultRenderer()
+	client := &sequenceLLM{responses: []*domain.GenerateResponse{{
+		Content: `{"broken":true}`,
+		Usage: domain.TokenUsage{
+			PromptTokens: 13, CompletionTokens: 17, TotalTokens: 19,
+			CachedTokens: 23, CacheWriteTokens: 29,
+		},
+	}}}
 	runner := NewRunnerWithRepairer(
 		&fakePromptRepo{def: promptDef(domain.FormatJSON, domain.ValidationJSON, 1)},
 		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
 		nil,
 		reader,
 		renderer,
-		&fakeLLM{resp: &domain.GenerateResponse{Content: `{"broken":true}`}},
+		client,
 		validator,
 		repairer,
 		admitter,
@@ -306,9 +424,20 @@ func TestRunnerRunPreparedUsesFrozenValidationForInitialAndRepairOutputs(t *test
 	if validator.directValidateCalls != 0 || repairer.calls != 1 {
 		t.Fatalf("validation/repair calls=(direct=%d repair=%d), want (0, 1)", validator.directValidateCalls, repairer.calls)
 	}
+	if len(client.requests) != 1 || len(repairer.reqs) != 1 ||
+		!reflect.DeepEqual(repairer.reqs[0].OriginalMessages, client.requests[0].Prompt.Messages) {
+		t.Fatalf("initial and repair messages = (%#v, %#v)", client.requests, repairer.reqs)
+	}
 	if result.Validation.Status != domain.ValidationPassed || result.Validation.RepairAttempts != 1 {
 		t.Fatalf("unexpected repaired validation result: %+v", result.Validation)
 	}
+	wantUsage := domain.TokenUsage{
+		PromptTokens: 15, CompletionTokens: 20, TotalTokens: 24,
+		CachedTokens: 30, CacheWriteTokens: 40,
+	}
+	if result.Usage != wantUsage {
+		t.Fatalf("prepared cumulative usage = %+v, want %+v", result.Usage, wantUsage)
+	}
 	if admitter.releaseCalls != 1 {
 		t.Fatalf("admission releases=%d, want 1", admitter.releaseCalls)
 	}
@@ -321,6 +450,7 @@ func TestRunnerRunPreparedReleasesAdmissionAcrossExecutionErrors(t *testing.T) {
 	generationFailure := errors.New("generation failed")
 	validationFailure := errors.New("validation failed")
 	repairFailure := errors.New("repair failed")
+	repairInvalidRequest := fmt.Errorf("repair request: %w", llm.ErrInvalidRequest)
 
 	tests := []struct {
 		name          string
@@ -328,6 +458,7 @@ func TestRunnerRunPreparedReleasesAdmissionAcrossExecutionErrors(t *testing.T) {
 		validation    *recordingPreparedValidation
 		repairer      *fakeRepairer
 		wantError     error
+		wantSource    error
 	}{
 		{
 			name:          "generation failure",
@@ -353,7 +484,50 @@ func TestRunnerRunPreparedReleasesAdmissionAcrossExecutionErrors(t *testing.T) {
 				}},
 			},
 			repairer:   &fakeRepairer{err: repairFailure},
-			wantError: ErrValidation,
+			wantError:  ErrLLMGenerate,
+			wantSource: repairFailure,
+		},
+		{
+			name: "repair invalid request",
+			validation: &recordingPreparedValidation{
+				results: []domain.ValidationResult{{
+					Status:  domain.ValidationFailed,
+					Mode:    domain.ValidationJSON,
+					Errors:  []string{"invalid"},
+					IsValid: false,
+				}},
+			},
+			repairer:   &fakeRepairer{err: repairInvalidRequest},
+			wantError:  ErrInvalidRequest,
+			wantSource: repairInvalidRequest,
+		},
+		{
+			name: "repair cancellation",
+			validation: &recordingPreparedValidation{
+				results: []domain.ValidationResult{{
+					Status:  domain.ValidationFailed,
+					Mode:    domain.ValidationJSON,
+					Errors:  []string{"invalid"},
+					IsValid: false,
+				}},
+			},
+			repairer:   &fakeRepairer{err: context.Canceled},
+			wantError:  ErrLLMGenerate,
+			wantSource: context.Canceled,
+		},
+		{
+			name: "repair deadline",
+			validation: &recordingPreparedValidation{
+				results: []domain.ValidationResult{{
+					Status:  domain.ValidationFailed,
+					Mode:    domain.ValidationJSON,
+					Errors:  []string{"invalid"},
+					IsValid: false,
+				}},
+			},
+			repairer:   &fakeRepairer{err: context.DeadlineExceeded},
+			wantError:  ErrLLMGenerate,
+			wantSource: context.DeadlineExceeded,
 		},
 	}
 
@@ -392,6 +566,9 @@ func TestRunnerRunPreparedReleasesAdmissionAcrossExecutionErrors(t *testing.T) {
 			if result != nil || !errors.Is(err, test.wantError) {
 				t.Fatalf("run prepared=(%+v, %v), want %v", result, err, test.wantError)
 			}
+			if test.wantSource != nil && !errors.Is(err, test.wantSource) {
+				t.Fatalf("run prepared error = %v, want source %v", err, test.wantSource)
+			}
 			if len(admitter.backendIDs) != 1 || admitter.releaseCalls != 1 {
 				t.Fatalf(
 					"admission calls=%#v releases=%d, want one each",
@@ -492,7 +669,7 @@ func TestRunnerPrepareExecutionRequiresValidationPreparer(t *testing.T) {
 		reader,
 		defaultRenderer(),
 		&fakeLLM{forbid: true},
-		&fakeValidator{},
+		validationOnly{},
 		nil,
 	)
 

internal/usecase/profile_inspection.go

@@ -57,14 +57,19 @@ func (r *Runner) resolveProfileSelection(
 	}, nil
 }
 
-func validateResolvedExecutionTarget(target domain.ExecutionTarget) error {
+func normalizeResolvedExecutionTarget(target domain.ExecutionTarget) (domain.ExecutionTarget, error) {
-	if strings.TrimSpace(target.Endpoint) == "" {
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(target.Endpoint)
-		return errors.New("execution endpoint is required")
+	if err != nil {
+		return domain.ExecutionTarget{}, fmt.Errorf("execution endpoint: %w", err)
 	}
+	target.Endpoint = endpoint
 	if strings.TrimSpace(target.Model) == "" {
-		return errors.New("execution model is required")
+		return domain.ExecutionTarget{}, errors.New("execution model is required")
 	}
-	return nil
+	if err := domain.ValidateExecutionTargetSettings(target); err != nil {
+		return domain.ExecutionTarget{}, err
+	}
+	return target, nil
 }
 
 // InspectProfile resolves one explicit profile without prompt or execution work.
@@ -86,13 +91,11 @@ func (r *Runner) InspectProfile(
 	if err != nil {
 		return nil, err
 	}
-	target, _, err := resolveExecutionTarget(selection.backend, selection.profile, nil)
+	target, _ := resolveExecutionTarget(selection.backend, selection.profile, nil)
+	target, err = normalizeResolvedExecutionTarget(target)
 	if err != nil {
 		return nil, fmt.Errorf("%w: %w", ErrProfileLoad, err)
 	}
-	if err := validateResolvedExecutionTarget(target); err != nil {
-		return nil, fmt.Errorf("%w: %w", ErrProfileLoad, err)
-	}
 	target.APIKey = ""
 
 	return &domain.ProfileInspection{

internal/usecase/repairer.go

@@ -2,23 +2,34 @@ package usecase
 
 import (
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"strings"
+	"unicode/utf8"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 )
 
+const (
+	maxRepairDiagnosticBytes = 64 * 1024
+	omittedRepairDiagnostics = "additional validation diagnostics were omitted"
+)
+
+// OutputRepairer generates a corrected candidate after validation fails.
 type OutputRepairer interface {
 	Repair(ctx context.Context, req RepairRequest) (*domain.GenerateResponse, error)
 }
 
+// RepairRequest contains the immutable execution state needed for one correction.
 type RepairRequest struct {
+	OriginalMessages []domain.RenderedMessage
 	PreviousOutput   string
 	ValidationErrors []string
 	SessionID        string
 	Target           domain.ExecutionTarget
+	TargetPresence   domain.ExecutionTargetPresence
 	StructuredOutput *domain.StructuredOutputSpec
 	Attempt          int
 	MaxAttempts      int
@@ -29,6 +40,7 @@ type defaultOutputRepairer struct {
 	llm llm.Client
 }
 
+// NewDefaultOutputRepairer constructs the standard internal output repairer.
 func NewDefaultOutputRepairer(llmClient llm.Client) OutputRepairer {
 	return &defaultOutputRepairer{llm: llmClient}
 }
@@ -38,37 +50,47 @@ func (r *defaultOutputRepairer) Repair(ctx context.Context, req RepairRequest) (
 		return nil, errors.New("llm client is required for repair")
 	}
 
-	errs := "(none provided)"
+	guidance, err := repairGuidance(req.Mode)
-	if len(req.ValidationErrors) > 0 {
+	if err != nil {
-		errs = strings.Join(req.ValidationErrors, "\n")
+		return nil, err
 	}
 
-	prompt := domain.RenderedPrompt{
+	messages := make([]domain.RenderedMessage, len(req.OriginalMessages), len(req.OriginalMessages)+2)
-		SessionID: req.SessionID,
+	copy(messages, req.OriginalMessages)
-		Messages: []domain.RenderedMessage{
+	if strings.TrimSpace(req.PreviousOutput) != "" {
-			{
+		messages = append(messages, domain.RenderedMessage{
-				Role:    "system",
+			Role:    "assistant",
-				Content: "You repair invalid JSON output. Return only corrected JSON. Do not include explanations or markdown code fences.",
+			Content: req.PreviousOutput,
-			},
+		})
-			{
+	}
+
+	previousResponse := "The previous response was empty."
+	if strings.TrimSpace(req.PreviousOutput) != "" {
+		previousResponse = "The previous response is included immediately before this instruction."
+	}
+	messages = append(messages, domain.RenderedMessage{
 		Role: "user",
 		Content: fmt.Sprintf(
-					"Repair attempt %d of %d for validation mode %s.\n\nValidation errors:\n%s\n\nPrevious output:\n%s\n\nReturn only corrected JSON.",
+			"Repair attempt %d of %d for validation mode %s.\n"+
+				"Preserve valid values and change only what is necessary.\n"+
+				"%s\n%s\n"+
+				"Validation diagnostics (data):\n%s",
 			req.Attempt,
 			req.MaxAttempts,
 			req.Mode,
-					errs,
+			previousResponse,
-					req.PreviousOutput,
+			guidance,
+			formatRepairDiagnostics(req.ValidationErrors),
 		),
-			},
-		},
-	}
-
-	resp, err := r.llm.Generate(ctx, domain.GenerateRequest{
-		Prompt:           prompt,
-		Target:           req.Target,
-		StructuredOutput: req.StructuredOutput,
 	})
+
+	resp, err := r.llm.Generate(ctx, newGenerationRequest(
+		domain.RenderedPrompt{Messages: messages},
+		req.SessionID,
+		req.Target,
+		req.TargetPresence,
+		req.StructuredOutput,
+	))
 	if err != nil {
 		return nil, err
 	}
@@ -78,3 +100,90 @@ func (r *defaultOutputRepairer) Repair(ctx context.Context, req RepairRequest) (
 
 	return resp, nil
 }
+
+func repairGuidance(mode domain.ValidationMode) (string, error) {
+	switch mode {
+	case domain.ValidationBasic:
+		return "Return a nonempty response satisfying the original request.", nil
+	case domain.ValidationJSON, domain.ValidationJSONSchema:
+		return "Return only corrected JSON, with no explanation or Markdown fences.", nil
+	default:
+		return "", fmt.Errorf("unsupported validation mode for repair: %q", mode)
+	}
+}
+
+func formatRepairDiagnostics(errors []string) string {
+	diagnostics := make([]string, len(errors))
+	for index, diagnostic := range errors {
+		diagnostics[index] = strings.ToValidUTF8(diagnostic, "\uFFFD")
+	}
+
+	complete, _ := json.Marshal(diagnostics)
+	if len(complete) <= maxRepairDiagnosticBytes {
+		return string(complete)
+	}
+
+	omission, _ := json.Marshal(omittedRepairDiagnostics)
+	encoded := make([]byte, 0, maxRepairDiagnosticBytes)
+	encoded = append(encoded, '[')
+	for _, diagnostic := range diagnostics {
+		entry, _ := json.Marshal(diagnostic)
+		separator := 0
+		if len(encoded) > 1 {
+			separator = 1
+		}
+		available := maxRepairDiagnosticBytes - len(encoded) - separator - 1 - len(omission) - 1
+		if len(entry) <= available {
+			if separator != 0 {
+				encoded = append(encoded, ',')
+			}
+			encoded = append(encoded, entry...)
+			continue
+		}
+
+		if available < len(`""`) {
+			break
+		}
+		if separator != 0 {
+			encoded = append(encoded, ',')
+		}
+		encoded = append(encoded, truncateDiagnosticJSONValue(diagnostic, available)...)
+		break
+	}
+	if len(encoded) > 1 {
+		encoded = append(encoded, ',')
+	}
+	encoded = append(encoded, omission...)
+	encoded = append(encoded, ']')
+	return string(encoded)
+}
+
+func truncateDiagnosticJSONValue(value string, maxBytes int) []byte {
+	if maxBytes < len(`""`) {
+		return nil
+	}
+
+	boundaries := []int{0}
+	for end := 0; end < len(value); {
+		_, size := utf8.DecodeRuneInString(value[end:])
+		end += size
+		if end+len(`""`) > maxBytes {
+			break
+		}
+		boundaries = append(boundaries, end)
+	}
+
+	low, high := 0, len(boundaries)-1
+	best := []byte(`""`)
+	for low <= high {
+		mid := low + (high-low)/2
+		candidate, _ := json.Marshal(value[:boundaries[mid]])
+		if len(candidate) <= maxBytes {
+			best = candidate
+			low = mid + 1
+			continue
+		}
+		high = mid - 1
+	}
+	return best
+}

internal/usecase/repairer_test.go

@@ -0,0 +1,349 @@
+package usecase
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+	"unicode/utf8"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+type recordingRepairClient struct {
+	requests []domain.GenerateRequest
+	response *domain.GenerateResponse
+	err      error
+}
+
+func (c *recordingRepairClient) Generate(_ context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
+	c.requests = append(c.requests, req)
+	return c.response, c.err
+}
+
+func TestDefaultOutputRepairerBuildsFullContextRequest(t *testing.T) {
+	client := &recordingRepairClient{response: &domain.GenerateResponse{Content: "corrected"}}
+	repairer := NewDefaultOutputRepairer(client)
+	original := []domain.RenderedMessage{
+		{Role: "system", Content: "Follow the task.", CacheControl: &domain.CacheControl{Type: domain.CacheControlEphemeral, TTL: "1h"}},
+		{Role: "user", Content: "Summarize the report."},
+	}
+	before := append([]domain.RenderedMessage(nil), original...)
+	previous := strings.Repeat("candidate ", 12_000)
+	target := domain.ExecutionTarget{BackendID: "backend", Endpoint: "https://provider.example/v1", Model: "model"}
+	presence := domain.ExecutionTargetPresence{Temperature: true, TopP: true}
+	structured := &domain.StructuredOutputSpec{}
+
+	response, err := repairer.Repair(context.Background(), RepairRequest{
+		OriginalMessages: original,
+		PreviousOutput:   previous,
+		ValidationErrors: []string{"invalid JSON"},
+		SessionID:        "session",
+		Target:           target,
+		TargetPresence:   presence,
+		StructuredOutput: structured,
+		Attempt:          1,
+		MaxAttempts:      3,
+		Mode:             domain.ValidationJSONSchema,
+	})
+	if err != nil {
+		t.Fatalf("repair: %v", err)
+	}
+	if response == nil || response.Content != "corrected" {
+		t.Fatalf("response = %+v", response)
+	}
+	if !reflect.DeepEqual(original, before) {
+		t.Fatalf("original messages changed: got %#v, want %#v", original, before)
+	}
+	if len(client.requests) != 1 {
+		t.Fatalf("generation requests = %d, want 1", len(client.requests))
+	}
+
+	request := client.requests[0]
+	if request.Prompt.SessionID != "session" || !reflect.DeepEqual(request.Target, target) || request.TargetPresence != presence || request.StructuredOutput != structured {
+		t.Fatalf("generation request fields = %+v", request)
+	}
+	if len(request.Prompt.Messages) != len(original)+2 {
+		t.Fatalf("message count = %d, want %d", len(request.Prompt.Messages), len(original)+2)
+	}
+	if !reflect.DeepEqual(request.Prompt.Messages[:len(original)], original) {
+		t.Fatalf("original messages = %#v, want %#v", request.Prompt.Messages[:len(original)], original)
+	}
+	assistant := request.Prompt.Messages[len(original)]
+	if assistant.Role != "assistant" || assistant.Content != previous {
+		t.Fatalf("assistant candidate = %+v", assistant)
+	}
+	correction := request.Prompt.Messages[len(original)+1]
+	if correction.Role != "user" || !strings.Contains(correction.Content, "Repair attempt 1 of 3") ||
+		!strings.Contains(correction.Content, "Preserve valid values") ||
+		!strings.Contains(correction.Content, "Return only corrected JSON") {
+		t.Fatalf("correction message = %q", correction.Content)
+	}
+	if diagnostics := repairDiagnosticsFromMessage(t, correction.Content); !reflect.DeepEqual(diagnostics, []string{"invalid JSON"}) {
+		t.Fatalf("diagnostics = %#v", diagnostics)
+	}
+}
+
+func TestDefaultOutputRepairerDoesNotAccumulateCandidates(t *testing.T) {
+	client := &recordingRepairClient{response: &domain.GenerateResponse{Content: "corrected"}}
+	repairer := NewDefaultOutputRepairer(client)
+	backing := make([]domain.RenderedMessage, 1, 4)
+	backing[0] = domain.RenderedMessage{Role: "user", Content: "Original task"}
+	before := append([]domain.RenderedMessage(nil), backing...)
+
+	for _, candidate := range []string{"first invalid", "second invalid"} {
+		_, err := repairer.Repair(context.Background(), RepairRequest{
+			OriginalMessages: backing,
+			PreviousOutput:   candidate,
+			Attempt:          1,
+			MaxAttempts:      3,
+			Mode:             domain.ValidationJSON,
+		})
+		if err != nil {
+			t.Fatalf("repair %q: %v", candidate, err)
+		}
+	}
+	if !reflect.DeepEqual(backing, before) {
+		t.Fatalf("caller messages changed: got %#v, want %#v", backing, before)
+	}
+	if len(client.requests) != 2 {
+		t.Fatalf("generation requests = %d, want 2", len(client.requests))
+	}
+	for index, request := range client.requests {
+		messages := request.Prompt.Messages
+		if len(messages) != 3 || messages[0] != backing[0] || messages[1].Role != "assistant" || messages[1].Content != []string{"first invalid", "second invalid"}[index] {
+			t.Fatalf("request %d messages = %#v", index, messages)
+		}
+	}
+}
+
+func TestDefaultOutputRepairerOmitsEmptyCandidateMessage(t *testing.T) {
+	for _, tc := range []struct {
+		name      string
+		candidate string
+	}{
+		{name: "empty", candidate: ""},
+		{name: "whitespace", candidate: " \n\t "},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			client := &recordingRepairClient{response: &domain.GenerateResponse{Content: "corrected"}}
+			repairer := NewDefaultOutputRepairer(client)
+			_, err := repairer.Repair(context.Background(), RepairRequest{
+				OriginalMessages: []domain.RenderedMessage{{Role: "user", Content: "Original task"}},
+				PreviousOutput:   tc.candidate,
+				Attempt:          1,
+				MaxAttempts:      1,
+				Mode:             domain.ValidationBasic,
+			})
+			if err != nil {
+				t.Fatalf("repair: %v", err)
+			}
+			messages := client.requests[0].Prompt.Messages
+			if len(messages) != 2 || messages[1].Role != "user" || !strings.Contains(messages[1].Content, "previous response was empty") {
+				t.Fatalf("messages = %#v", messages)
+			}
+			if strings.Contains(messages[1].Content, tc.candidate) && tc.candidate != "" {
+				t.Fatalf("correction message repeated whitespace candidate: %q", messages[1].Content)
+			}
+		})
+	}
+}
+
+func TestDefaultOutputRepairerUsesModeSpecificGuidance(t *testing.T) {
+	tests := []struct {
+		mode domain.ValidationMode
+		want string
+	}{
+		{mode: domain.ValidationBasic, want: "Return a nonempty response"},
+		{mode: domain.ValidationJSON, want: "Return only corrected JSON"},
+		{mode: domain.ValidationJSONSchema, want: "Return only corrected JSON"},
+	}
+
+	for _, tc := range tests {
+		t.Run(string(tc.mode), func(t *testing.T) {
+			client := &recordingRepairClient{response: &domain.GenerateResponse{Content: "corrected"}}
+			_, err := NewDefaultOutputRepairer(client).Repair(context.Background(), RepairRequest{Mode: tc.mode, Attempt: 1, MaxAttempts: 1})
+			if err != nil {
+				t.Fatalf("repair: %v", err)
+			}
+			if !strings.Contains(client.requests[0].Prompt.Messages[0].Content, tc.want) {
+				t.Fatalf("correction message = %q", client.requests[0].Prompt.Messages[0].Content)
+			}
+		})
+	}
+
+	client := &recordingRepairClient{response: &domain.GenerateResponse{Content: "unexpected"}}
+	_, err := NewDefaultOutputRepairer(client).Repair(context.Background(), RepairRequest{Mode: domain.ValidationNone})
+	if err == nil || !strings.Contains(err.Error(), "unsupported validation mode") {
+		t.Fatalf("unsupported mode error = %v", err)
+	}
+	if len(client.requests) != 0 {
+		t.Fatalf("generation requests = %d, want 0", len(client.requests))
+	}
+}
+
+func TestFormatRepairDiagnosticsBoundsAndPreservesData(t *testing.T) {
+	tests := []struct {
+		name         string
+		errors       []string
+		wantOmission bool
+		check        func(*testing.T, []string)
+	}{
+		{
+			name:   "below limit",
+			errors: []string{"first", "second"},
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if !reflect.DeepEqual(got, []string{"first", "second"}) {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+		{
+			name:   "at limit",
+			errors: []string{strings.Repeat("x", maxRepairDiagnosticBytes-4)},
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if len(got) != 1 || len(got[0]) != maxRepairDiagnosticBytes-4 {
+					t.Fatalf("diagnostics lengths = %#v", got)
+				}
+			},
+		},
+		{
+			name:         "multibyte truncation preserves prior entries",
+			errors:       []string{"first", strings.Repeat("界", maxRepairDiagnosticBytes)},
+			wantOmission: true,
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if len(got) != 3 || got[0] != "first" || !strings.HasPrefix(strings.Repeat("界", maxRepairDiagnosticBytes), got[1]) {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+		{
+			name:         "many diagnostics",
+			errors:       manyRepairDiagnostics(),
+			wantOmission: true,
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if len(got) < 2 || got[0] != manyRepairDiagnostics()[0] {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+		{
+			name: "no room for partial diagnostic",
+			errors: func() []string {
+				omission, _ := json.Marshal(omittedRepairDiagnostics)
+				return []string{
+					strings.Repeat("x", maxRepairDiagnosticBytes-len(omission)-5),
+					strings.Repeat("y", 128),
+				}
+			}(),
+			wantOmission: true,
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if len(got) != 2 || got[1] != omittedRepairDiagnostics {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+		{
+			name:   "invalid UTF-8",
+			errors: []string{"broken\xffinput"},
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if !reflect.DeepEqual(got, []string{"broken\uFFFDinput"}) {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+		{
+			name:         "one huge diagnostic",
+			errors:       []string{strings.Repeat("x", maxRepairDiagnosticBytes*2)},
+			wantOmission: true,
+			check: func(t *testing.T, got []string) {
+				t.Helper()
+				if len(got) != 2 || !strings.HasPrefix(strings.Repeat("x", maxRepairDiagnosticBytes*2), got[0]) {
+					t.Fatalf("diagnostics = %#v", got)
+				}
+			},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			before := append([]string(nil), tc.errors...)
+			encoded := formatRepairDiagnostics(tc.errors)
+			if len(encoded) > maxRepairDiagnosticBytes || !utf8.ValidString(encoded) {
+				t.Fatalf("encoded diagnostic length/UTF-8 = (%d, %v)", len(encoded), utf8.ValidString(encoded))
+			}
+			var got []string
+			if err := json.Unmarshal([]byte(encoded), &got); err != nil {
+				t.Fatalf("decode diagnostics: %v; encoded=%q", err, encoded)
+			}
+			if !reflect.DeepEqual(tc.errors, before) {
+				t.Fatalf("input diagnostics changed: got %#v, want %#v", tc.errors, before)
+			}
+			if hasOmission := len(got) > 0 && got[len(got)-1] == omittedRepairDiagnostics; hasOmission != tc.wantOmission {
+				t.Fatalf("omission = %v, want %v; diagnostics=%#v", hasOmission, tc.wantOmission, got)
+			}
+			tc.check(t, got)
+		})
+	}
+}
+
+func manyRepairDiagnostics() []string {
+	diagnostics := make([]string, 1_000)
+	for index := range diagnostics {
+		diagnostics[index] = fmt.Sprintf("diagnostic %04d %s", index, strings.Repeat("x", 128))
+	}
+	return diagnostics
+}
+
+func TestDefaultOutputRepairerPropagatesGenerationFailures(t *testing.T) {
+	expected := errors.New("generation failed")
+	for _, tc := range []struct {
+		name   string
+		client *recordingRepairClient
+		want   error
+	}{
+		{name: "nil client", want: nil},
+		{name: "generation error", client: &recordingRepairClient{err: expected}, want: expected},
+		{name: "nil response", client: &recordingRepairClient{}, want: nil},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			var repairer OutputRepairer
+			if tc.client == nil {
+				repairer = NewDefaultOutputRepairer(nil)
+			} else {
+				repairer = NewDefaultOutputRepairer(tc.client)
+			}
+			response, err := repairer.Repair(context.Background(), RepairRequest{Mode: domain.ValidationJSON})
+			if response != nil || err == nil {
+				t.Fatalf("response/error = (%+v, %v)", response, err)
+			}
+			if tc.want != nil && !errors.Is(err, tc.want) {
+				t.Fatalf("error = %v, want %v", err, tc.want)
+			}
+		})
+	}
+}
+
+func repairDiagnosticsFromMessage(t *testing.T, message string) []string {
+	t.Helper()
+	const marker = "Validation diagnostics (data):\n"
+	index := strings.Index(message, marker)
+	if index < 0 {
+		t.Fatalf("missing diagnostics marker in %q", message)
+	}
+	encoded := message[index+len(marker):]
+	var diagnostics []string
+	if err := json.Unmarshal([]byte(encoded), &diagnostics); err != nil {
+		t.Fatalf("decode diagnostics: %v", err)
+	}
+	return diagnostics
+}

internal/usecase/runner.go

@@ -17,6 +17,7 @@ import (
 	"gitea.maximumdirect.net/eric/promptkit/internal/capacity"
 	"gitea.maximumdirect.net/eric/promptkit/internal/defaults"
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
 	"gitea.maximumdirect.net/eric/promptkit/internal/llm"
 	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
 	"gitea.maximumdirect.net/eric/promptkit/internal/prompt"
@@ -71,6 +72,11 @@ type preparationState struct {
 	start                time.Time
 }
 
+type preparedOperation struct {
+	run        *domain.PreparedRun
+	validation validate.PreparedValidation
+}
+
 func NewRunner(
 	promptDefs promptdef.Repository,
 	profiles profile.Repository,
@@ -137,18 +143,20 @@ func (r *Runner) Run(ctx context.Context, req domain.RunRequest) (*domain.RunRes
 	}
 	defer release()
 
-	prepared, err := r.completePreparation(ctx, req, state)
+	operation, err := r.completePreparation(ctx, req, state)
 	if err != nil {
 		return nil, err
 	}
 	directAPIKey := state.effectiveModel.APIKey
 
-	return r.executePreparedRun(ctx, prepared, directAPIKey, runID, start, func(
+	return r.executePreparedRun(ctx, operation.run, directAPIKey, runID, start, func(
 		ctx context.Context,
 		artifact *domain.Artifact,
 		attemptsUsed int,
 	) (domain.ValidationResult, error) {
-		return r.validateOutput(ctx, artifact, prepared.OutputContract, attemptsUsed)
+		result, err := operation.validation.Validate(ctx, artifact)
+		result.RepairAttempts = attemptsUsed
+		return result, err
 	})
 }
 
@@ -168,18 +176,20 @@ func (r *Runner) executePreparedRun(
 ) (*domain.RunResult, error) {
 	executionTarget := prepared.EffectiveModelParams
 	executionTarget.APIKey = directAPIKey
-	genResp, err := r.llm.Generate(ctx, domain.GenerateRequest{
+	genResp, err := r.llm.Generate(ctx, newGenerationRequest(
-		Prompt:           domain.RenderedPrompt{SessionID: prepared.SessionID, Messages: prepared.Messages},
+		domain.RenderedPrompt{Messages: prepared.Messages},
-		Target:           executionTarget,
+		prepared.SessionID,
-		TargetPresence:   prepared.TargetPresence,
+		executionTarget,
-		StructuredOutput: prepared.StructuredOutput,
+		prepared.TargetPresence,
-	})
+		prepared.StructuredOutput,
+	))
 	if err != nil {
-		if errors.Is(err, llm.ErrInvalidRequest) {
+		return nil, wrapGenerationError(err)
-			return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
 	}
-		return nil, fmt.Errorf("%w: %w", ErrLLMGenerate, err)
+	if genResp == nil {
+		return nil, fmt.Errorf("%w: model returned nil response", ErrLLMGenerate)
 	}
+	usage := genResp.Usage
 
 	outputArtifact := buildOutputArtifact(genResp.Content, prepared.OutputContract.Format)
 	validationResult, err := validateArtifact(ctx, &outputArtifact, 0)
@@ -193,23 +203,26 @@ func (r *Runner) executePreparedRun(
 			attemptsUsed++
 
 			repairResp, repairErr := r.repairer.Repair(ctx, RepairRequest{
+				OriginalMessages: prepared.Messages,
 				PreviousOutput:   genResp.Content,
 				ValidationErrors: validationResult.Errors,
 				SessionID:        prepared.SessionID,
 				Target:           executionTarget,
+				TargetPresence:   prepared.TargetPresence,
 				StructuredOutput: prepared.StructuredOutput,
 				Attempt:          attemptsUsed,
 				MaxAttempts:      prepared.OutputContract.RepairAttempts,
 				Mode:             prepared.OutputContract.ValidationMode,
 			})
 			if repairErr != nil {
-				return nil, fmt.Errorf("%w: %w", ErrValidation, repairErr)
+				return nil, wrapGenerationError(repairErr)
 			}
 			if repairResp == nil {
-				return nil, fmt.Errorf("%w: repairer returned nil response", ErrValidation)
+				return nil, fmt.Errorf("%w: repairer returned nil response", ErrLLMGenerate)
 			}
 
 			genResp = repairResp
+			usage = addTokenUsage(usage, repairResp.Usage)
 			outputArtifact = buildOutputArtifact(genResp.Content, prepared.OutputContract.Format)
 
 			validationResult, err = validateArtifact(ctx, &outputArtifact, attemptsUsed)
@@ -238,19 +251,40 @@ func (r *Runner) executePreparedRun(
 		Endpoint:             prepared.EffectiveModelParams.Endpoint,
 		EffectiveModelParams: executionTarget,
 		InputHashes:          prepared.InputHashes,
-		Usage:                genResp.Usage,
+		Usage:                usage,
 		StartTime:            start,
 		EndTime:              end,
 		Duration:             end.Sub(start),
 	}, nil
 }
 
+func wrapGenerationError(err error) error {
+	if errors.Is(err, llm.ErrInvalidRequest) {
+		return fmt.Errorf("%w: %w", ErrInvalidRequest, err)
+	}
+	return fmt.Errorf("%w: %w", ErrLLMGenerate, err)
+}
+
+func addTokenUsage(total, next domain.TokenUsage) domain.TokenUsage {
+	return domain.TokenUsage{
+		PromptTokens:     total.PromptTokens + next.PromptTokens,
+		CompletionTokens: total.CompletionTokens + next.CompletionTokens,
+		TotalTokens:      total.TotalTokens + next.TotalTokens,
+		CachedTokens:     total.CachedTokens + next.CachedTokens,
+		CacheWriteTokens: total.CacheWriteTokens + next.CacheWriteTokens,
+	}
+}
+
 func (r *Runner) Prepare(ctx context.Context, req domain.RunRequest) (*domain.PreparedRun, error) {
 	state, err := r.resolvePreparation(ctx, req, time.Now().UTC())
 	if err != nil {
 		return nil, err
 	}
-	return r.completePreparation(ctx, req, state)
+	operation, err := r.completePreparation(ctx, req, state)
+	if err != nil {
+		return nil, err
+	}
+	return operation.run, nil
 }
 
 func (r *Runner) resolvePreparation(
@@ -272,6 +306,10 @@ func (r *Runner) resolvePreparation(
 	}
 	def := promptSelection.definition
 	promptDefinitionHash := promptSelection.hash
+	effectiveContract, err := resolveOutputContract(def, req.Validation)
+	if err != nil {
+		return nil, fmt.Errorf("%w: output contract: %v", ErrInvalidRequest, err)
+	}
 
 	selectedProfileID := strings.TrimSpace(req.ProfileID)
 	if selectedProfileID == "" {
@@ -286,19 +324,16 @@ func (r *Runner) resolvePreparation(
 		return nil, err
 	}
 
-	effectiveModel, targetPresence, err := resolveExecutionTarget(selection.backend, selection.profile, req.Execution)
+	effectiveModel, targetPresence := resolveExecutionTarget(selection.backend, selection.profile, req.Execution)
-	if err != nil {
-		return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
-	}
 	effectiveModel.APIKey = req.APIKey
-	if err := validateResolvedExecutionTarget(effectiveModel); err != nil {
+	effectiveModel, err = normalizeResolvedExecutionTarget(effectiveModel)
+	if err != nil {
 		return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
 	}
 	if err := validateAPIKey(effectiveModel.APIKeyEnv, effectiveModel.APIKey, effectiveModel.APIKeyRequired); err != nil {
 		return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
 	}
 
-	effectiveContract := resolveOutputContract(def, req.Validation)
 	return &preparationState{
 		definition:           def,
 		directSessionID:      directSessionID,
@@ -315,16 +350,68 @@ func (r *Runner) completePreparation(
 	ctx context.Context,
 	req domain.RunRequest,
 	state *preparationState,
-) (*domain.PreparedRun, error) {
+) (*preparedOperation, error) {
-	structuredOutput, err := r.resolveStructuredOutput(
+	validationPlan, err := r.prepareValidation(ctx, state.effectiveContract)
-		ctx,
+	if err != nil {
+		return nil, err
+	}
+	structuredOutput, err := r.structuredOutputFromValidationPlan(
 		state.definition,
 		state.effectiveContract,
+		validationPlan,
 	)
 	if err != nil {
 		return nil, err
 	}
-	return r.completePreparationWithStructuredOutput(ctx, req, state, structuredOutput)
+	prepared, err := r.completePreparationWithStructuredOutput(ctx, req, state, structuredOutput)
+	if err != nil {
+		return nil, err
+	}
+	return &preparedOperation{run: prepared, validation: validationPlan}, nil
+}
+
+func (r *Runner) prepareValidation(
+	ctx context.Context,
+	contract domain.OutputContract,
+) (validate.PreparedValidation, error) {
+	if r.validator == nil {
+		return noOpPreparedValidation{contract: contract}, nil
+	}
+	preparer, ok := r.validator.(validate.ValidationPreparer)
+	if !ok {
+		return nil, fmt.Errorf("%w: validator does not support prepared validation", ErrValidation)
+	}
+	plan, err := preparer.PrepareValidation(ctx, contract)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %w", ErrValidation, err)
+	}
+	if plan == nil {
+		return nil, fmt.Errorf("%w: validator returned nil prepared validation", ErrValidation)
+	}
+	return plan, nil
+}
+
+func (r *Runner) structuredOutputFromValidationPlan(
+	def *domain.PromptDefinition,
+	contract domain.OutputContract,
+	plan validate.PreparedValidation,
+) (*domain.StructuredOutputSpec, error) {
+	if contract.ValidationMode != domain.ValidationJSONSchema {
+		return nil, nil
+	}
+
+	schemaDocument := plan.SchemaDocument()
+	if schemaDocument == nil {
+		if r.validator == nil {
+			return nil, nil
+		}
+		return nil, fmt.Errorf("%w: prepared json_schema validation has no schema document", ErrValidation)
+	}
+	schemaDocument, err := jsonvalue.Copy(schemaDocument)
+	if err != nil {
+		return nil, fmt.Errorf("%w: invalid prepared json_schema schema document: %v", ErrValidation, err)
+	}
+	return structuredOutputSpec(def, schemaDocument), nil
 }
 
 func (r *Runner) completePreparationWithStructuredOutput(
@@ -398,24 +485,6 @@ func (r *Runner) admitRun(ctx context.Context, backendID string) (func(), error)
 	return release, nil
 }
 
-func (r *Runner) resolveStructuredOutput(ctx context.Context, def *domain.PromptDefinition, contract domain.OutputContract) (*domain.StructuredOutputSpec, error) {
-	if contract.ValidationMode != domain.ValidationJSONSchema {
-		return nil, nil
-	}
-
-	loader, ok := r.validator.(validate.SchemaDocumentLoader)
-	if !ok || loader == nil {
-		return nil, fmt.Errorf("%w: json_schema output requires schema document loader", ErrValidation)
-	}
-
-	schemaDoc, err := loader.LoadSchemaDocument(ctx, contract.SchemaPath)
-	if err != nil {
-		return nil, fmt.Errorf("%w: failed to load json schema for structured output: %v", ErrValidation, err)
-	}
-
-	return structuredOutputSpec(def, schemaDoc), nil
-}
-
 func structuredOutputSpec(def *domain.PromptDefinition, schemaDocument any) *domain.StructuredOutputSpec {
 	return &domain.StructuredOutputSpec{
 		Type: domain.StructuredOutputJSONSchema,
@@ -453,25 +522,6 @@ func deriveStructuredSchemaName(promptID string, promptVersion string) string {
 	return name
 }
 
-func (r *Runner) validateOutput(ctx context.Context, artifact *domain.Artifact, contract domain.OutputContract, attemptsUsed int) (domain.ValidationResult, error) {
-	if r.validator == nil || contract.ValidationMode == domain.ValidationNone {
-		return domain.ValidationResult{
-			Status:         domain.ValidationSkipped,
-			Mode:           contract.ValidationMode,
-			SchemaPath:     contract.SchemaPath,
-			RepairAttempts: attemptsUsed,
-			IsValid:        true,
-		}, nil
-	}
-
-	res, err := r.validator.Validate(ctx, artifact, contract)
-	if err != nil {
-		return domain.ValidationResult{}, err
-	}
-	res.RepairAttempts = attemptsUsed
-	return res, nil
-}
-
 func (r *Runner) shouldAttemptRepair(contract domain.OutputContract, validationResult domain.ValidationResult) bool {
 	if r.repairer == nil {
 		return false
@@ -482,7 +532,12 @@ func (r *Runner) shouldAttemptRepair(contract domain.OutputContract, validationR
 	if validationResult.Status != domain.ValidationFailed {
 		return false
 	}
-	return contract.ValidationMode == domain.ValidationJSON || contract.ValidationMode == domain.ValidationJSONSchema
+	switch contract.ValidationMode {
+	case domain.ValidationBasic, domain.ValidationJSON, domain.ValidationJSONSchema:
+		return true
+	default:
+		return false
+	}
 }
 
 func mergeExecutionTarget(base domain.ExecutionTarget, override domain.ExecutionTarget) domain.ExecutionTarget {
@@ -527,7 +582,7 @@ func mergeExecutionTarget(base domain.ExecutionTarget, override domain.Execution
 	return out
 }
 
-func mergeExecutionTargetOverride(base domain.ExecutionTarget, override domain.ExecutionTargetOverride) (domain.ExecutionTarget, domain.ExecutionTargetPresence, error) {
+func mergeExecutionTargetOverride(base domain.ExecutionTarget, override domain.ExecutionTargetOverride) (domain.ExecutionTarget, domain.ExecutionTargetPresence) {
 	out := base
 	var presence domain.ExecutionTargetPresence
 	if override.Endpoint != "" {
@@ -537,30 +592,18 @@ func mergeExecutionTargetOverride(base domain.ExecutionTarget, override domain.E
 		out.Model = override.Model
 	}
 	if override.Temperature != nil {
-		if *override.Temperature < 0 || *override.Temperature > 2 {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, errors.New("temperature must be between 0 and 2")
-		}
 		out.Temperature = *override.Temperature
 		presence.Temperature = true
 	}
 	if override.MaxTokens != nil {
-		if *override.MaxTokens < 0 {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, errors.New("max_tokens must be greater than or equal to 0")
-		}
 		out.MaxTokens = *override.MaxTokens
 		presence.MaxTokens = true
 	}
 	if override.TopP != nil {
-		if *override.TopP < 0 || *override.TopP > 1 {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, errors.New("top_p must be between 0 and 1")
-		}
 		out.TopP = *override.TopP
 		presence.TopP = true
 	}
 	if override.TimeoutSeconds != nil {
-		if *override.TimeoutSeconds < 0 {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, errors.New("timeout_seconds must be greater than or equal to 0")
-		}
 		out.TimeoutSeconds = *override.TimeoutSeconds
 		presence.TimeoutSeconds = true
 	}
@@ -576,35 +619,31 @@ func mergeExecutionTargetOverride(base domain.ExecutionTarget, override domain.E
 	if len(override.ExtraParams) > 0 {
 		out.ExtraParams = copyExtraParams(override.ExtraParams)
 	}
-	return out, presence, nil
+	return out, presence
 }
 
-func resolveExecutionTarget(backendValue *domain.Backend, profileValue *domain.ExecutionProfile, override *domain.ExecutionTargetOverride) (domain.ExecutionTarget, domain.ExecutionTargetPresence, error) {
+func resolveExecutionTarget(backendValue *domain.Backend, profileValue *domain.ExecutionProfile, override *domain.ExecutionTargetOverride) (domain.ExecutionTarget, domain.ExecutionTargetPresence) {
 	out := defaults.ExecutionTargetDefault()
 	out = mergeExecutionTarget(out, backendToTarget(backendValue))
 	out = mergeExecutionTarget(out, executionProfileToTarget(profileValue))
 	var presence domain.ExecutionTargetPresence
 	if override != nil {
-		var err error
+		out, presence = mergeExecutionTargetOverride(out, *override)
-		out, presence, err = mergeExecutionTargetOverride(out, *override)
-		if err != nil {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, err
 	}
-	}
+	return out, presence
-	return out, presence, nil
 }
 
 func validateAPIKey(apiKeyEnv string, apiKey string, apiKeyRequired bool) error {
 	if strings.TrimSpace(apiKey) != "" {
 		return nil
 	}
+	if !apiKeyRequired {
+		return nil
+	}
 	envName := strings.TrimSpace(apiKeyEnv)
 	if envName == "" {
-		if apiKeyRequired {
 		return ErrAPIKeyRequired
 	}
-		return nil
-	}
 	if strings.TrimSpace(os.Getenv(envName)) == "" {
 		return fmt.Errorf("%w: api key environment variable %q is not set", ErrAPIKeyEnvMissing, envName)
 	}
@@ -658,18 +697,21 @@ func copyExtraParams(src map[string]any) map[string]any {
 	return cp
 }
 
-func resolveOutputContract(def *domain.PromptDefinition, override *domain.OutputContract) domain.OutputContract {
+func resolveOutputContract(def *domain.PromptDefinition, override *domain.OutputContract) (domain.OutputContract, error) {
 	contract := def.Validation
 	if contract.Format == "" {
 		contract.Format = def.OutputFormat
 	}
 	if override != nil {
 		contract = *override
-	}
 		if contract.Format == "" {
 			contract.Format = domain.FormatText
 		}
-	return contract
+	}
+	if err := domain.ValidateOutputContract(contract); err != nil {
+		return domain.OutputContract{}, err
+	}
+	return contract, nil
 }
 
 func hashRenderedPrompt(p domain.RenderedPrompt) string {

internal/usecase/runner_test.go

@@ -6,9 +6,11 @@ import (
 	"encoding/hex"
 	"errors"
 	"fmt"
+	"math"
 	"path/filepath"
 	"reflect"
 	"regexp"
+	"strconv"
 	"strings"
 	"sync"
 	"testing"
@@ -189,16 +191,34 @@ func (f *fakeValidator) Validate(ctx context.Context, artifact *domain.Artifact,
 	return f.result, nil
 }
 
-func (f *fakeValidator) LoadSchemaDocument(ctx context.Context, schemaPath string) (any, error) {
+func (f *fakeValidator) PrepareValidation(_ context.Context, contract domain.OutputContract) (validate.PreparedValidation, error) {
+	var schemaDocument any
+	if contract.ValidationMode == domain.ValidationJSONSchema {
 		f.schemaLoads++
-	f.schemaLoadPath = schemaPath
+		f.schemaLoadPath = contract.SchemaPath
 		if f.schemaErr != nil {
 			return nil, f.schemaErr
 		}
-	if f.schemaDoc != nil {
+		schemaDocument = f.schemaDoc
-		return f.schemaDoc, nil
+		if schemaDocument == nil {
+			schemaDocument = map[string]any{"type": "object"}
 		}
-	return map[string]any{"type": "object"}, nil
+	}
+	return &fakePreparedValidator{validator: f, contract: contract, schemaDocument: schemaDocument}, nil
+}
+
+type fakePreparedValidator struct {
+	validator      *fakeValidator
+	contract       domain.OutputContract
+	schemaDocument any
+}
+
+func (p *fakePreparedValidator) Validate(ctx context.Context, artifact *domain.Artifact) (domain.ValidationResult, error) {
+	return p.validator.Validate(ctx, artifact, p.contract)
+}
+
+func (p *fakePreparedValidator) SchemaDocument() any {
+	return p.schemaDocument
 }
 
 type fakeRepairer struct {
@@ -208,6 +228,30 @@ type fakeRepairer struct {
 	reqs      []RepairRequest
 }
 
+type sequenceLLM struct {
+	responses []*domain.GenerateResponse
+	requests  []domain.GenerateRequest
+}
+
+func (c *sequenceLLM) Generate(_ context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
+	c.requests = append(c.requests, req)
+	index := len(c.requests) - 1
+	if index >= len(c.responses) {
+		return nil, errors.New("no generation response configured")
+	}
+	return c.responses[index], nil
+}
+
+type recordingRepairer struct {
+	next OutputRepairer
+	reqs []RepairRequest
+}
+
+func (r *recordingRepairer) Repair(ctx context.Context, req RepairRequest) (*domain.GenerateResponse, error) {
+	r.reqs = append(r.reqs, req)
+	return r.next.Repair(ctx, req)
+}
+
 type fakeRunAdmitter struct {
 	backendIDs   []string
 	err          error
@@ -251,8 +295,10 @@ func (c *controlledRepairLLM) Generate(
 	ctx context.Context,
 	req domain.GenerateRequest,
 ) (*domain.GenerateResponse, error) {
-	isRepair := len(req.Prompt.Messages) > 0 &&
+	messages := req.Prompt.Messages
-		strings.HasPrefix(req.Prompt.Messages[0].Content, "You repair invalid JSON")
+	isRepair := len(messages) >= 2 &&
+		messages[len(messages)-2].Role == "assistant" &&
+		messages[len(messages)-1].Role == "user"
 	c.mu.Lock()
 	c.calls++
 	c.active++
@@ -502,6 +548,35 @@ func TestRunnerDirectSessionResolution(t *testing.T) {
 			t.Fatalf("invalid direct session invoked generation %d times", llmClient.calls)
 		}
 	})
+
+	t.Run("malformed direct value fails before loading or generation", func(t *testing.T) {
+		promptRepo := &fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)}
+		llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: "unexpected"}}
+		runner := NewRunner(
+			promptRepo,
+			&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+			nil,
+			defaultArtifactReader(),
+			defaultRenderer(),
+			llmClient,
+			nil, nil)
+
+		_, err := runner.Run(context.Background(), domain.RunRequest{
+			PromptID:  "p",
+			ProfileID: "exec",
+			SessionID: "session" + string([]byte{0xff}),
+			Inputs:    singleInputRef(),
+		})
+		if !errors.Is(err, ErrInvalidRequest) {
+			t.Fatalf("expected ErrInvalidRequest, got %v", err)
+		}
+		if promptRepo.lastID != "" {
+			t.Fatalf("invalid direct session loaded prompt %q", promptRepo.lastID)
+		}
+		if llmClient.calls != 0 {
+			t.Fatalf("invalid direct session invoked generation %d times", llmClient.calls)
+		}
+	})
 }
 
 func TestRunnerPrepareUsesPromptDefaultProfileWhenNoExplicitProfileID(t *testing.T) {
@@ -704,16 +779,23 @@ func TestRunnerPrepareRequestNumericOverridePresence(t *testing.T) {
 }
 
 func TestRunnerPrepareInvalidRequestNumericOverridesFail(t *testing.T) {
-	tests := []struct {
+	type testCase struct {
 		name     string
 		override *domain.ExecutionTargetOverride
-	}{
+	}
+	tests := []testCase{
 		{name: "temperature below range", override: &domain.ExecutionTargetOverride{Temperature: float64Ptr(-0.1)}},
 		{name: "temperature above range", override: &domain.ExecutionTargetOverride{Temperature: float64Ptr(2.1)}},
 		{name: "max tokens below range", override: &domain.ExecutionTargetOverride{MaxTokens: intPtr(-1)}},
 		{name: "top p below range", override: &domain.ExecutionTargetOverride{TopP: float64Ptr(-0.1)}},
 		{name: "top p above range", override: &domain.ExecutionTargetOverride{TopP: float64Ptr(1.1)}},
 		{name: "timeout below range", override: &domain.ExecutionTargetOverride{TimeoutSeconds: intPtr(-1)}},
+		{name: "temperature is not finite", override: &domain.ExecutionTargetOverride{Temperature: float64Ptr(math.NaN())}},
+		{name: "top p is not finite", override: &domain.ExecutionTargetOverride{TopP: float64Ptr(math.Inf(1))}},
+	}
+	if strconv.IntSize == 64 {
+		durationLimit := int64(math.MaxInt64 / int64(time.Second))
+		tests = append(tests, testCase{name: "timeout cannot be represented as a duration", override: &domain.ExecutionTargetOverride{TimeoutSeconds: intPtr(int(durationLimit) + 1)}})
 	}
 
 	for _, tc := range tests {
@@ -1690,36 +1772,6 @@ func TestRunnerRunSelectedProfileBeatsBuiltInDefault(t *testing.T) {
 	}
 }
 
-func TestRunnerRunBuiltInDefaultsUsedWhenProfileOmitsOptionalFields(t *testing.T) {
-	promptRepo := &fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)}
-	execRepo := &fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
-		"exec": {ID: "exec", Endpoint: "http://profile/v1", Model: "profile-model"},
-	}}
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: "ok"}}
-	runner := NewRunner(promptRepo, execRepo, nil, defaultArtifactReader(), defaultRenderer(), llmClient, nil, nil)
-
-	res, err := runner.Run(context.Background(), domain.RunRequest{
-		PromptID:  "p",
-		ProfileID: "exec",
-		Inputs:    singleInputRef(),
-	})
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-	if res.EffectiveModelParams.Temperature != defaults.ExecutionDefaultTemperature {
-		t.Fatalf("expected default temperature %v, got %v", defaults.ExecutionDefaultTemperature, res.EffectiveModelParams.Temperature)
-	}
-	if res.EffectiveModelParams.TopP != defaults.ExecutionDefaultTopP {
-		t.Fatalf("expected default top_p %v, got %v", defaults.ExecutionDefaultTopP, res.EffectiveModelParams.TopP)
-	}
-	if res.EffectiveModelParams.MaxTokens != defaults.ExecutionDefaultMaxTokens {
-		t.Fatalf("expected default max_tokens %d, got %d", defaults.ExecutionDefaultMaxTokens, res.EffectiveModelParams.MaxTokens)
-	}
-	if res.EffectiveModelParams.TimeoutSeconds != defaults.ExecutionDefaultTimeoutSeconds {
-		t.Fatalf("expected default timeout_seconds %d, got %d", defaults.ExecutionDefaultTimeoutSeconds, res.EffectiveModelParams.TimeoutSeconds)
-	}
-}
-
 func TestRunnerRunAPIKeyEnvResolvesFromEnvironment(t *testing.T) {
 	t.Setenv("PROMPTKIT_TEST_API_KEY", "secret")
 	promptRepo := &fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)}
@@ -1737,22 +1789,26 @@ func TestRunnerRunAPIKeyEnvResolvesFromEnvironment(t *testing.T) {
 	}
 }
 
-func TestRunnerRunAPIKeyEnvMissingEnvironmentValueFailsClearly(t *testing.T) {
+func TestRunnerRunOptionalAPIKeyEnvMissingEnvironmentValueReachesLLM(t *testing.T) {
+	const environmentName = "PROMPTKIT_MISSING_KEY"
+	t.Setenv(environmentName, "")
+
 	promptRepo := &fakePromptRepo{def: promptDef(domain.FormatText, domain.ValidationNone, 0)}
 	execRepo := &fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
-		"exec": {ID: "exec", Endpoint: "http://profile/v1", Model: "profile-model", APIKeyEnv: "PROMPTKIT_MISSING_KEY"},
+		"exec": {ID: "exec", Endpoint: "http://profile/v1", Model: "profile-model", APIKeyEnv: environmentName},
 	}}
-	runner := NewRunner(promptRepo, execRepo, nil, defaultArtifactReader(), defaultRenderer(), &fakeLLM{resp: &domain.GenerateResponse{Content: "ok"}}, nil, nil)
+	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: "ok"}}
+	runner := NewRunner(promptRepo, execRepo, nil, defaultArtifactReader(), defaultRenderer(), llmClient, nil, nil)
 
-	_, err := runner.Run(context.Background(), domain.RunRequest{PromptID: "p", ProfileID: "exec", Inputs: singleInputRef()})
+	result, err := runner.Run(context.Background(), domain.RunRequest{PromptID: "p", ProfileID: "exec", Inputs: singleInputRef()})
-	if !errors.Is(err, ErrInvalidRequest) {
+	if err != nil || result == nil {
-		t.Fatalf("expected ErrInvalidRequest, got %v", err)
+		t.Fatalf("optional credential run = (%+v, %v), want success", result, err)
 	}
-	if !errors.Is(err, ErrAPIKeyEnvMissing) {
+	if llmClient.calls != 1 {
-		t.Fatalf("expected ErrAPIKeyEnvMissing, got %v", err)
+		t.Fatalf("LLM calls = %d, want 1", llmClient.calls)
 	}
-	if !strings.Contains(err.Error(), "PROMPTKIT_MISSING_KEY") {
+	if llmClient.lastReq.Target.APIKeyEnv != environmentName {
-		t.Fatalf("expected missing env name in error, got %v", err)
+		t.Fatalf("LLM api_key_env = %q, want %q", llmClient.lastReq.Target.APIKeyEnv, environmentName)
 	}
 }
 
@@ -2044,50 +2100,272 @@ func TestRunnerRunValidationStillWorks(t *testing.T) {
 	}
 }
 
-func TestRunnerRunStructuredRepairRemainsBoundedAndUsesEffectiveModelSettings(t *testing.T) {
+func TestRunnerRepairStateMachine(t *testing.T) {
-	repairer := &fakeRepairer{responses: []*domain.GenerateResponse{{Content: `{"broken":`}, {Content: `{"still":`}}}
+	failed := func(mode domain.ValidationMode, diagnostic string) domain.ValidationResult {
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: `{"initial":`}}
+		return domain.ValidationResult{
+			Status:  domain.ValidationFailed,
+			Mode:    mode,
+			Errors:  []string{diagnostic},
+			IsValid: false,
+		}
+	}
+	passed := func(mode domain.ValidationMode) domain.ValidationResult {
+		return domain.ValidationResult{
+			Status:  domain.ValidationPassed,
+			Mode:    mode,
+			IsValid: true,
+		}
+	}
+	responses := func(count int) []*domain.GenerateResponse {
+		values := make([]*domain.GenerateResponse, count)
+		for index := range values {
+			unit := index + 1
+			values[index] = &domain.GenerateResponse{
+				Content: fmt.Sprintf(`{"candidate":%d}`, index),
+				Usage: domain.TokenUsage{
+					PromptTokens:     unit,
+					CompletionTokens: unit * 10,
+					TotalTokens:      unit * 100,
+					CachedTokens:     unit * 1000,
+					CacheWriteTokens: unit * 10000,
+				},
+			}
+		}
+		return values
+	}
+	zeroOverrides := &domain.ExecutionTargetOverride{
+		Temperature:    float64Ptr(0),
+		MaxTokens:      intPtr(0),
+		TopP:           float64Ptr(0),
+		TimeoutSeconds: intPtr(0),
+	}
+	allPresent := domain.ExecutionTargetPresence{
+		Temperature:    true,
+		MaxTokens:      true,
+		TopP:           true,
+		TimeoutSeconds: true,
+	}
+	emptyThenValid := responses(2)
+	emptyThenValid[0].Content = " \t "
 
+	tests := []struct {
+		name              string
+		mode              domain.ValidationMode
+		budget            int
+		validationResults []domain.ValidationResult
+		responses         []*domain.GenerateResponse
+		execution         *domain.ExecutionTargetOverride
+		wantPresence      domain.ExecutionTargetPresence
+		wantRepairs       int
+		wantStatus        domain.ValidationStatus
+		structured        bool
+	}{
+		{
+			name:              "initial success does not repair",
+			mode:              domain.ValidationJSON,
+			budget:            3,
+			validationResults: []domain.ValidationResult{passed(domain.ValidationJSON)},
+			responses:         responses(1),
+			wantStatus:        domain.ValidationPassed,
+		},
+		{
+			name:              "empty basic output repairs successfully",
+			mode:              domain.ValidationBasic,
+			budget:            3,
+			validationResults: []domain.ValidationResult{failed(domain.ValidationBasic, "empty output"), passed(domain.ValidationBasic)},
+			responses:         emptyThenValid,
+			wantRepairs:       1,
+			wantStatus:        domain.ValidationPassed,
+		},
+		{
+			name:   "inherited numeric values remain absent",
+			mode:   domain.ValidationJSON,
+			budget: 1,
+			validationResults: []domain.ValidationResult{
+				failed(domain.ValidationJSON, "initial syntax"),
+				passed(domain.ValidationJSON),
+			},
+			responses:   responses(2),
+			wantRepairs: 1,
+			wantStatus:  domain.ValidationPassed,
+		},
+		{
+			name:       "explicit numeric zeros remain present",
+			mode:       domain.ValidationJSONSchema,
+			budget:     1,
+			execution:  zeroOverrides,
+			structured: true,
+			validationResults: []domain.ValidationResult{
+				failed(domain.ValidationJSONSchema, "initial schema mismatch"),
+				passed(domain.ValidationJSONSchema),
+			},
+			responses:    responses(2),
+			wantPresence: allPresent,
+			wantRepairs:  1,
+			wantStatus:   domain.ValidationPassed,
+		},
+		{
+			name:   "successful repair stops below larger budget",
+			mode:   domain.ValidationJSON,
+			budget: 3,
+			validationResults: []domain.ValidationResult{
+				failed(domain.ValidationJSON, "candidate zero"),
+				failed(domain.ValidationJSON, "candidate one"),
+				passed(domain.ValidationJSON),
+			},
+			responses:   responses(3),
+			wantRepairs: 2,
+			wantStatus:  domain.ValidationPassed,
+		},
+		{
+			name:   "failed repairs exhaust exact larger budget",
+			mode:   domain.ValidationJSON,
+			budget: 3,
+			validationResults: []domain.ValidationResult{
+				failed(domain.ValidationJSON, "candidate zero"),
+				failed(domain.ValidationJSON, "candidate one"),
+				failed(domain.ValidationJSON, "candidate two"),
+				failed(domain.ValidationJSON, "candidate three"),
+			},
+			responses:   responses(4),
+			wantRepairs: 3,
+			wantStatus:  domain.ValidationFailed,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			definition := promptDef(domain.FormatJSON, tc.mode, tc.budget)
+			plan := &recordingPreparedValidation{results: tc.validationResults}
+			if tc.structured {
+				definition.Validation.SchemaPath = "schema.json"
+				plan.schemaDocument = map[string]any{"type": "object"}
+			}
+			validator := &recordingValidationPreparer{plan: plan}
+			client := &sequenceLLM{responses: tc.responses}
+			repairer := &recordingRepairer{next: NewDefaultOutputRepairer(client)}
 			runner := NewRunnerWithRepairer(
-		&fakePromptRepo{def: promptDef(domain.FormatJSON, domain.ValidationJSON, 1)},
+				&fakePromptRepo{def: definition},
 				&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
-			"exec": {ID: "exec", BackendID: "custom", Model: "profile-model", TimeoutSeconds: 55},
+					"exec": {ID: "exec", BackendID: "custom", Model: "profile-model"},
-		}}, fakeBackendResolver{backends: map[string]domain.Backend{
+				}},
-			"custom": {ID: "custom", Endpoint: "http://backend/v1"},
+				fakeBackendResolver{backends: map[string]domain.Backend{
+					"custom": {ID: "custom", Endpoint: "http://backend.example/v1"},
 				}},
-
 				defaultArtifactReader(),
 				defaultRenderer(),
-		llmClient,
+				client,
-		validate.NewStandardValidator("."),
+				validator,
-		repairer, nil)
+				repairer,
+				nil,
+			)
 
-	res, err := runner.Run(context.Background(), domain.RunRequest{
+			result, err := runner.Run(context.Background(), domain.RunRequest{
 				PromptID:  "p",
 				ProfileID: "exec",
+				SessionID: " repair-session ",
+				APIKey:    "direct-secret",
 				Inputs:    singleInputRef(),
-		Execution: &domain.ExecutionTargetOverride{Endpoint: "http://override/v1", Model: "override-model", TimeoutSeconds: intPtr(22)},
+				Execution: tc.execution,
 			})
 			if err != nil {
-		t.Fatalf("expected no error, got %v", err)
+				t.Fatalf("run: %v", err)
 			}
-	if repairer.calls != 1 || res.Validation.RepairAttempts != 1 {
+			if len(client.requests) != tc.wantRepairs+1 || len(repairer.reqs) != tc.wantRepairs {
-		t.Fatalf("expected one bounded repair, calls=%d attempts=%d", repairer.calls, res.Validation.RepairAttempts)
+				t.Fatalf(
+					"generation/repair calls = (%d, %d), want (%d, %d)",
+					len(client.requests), len(repairer.reqs), tc.wantRepairs+1, tc.wantRepairs,
+				)
 			}
-	if len(repairer.reqs) != 1 {
+			if len(plan.artifacts) != tc.wantRepairs+1 {
-		t.Fatalf("expected one repair request, got %d", len(repairer.reqs))
+				t.Fatalf("validation calls = %d, want %d", len(plan.artifacts), tc.wantRepairs+1)
 			}
-	if repairer.reqs[0].Target.Endpoint != "http://override/v1" || repairer.reqs[0].Target.Model != "override-model" {
+
-		t.Fatalf("expected repair to use effective target, got %+v", repairer.reqs[0].Target)
+			initialRequest := client.requests[0]
+			if initialRequest.TargetPresence != tc.wantPresence {
+				t.Fatalf("initial target presence = %+v, want %+v", initialRequest.TargetPresence, tc.wantPresence)
 			}
-	if repairer.reqs[0].Target.TimeoutSeconds != 22 {
+			if initialRequest.Target.APIKey != "direct-secret" || initialRequest.Target.BackendID != "custom" ||
-		t.Fatalf("expected repair to use effective timeout, got %d", repairer.reqs[0].Target.TimeoutSeconds)
+				initialRequest.Prompt.SessionID != "repair-session" {
+				t.Fatalf("initial common request fields = %+v", initialRequest)
 			}
-	if llmClient.lastReq.Target.BackendID != "custom" ||
+			if initialRequest.Target.Endpoint != "http://backend.example/v1" ||
-		repairer.reqs[0].Target.BackendID != "custom" ||
+				initialRequest.Target.Model != "profile-model" ||
-		res.SelectedBackendID != "custom" {
+				initialRequest.Target.Temperature != 0 || initialRequest.Target.MaxTokens != 0 || initialRequest.Target.TopP != 0 {
-		t.Fatalf("expected backend identity in generation, repair, and result: generate=%q repair=%q result=%q",
+				t.Fatalf("initial effective target = %+v", initialRequest.Target)
-			llmClient.lastReq.Target.BackendID, repairer.reqs[0].Target.BackendID, res.SelectedBackendID)
+			}
+			if tc.structured != (initialRequest.StructuredOutput != nil) {
+				t.Fatalf("initial structured output = %+v, want present %v", initialRequest.StructuredOutput, tc.structured)
+			}
+			if tc.structured && (initialRequest.StructuredOutput.JSONSchema == nil ||
+				initialRequest.StructuredOutput.JSONSchema.Name != "p_1") {
+				t.Fatalf("initial JSON Schema metadata = %+v, want derived schema name p_1", initialRequest.StructuredOutput)
+			}
+
+			for index, req := range repairer.reqs {
+				if req.Attempt != index+1 || req.MaxAttempts != tc.budget || req.Mode != tc.mode {
+					t.Fatalf("repair request %d progression = %+v", index, req)
+				}
+				if req.PreviousOutput != tc.responses[index].Content ||
+					!reflect.DeepEqual(req.ValidationErrors, tc.validationResults[index].Errors) {
+					t.Fatalf("repair request %d prior state = %+v", index, req)
+				}
+				if !reflect.DeepEqual(req.OriginalMessages, initialRequest.Prompt.Messages) {
+					t.Fatalf("repair request %d original messages drifted: %#v", index, req.OriginalMessages)
+				}
+				if req.TargetPresence != tc.wantPresence || !reflect.DeepEqual(req.Target, initialRequest.Target) ||
+					req.SessionID != initialRequest.Prompt.SessionID ||
+					!reflect.DeepEqual(req.StructuredOutput, initialRequest.StructuredOutput) {
+					t.Fatalf("repair request %d common fields drifted: %+v", index, req)
+				}
+
+				generated := client.requests[index+1]
+				if generated.TargetPresence != initialRequest.TargetPresence ||
+					!reflect.DeepEqual(generated.Target, initialRequest.Target) ||
+					generated.Prompt.SessionID != initialRequest.Prompt.SessionID ||
+					!reflect.DeepEqual(generated.StructuredOutput, initialRequest.StructuredOutput) {
+					t.Fatalf("repair generation request %d common fields drifted: %+v", index, generated)
+				}
+				expectedMessages := len(initialRequest.Prompt.Messages) + 1
+				if strings.TrimSpace(tc.responses[index].Content) != "" {
+					expectedMessages++
+				}
+				if len(generated.Prompt.Messages) != expectedMessages ||
+					generated.Prompt.Messages[len(generated.Prompt.Messages)-1].Role != "user" {
+					t.Fatalf("repair generation request %d messages = %#v", index, generated.Prompt.Messages)
+				}
+				if strings.TrimSpace(tc.responses[index].Content) != "" {
+					assistant := generated.Prompt.Messages[len(generated.Prompt.Messages)-2]
+					if assistant.Role != "assistant" || assistant.Content != tc.responses[index].Content {
+						t.Fatalf("repair generation request %d candidate = %+v", index, assistant)
+					}
+				}
+			}
+
+			lastResponse := tc.responses[tc.wantRepairs]
+			if result.RawOutput != lastResponse.Content || string(result.Artifact.Body) != lastResponse.Content {
+				t.Fatalf("final output = (%q, %q), want %q", result.RawOutput, result.Artifact.Body, lastResponse.Content)
+			}
+			if result.Validation.Status != tc.wantStatus || result.Validation.RepairAttempts != tc.wantRepairs {
+				t.Fatalf("final validation = %+v, want status %q and %d repairs", result.Validation, tc.wantStatus, tc.wantRepairs)
+			}
+			if result.SelectedBackendID != "custom" || result.SessionID != "repair-session" ||
+				result.EffectiveModelParams.APIKey != "" {
+				t.Fatalf("result execution metadata = %+v", result)
+			}
+
+			var wantUsage domain.TokenUsage
+			for _, response := range tc.responses[:tc.wantRepairs+1] {
+				wantUsage.PromptTokens += response.Usage.PromptTokens
+				wantUsage.CompletionTokens += response.Usage.CompletionTokens
+				wantUsage.TotalTokens += response.Usage.TotalTokens
+				wantUsage.CachedTokens += response.Usage.CachedTokens
+				wantUsage.CacheWriteTokens += response.Usage.CacheWriteTokens
+			}
+			if result.Usage != wantUsage {
+				t.Fatalf("cumulative usage = %+v, want %+v", result.Usage, wantUsage)
+			}
+		})
 	}
 }
 
@@ -2182,96 +2460,6 @@ func TestRunnerSchedulesInitialAndRepairGenerationThroughOneBackendPool(t *testi
 	}
 }
 
-func TestRunnerRunRepairCarriesEffectiveSessionID(t *testing.T) {
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: `{"broken":`}}
-	runner := NewRunnerWithRepairer(
-		&fakePromptRepo{def: promptDef(domain.FormatJSON, domain.ValidationJSON, 1)},
-		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
-			"exec": {ID: "exec", Endpoint: "http://example.test/v1", Model: "model"},
-		}},
-		nil,
-		defaultArtifactReader(),
-		defaultRenderer(),
-		llmClient,
-		validate.NewStandardValidator("."),
-		NewDefaultOutputRepairer(llmClient), nil)
-
-	result, err := runner.Run(context.Background(), domain.RunRequest{
-		PromptID:  "p",
-		ProfileID: "exec",
-		SessionID: " repair-session ",
-		Inputs:    singleInputRef(),
-	})
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-	if llmClient.calls != 2 {
-		t.Fatalf("expected initial generation and one repair, got %d calls", llmClient.calls)
-	}
-	if llmClient.lastReq.Prompt.SessionID != "repair-session" {
-		t.Fatalf("expected repair generation to retain effective session, got %q", llmClient.lastReq.Prompt.SessionID)
-	}
-	if result.SessionID != "repair-session" {
-		t.Fatalf("expected result to retain effective session, got %q", result.SessionID)
-	}
-}
-
-func TestRunnerRunJSONSchemaRepairCarriesStructuredOutputSpec(t *testing.T) {
-	def := promptDef(domain.FormatJSON, domain.ValidationJSONSchema, 1)
-	def.Validation.SchemaPath = "events.schema.json"
-
-	validator := &fakeValidator{
-		result: domain.ValidationResult{
-			Status:  domain.ValidationFailed,
-			Mode:    domain.ValidationJSONSchema,
-			Errors:  []string{"schema mismatch"},
-			IsValid: false,
-		},
-		schemaDoc: map[string]any{
-			"type": "object",
-			"properties": map[string]any{
-				"events": map[string]any{"type": "array"},
-			},
-		},
-	}
-	repairer := &fakeRepairer{
-		responses: []*domain.GenerateResponse{
-			{Content: `{"events":[]}`},
-		},
-	}
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: `{"events":[1]}`}}
-	runner := NewRunnerWithRepairer(
-		&fakePromptRepo{def: def},
-		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}}, nil,
-
-		defaultArtifactReader(),
-		defaultRenderer(),
-		llmClient,
-		validator,
-		repairer, nil)
-
-	_, err := runner.Run(context.Background(), domain.RunRequest{
-		PromptID:  "p",
-		ProfileID: "exec",
-		Inputs:    singleInputRef(),
-	})
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-	if llmClient.lastReq.StructuredOutput == nil || llmClient.lastReq.StructuredOutput.JSONSchema == nil {
-		t.Fatalf("expected initial llm request to include structured output, got %+v", llmClient.lastReq.StructuredOutput)
-	}
-	if len(repairer.reqs) != 1 {
-		t.Fatalf("expected one repair request, got %d", len(repairer.reqs))
-	}
-	if repairer.reqs[0].StructuredOutput == nil || repairer.reqs[0].StructuredOutput.JSONSchema == nil {
-		t.Fatalf("expected repair request structured output, got %+v", repairer.reqs[0].StructuredOutput)
-	}
-	if repairer.reqs[0].StructuredOutput.JSONSchema.Name != "p_1" {
-		t.Fatalf("expected derived schema name p_1, got %q", repairer.reqs[0].StructuredOutput.JSONSchema.Name)
-	}
-}
-
 func TestExecutionProfileToTargetPopulatesAllFieldsAndCopiesExtraParams(t *testing.T) {
 	src := &domain.ExecutionProfile{
 		ID:              "exec",
@@ -2330,10 +2518,7 @@ func TestResolveExecutionTargetProfileValuesPopulateAllSupportedFields(t *testin
 		},
 	}
 
-	target, presence, err := resolveExecutionTarget(nil, profileValue, nil)
+	target, presence := resolveExecutionTarget(nil, profileValue, nil)
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
 	if presence != (domain.ExecutionTargetPresence{}) {
 		t.Fatalf("expected no request override presence, got %+v", presence)
 	}
@@ -2385,10 +2570,7 @@ func TestResolveExecutionTargetRuntimeOverridesBeatProfileForAllOverrideableFiel
 		},
 	}
 
-	target, presence, err := resolveExecutionTarget(nil, profileValue, override)
+	target, presence := resolveExecutionTarget(nil, profileValue, override)
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
 	if presence != (domain.ExecutionTargetPresence{Temperature: true, MaxTokens: true, TopP: true, TimeoutSeconds: true}) {
 		t.Fatalf("unexpected override presence: %+v", presence)
 	}
@@ -2435,12 +2617,9 @@ func TestResolveExecutionTargetReasoningOverrideStates(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			target, _, err := resolveExecutionTarget(nil, profileValue, &domain.ExecutionTargetOverride{
+			target, _ := resolveExecutionTarget(nil, profileValue, &domain.ExecutionTargetOverride{
 				ReasoningEffort: tt.override,
 			})
-			if err != nil {
-				t.Fatalf("resolve execution target: %v", err)
-			}
 			if target.ReasoningEffort != tt.want {
 				t.Fatalf("reasoning effort = %q, want %q", target.ReasoningEffort, tt.want)
 			}
@@ -2569,10 +2748,7 @@ func TestResolveExecutionTargetUsesBackendProfileAndRequestPrecedence(t *testing
 		ExtraParams: map[string]any{"request": true},
 	}
 
-	target, _, err := resolveExecutionTarget(backendValue, profileValue, override)
+	target, _ := resolveExecutionTarget(backendValue, profileValue, override)
-	if err != nil {
-		t.Fatalf("resolve target: %v", err)
-	}
 	if target.BackendID != "custom" {
 		t.Fatalf("endpoint override changed backend identity: %+v", target)
 	}
@@ -2583,12 +2759,9 @@ func TestResolveExecutionTargetUsesBackendProfileAndRequestPrecedence(t *testing
 		t.Fatalf("expected whole-map request replacement, got %#v", target.ExtraParams)
 	}
 
-	target, _, err = resolveExecutionTarget(backendValue, &domain.ExecutionProfile{
+	target, _ = resolveExecutionTarget(backendValue, &domain.ExecutionProfile{
 		ID: "exec", BackendID: "custom", Model: "profile-model",
 	}, nil)
-	if err != nil {
-		t.Fatalf("resolve backend defaults: %v", err)
-	}
 	if target.Endpoint != backendValue.Endpoint ||
 		target.APIKeyEnv != backendValue.APIKeyEnv ||
 		!reflect.DeepEqual(target.ExtraParams, backendValue.ExtraParams) {

internal/validate/cancellation_test.go

@@ -0,0 +1,311 @@
+package validate
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"testing/fstest"
+	"time"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+	"github.com/santhosh-tekuri/jsonschema/v6"
+)
+
+func TestValidationCancellationBeforeWorkDoesNotOpenSchemaSource(t *testing.T) {
+	source := &countingSchemaFS{FS: fstest.MapFS{
+		"schema.json": {Data: []byte(`{"type":"object"}`)},
+	}}
+	validator := NewFSValidator(source, ".").(ValidationPreparer)
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	plan, err := validator.PrepareValidation(ctx, domain.OutputContract{
+		ValidationMode: domain.ValidationJSONSchema,
+		SchemaPath:     "schema.json",
+	})
+	if plan != nil || !errors.Is(err, context.Canceled) {
+		t.Fatalf("PrepareValidation() = (%v, %v), want nil plan and context cancellation", plan, err)
+	}
+	if opens := source.opens.Load(); opens != 0 {
+		t.Fatalf("schema source opens = %d, want 0", opens)
+	}
+}
+
+func TestValidationCancellationBetweenReferencedSchemaReadChunks(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	source := &controlledSchemaFS{
+		FS: fstest.MapFS{
+			"root.json":  {Data: []byte(`{"$ref":"child.json"}`)},
+			"child.json": {Data: []byte(`{"type":"string"}` + strings.Repeat(" ", schemaReadChunkSize*2))},
+		},
+		target: "child.json",
+		cancel: cancel,
+	}
+	validator := NewFSValidator(source, ".").(ValidationPreparer)
+
+	plan, err := validator.PrepareValidation(ctx, domain.OutputContract{
+		ValidationMode: domain.ValidationJSONSchema,
+		SchemaPath:     "root.json",
+	})
+	if plan != nil || !errors.Is(err, context.Canceled) {
+		t.Fatalf("PrepareValidation() = (%v, %v), want nil plan and context cancellation", plan, err)
+	}
+	if reads := source.reads.Load(); reads != 1 {
+		t.Fatalf("controlled child reads = %d, want 1", reads)
+	}
+	if closes := source.closes.Load(); closes != 1 {
+		t.Fatalf("controlled child closes = %d, want 1", closes)
+	}
+}
+
+func TestDecodeCancellationAfterSynchronousCallWins(t *testing.T) {
+	ctx := newCheckpointContext(2)
+	value, err := decodeJSONValue(ctx, []byte(`{"value":1}`))
+	if value != nil || !errors.Is(err, context.Canceled) {
+		t.Fatalf("decodeJSONValue() = (%v, %v), want nil value and context cancellation", value, err)
+	}
+}
+
+func TestCompileCancellationAfterSynchronousCallWins(t *testing.T) {
+	for _, dependencyErr := range []error{nil, errors.New("compile failed")} {
+		ctx, cancel := context.WithCancel(context.Background())
+		compiled := &jsonschema.Schema{}
+		schema, err := compileJSONSchema(ctx, "promptkit-schema:/root.json", func(string) (*jsonschema.Schema, error) {
+			cancel()
+			return compiled, dependencyErr
+		})
+		if schema != nil || !errors.Is(err, context.Canceled) {
+			t.Fatalf("compileJSONSchema() = (%v, %v), want authoritative context cancellation", schema, err)
+		}
+		if dependencyErr != nil && errors.Is(err, dependencyErr) {
+			t.Fatalf("compileJSONSchema() error = %v, dependency error should not win", err)
+		}
+	}
+}
+
+func TestExecutionCancellationAfterSynchronousCallWins(t *testing.T) {
+	for _, dependencyErr := range []error{nil, errors.New("schema mismatch")} {
+		ctx, cancel := context.WithCancel(context.Background())
+		executor := &controlledSchemaExecutor{cancel: cancel, err: dependencyErr}
+		validationErrors, err := executeJSONSchema(ctx, executor, map[string]any{"ok": true})
+		if validationErrors != nil || !errors.Is(err, context.Canceled) {
+			t.Fatalf("executeJSONSchema() = (%v, %v), want no result and context cancellation", validationErrors, err)
+		}
+		if calls := executor.calls.Load(); calls != 1 {
+			t.Fatalf("schema execution calls = %d, want 1", calls)
+		}
+	}
+}
+
+func TestCancellationDoesNotDetachBlockedSchemaRead(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	started := make(chan struct{})
+	release := make(chan struct{})
+	source := &controlledSchemaFS{
+		FS: fstest.MapFS{
+			"schema.json": {Data: []byte(`{"type":"object"}`)},
+		},
+		target:  "schema.json",
+		started: started,
+		release: release,
+	}
+	validator := NewFSValidator(source, ".").(ValidationPreparer)
+	type outcome struct {
+		plan PreparedValidation
+		err  error
+	}
+	result := make(chan outcome, 1)
+	go func() {
+		plan, err := validator.PrepareValidation(ctx, domain.OutputContract{
+			ValidationMode: domain.ValidationJSONSchema,
+			SchemaPath:     "schema.json",
+		})
+		result <- outcome{plan: plan, err: err}
+	}()
+
+	waitForSignal(t, started, "schema read to start")
+	cancel()
+	select {
+	case got := <-result:
+		t.Fatalf("blocked dependency returned before release: (%v, %v)", got.plan, got.err)
+	default:
+	}
+	close(release)
+	got := waitForOutcome(t, result)
+	if got.plan != nil || !errors.Is(got.err, context.Canceled) {
+		t.Fatalf("PrepareValidation() after release = (%v, %v), want nil plan and context cancellation", got.plan, got.err)
+	}
+	if reads := source.reads.Load(); reads != 1 {
+		t.Fatalf("blocked reads = %d, want 1 completed read", reads)
+	}
+}
+
+func TestCancellationDoesNotDetachBlockedSchemaExecution(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	started := make(chan struct{})
+	release := make(chan struct{})
+	executor := &controlledSchemaExecutor{started: started, release: release}
+	result := make(chan error, 1)
+	go func() {
+		_, err := executeJSONSchema(ctx, executor, map[string]any{"ok": true})
+		result <- err
+	}()
+
+	waitForSignal(t, started, "schema execution to start")
+	cancel()
+	select {
+	case err := <-result:
+		t.Fatalf("blocked dependency returned before release: %v", err)
+	default:
+	}
+	close(release)
+	select {
+	case err := <-result:
+		if !errors.Is(err, context.Canceled) {
+			t.Fatalf("executeJSONSchema() after release = %v, want context cancellation", err)
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for released schema execution")
+	}
+	if calls := executor.calls.Load(); calls != 1 {
+		t.Fatalf("schema execution calls = %d, want 1 completed call", calls)
+	}
+}
+
+type countingSchemaFS struct {
+	fs.FS
+	opens atomic.Int32
+}
+
+func (f *countingSchemaFS) Open(name string) (fs.File, error) {
+	f.opens.Add(1)
+	return f.FS.Open(name)
+}
+
+type controlledSchemaFS struct {
+	fs.FS
+	target  string
+	cancel  context.CancelFunc
+	started chan struct{}
+	release chan struct{}
+	once    sync.Once
+	reads   atomic.Int32
+	closes  atomic.Int32
+}
+
+func (f *controlledSchemaFS) Open(name string) (fs.File, error) {
+	file, err := f.FS.Open(name)
+	if err != nil || name != f.target {
+		return file, err
+	}
+	return &controlledSchemaFile{File: file, owner: f}, nil
+}
+
+type controlledSchemaFile struct {
+	fs.File
+	owner *controlledSchemaFS
+}
+
+func (f *controlledSchemaFile) Read(buffer []byte) (int, error) {
+	f.owner.once.Do(func() {
+		if f.owner.started != nil {
+			close(f.owner.started)
+		}
+		if f.owner.release != nil {
+			<-f.owner.release
+		}
+	})
+	n, err := f.File.Read(buffer)
+	f.owner.reads.Add(1)
+	if f.owner.cancel != nil {
+		f.owner.cancel()
+	}
+	return n, err
+}
+
+func (f *controlledSchemaFile) Close() error {
+	f.owner.closes.Add(1)
+	return f.File.Close()
+}
+
+type controlledSchemaExecutor struct {
+	cancel  context.CancelFunc
+	err     error
+	started chan struct{}
+	release chan struct{}
+	calls   atomic.Int32
+}
+
+func (e *controlledSchemaExecutor) Validate(any) error {
+	e.calls.Add(1)
+	if e.started != nil {
+		close(e.started)
+	}
+	if e.release != nil {
+		<-e.release
+	}
+	if e.cancel != nil {
+		e.cancel()
+	}
+	return e.err
+}
+
+type checkpointContext struct {
+	context.Context
+	mu        sync.Mutex
+	remaining int
+	canceled  bool
+	done      chan struct{}
+}
+
+func newCheckpointContext(checksUntilCancel int) *checkpointContext {
+	return &checkpointContext{
+		Context:   context.Background(),
+		remaining: checksUntilCancel,
+		done:      make(chan struct{}),
+	}
+}
+
+func (c *checkpointContext) Done() <-chan struct{} {
+	return c.done
+}
+
+func (c *checkpointContext) Err() error {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.canceled {
+		return context.Canceled
+	}
+	c.remaining--
+	if c.remaining == 0 {
+		c.canceled = true
+		close(c.done)
+		return context.Canceled
+	}
+	return nil
+}
+
+func waitForSignal(t *testing.T, signal <-chan struct{}, description string) {
+	t.Helper()
+	select {
+	case <-signal:
+	case <-time.After(2 * time.Second):
+		t.Fatalf("timed out waiting for %s", description)
+	}
+}
+
+func waitForOutcome[T any](t *testing.T, result <-chan T) T {
+	t.Helper()
+	select {
+	case value := <-result:
+		return value
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for released dependency")
+		var zero T
+		return zero
+	}
+}

internal/validate/standard_validator.go

@@ -1,15 +1,18 @@
 package validate
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
 	"net/url"
 	"os"
 	"path"
 	"path/filepath"
+	"runtime"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
@@ -19,6 +22,8 @@ import (
 
 const jsonSchemaDraft2020 = "https://json-schema.org/draft/2020-12/schema"
 
+const schemaReadChunkSize = 64 * 1024
+
 // StandardValidator provides basic, JSON, and JSON Schema output validation.
 type StandardValidator struct {
 	schemaBaseDir string
@@ -48,7 +53,7 @@ func (v *FSValidator) Validate(ctx context.Context, artifact *domain.Artifact, c
 type preparedValidation struct {
 	contract       domain.OutputContract
 	schemaDocument any
-	schema         *jsonschema.Schema
+	schema         schemaExecutor
 }
 
 func (p *preparedValidation) Validate(ctx context.Context, artifact *domain.Artifact) (domain.ValidationResult, error) {
@@ -59,14 +64,11 @@ func (p *preparedValidation) SchemaDocument() any {
 	return p.schemaDocument
 }
 
-func (p *preparedValidation) validateJSONSchema(instance any, _ string) ([]string, error) {
+func (p *preparedValidation) validateJSONSchema(ctx context.Context, instance any, _ string) ([]string, error) {
 	if p.schema == nil {
 		return nil, errors.New("prepared JSON schema is unavailable")
 	}
-	if err := p.schema.Validate(instance); err != nil {
+	return executeJSONSchema(ctx, p.schema, instance)
-		return []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
 }
 
 func (v *StandardValidator) PrepareValidation(ctx context.Context, contract domain.OutputContract) (PreparedValidation, error) {
@@ -79,11 +81,11 @@ func (v *StandardValidator) PrepareValidation(ctx context.Context, contract doma
 		return prepared, nil
 	}
 
-	resolvedSchemaPath, err := v.resolveSchemaPath(contract.SchemaPath)
+	resolvedSchemaPath, err := v.resolveSchemaPath(ctx, contract.SchemaPath)
 	if err != nil {
 		return nil, err
 	}
-	schemaDocument, err := loadJSONSchemaFile(resolvedSchemaPath)
+	schemaDocument, err := loadJSONSchemaFile(ctx, resolvedSchemaPath)
 	if err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", resolvedSchemaPath, err)
 	}
@@ -91,11 +93,24 @@ func (v *StandardValidator) PrepareValidation(ctx context.Context, contract doma
 	if err != nil {
 		return nil, err
 	}
-	compiler := newSchemaCompiler(standardSchemaLoader{root: schemaRoot})
+	if err := ctx.Err(); err != nil {
-	if err := compiler.AddResource(resolvedSchemaPath, schemaDocument); err != nil {
+		return nil, err
+	}
+	compiler := newSchemaCompiler(standardSchemaLoader{ctx: ctx, root: schemaRoot})
+	resourceURL := fileSchemaResourceURL(resolvedSchemaPath)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if err := compiler.AddResource(resourceURL.String(), schemaDocument); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
 		return nil, fmt.Errorf("failed to register JSON schema %q: %w", resolvedSchemaPath, err)
 	}
-	schema, err := compiler.Compile(resolvedSchemaPath)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	schema, err := compileJSONSchema(ctx, resourceURL.String(), compiler.Compile)
 	if err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", resolvedSchemaPath, err)
 	}
@@ -118,16 +133,25 @@ func (v *FSValidator) PrepareValidation(ctx context.Context, contract domain.Out
 		return prepared, nil
 	}
 
-	schemaName, schemaDocument, err := v.loadSchemaDocument(contract.SchemaPath)
+	schemaName, schemaDocument, err := v.loadSchemaDocument(ctx, contract.SchemaPath)
 	if err != nil {
 		return nil, err
 	}
 	resourceURL := fsSchemaResourceURL(schemaName)
-	compiler := newSchemaCompiler(fsSchemaLoader{fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
+	compiler := newSchemaCompiler(fsSchemaLoader{ctx: ctx, fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
-	if err := compiler.AddResource(resourceURL, schemaDocument); err != nil {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if err := compiler.AddResource(resourceURL.String(), schemaDocument); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
 		return nil, fmt.Errorf("failed to register JSON schema %q: %w", schemaName, err)
 	}
-	schema, err := compiler.Compile(resourceURL)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	schema, err := compileJSONSchema(ctx, resourceURL.String(), compiler.Compile)
 	if err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", schemaName, err)
 	}
@@ -140,7 +164,7 @@ func (v *FSValidator) PrepareValidation(ctx context.Context, contract domain.Out
 	return prepared, nil
 }
 
-type schemaValidatorFunc func(instance any, schemaPath string) ([]string, error)
+type schemaValidatorFunc func(ctx context.Context, instance any, schemaPath string) ([]string, error)
 
 func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract domain.OutputContract, validateSchema schemaValidatorFunc) (domain.ValidationResult, error) {
 	select {
@@ -152,7 +176,6 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 	res := domain.ValidationResult{
 		Mode:       contract.ValidationMode,
 		SchemaPath: contract.SchemaPath,
-		RepairAttempts: contract.RepairAttempts,
 	}
 
 	if artifact == nil {
@@ -165,7 +188,11 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 		res.IsValid = true
 		return res, nil
 	case domain.ValidationBasic:
-		if strings.TrimSpace(string(artifact.Body)) == "" {
+		empty := strings.TrimSpace(string(artifact.Body)) == ""
+		if err := ctx.Err(); err != nil {
+			return domain.ValidationResult{}, err
+		}
+		if empty {
 			res.Status = domain.ValidationFailed
 			res.IsValid = false
 			res.Errors = []string{"output is empty"}
@@ -175,26 +202,32 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 		res.IsValid = true
 		return res, nil
 	case domain.ValidationJSON:
-		_, jsonErr := parseJSON(artifact.Body)
+		valid := json.Valid(artifact.Body)
-		if jsonErr != nil {
+		if err := ctx.Err(); err != nil {
+			return domain.ValidationResult{}, err
+		}
+		if !valid {
 			res.Status = domain.ValidationFailed
 			res.IsValid = false
-			res.Errors = []string{fmt.Sprintf("invalid JSON: %v", jsonErr)}
+			res.Errors = []string{"invalid JSON"}
 			return res, nil
 		}
 		res.Status = domain.ValidationPassed
 		res.IsValid = true
 		return res, nil
 	case domain.ValidationJSONSchema:
-		instance, jsonErr := parseJSON(artifact.Body)
+		instance, jsonErr := decodeJSONValue(ctx, artifact.Body)
 		if jsonErr != nil {
+			if contextErr := ctx.Err(); contextErr != nil {
+				return domain.ValidationResult{}, contextErr
+			}
 			res.Status = domain.ValidationFailed
 			res.IsValid = false
 			res.Errors = []string{fmt.Sprintf("invalid JSON: %v", jsonErr)}
 			return res, nil
 		}
 
-		validationErrors, err := validateSchema(instance, contract.SchemaPath)
+		validationErrors, err := validateSchema(ctx, instance, contract.SchemaPath)
 		if err != nil {
 			return domain.ValidationResult{}, err
 		}
@@ -213,8 +246,8 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 	}
 }
 
-func (v *StandardValidator) validateJSONSchema(instance any, schemaPath string) ([]string, error) {
+func (v *StandardValidator) validateJSONSchema(ctx context.Context, instance any, schemaPath string) ([]string, error) {
-	resolvedSchemaPath, err := v.resolveSchemaPath(schemaPath)
+	resolvedSchemaPath, err := v.resolveSchemaPath(ctx, schemaPath)
 	if err != nil {
 		return nil, err
 	}
@@ -223,20 +256,21 @@ func (v *StandardValidator) validateJSONSchema(instance any, schemaPath string)
 	if err != nil {
 		return nil, err
 	}
-	compiler := newSchemaCompiler(standardSchemaLoader{root: schemaRoot})
+	if err := ctx.Err(); err != nil {
-	schema, err := compiler.Compile(resolvedSchemaPath)
+		return nil, err
+	}
+	compiler := newSchemaCompiler(standardSchemaLoader{ctx: ctx, root: schemaRoot})
+	resourceURL := fileSchemaResourceURL(resolvedSchemaPath)
+	schema, err := compileJSONSchema(ctx, resourceURL.String(), compiler.Compile)
 	if err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", resolvedSchemaPath, err)
 	}
 
-	if err := schema.Validate(instance); err != nil {
+	return executeJSONSchema(ctx, schema, instance)
-		return []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
 }
 
-func (v *FSValidator) validateJSONSchema(instance any, schemaPath string) ([]string, error) {
+func (v *FSValidator) validateJSONSchema(ctx context.Context, instance any, schemaPath string) ([]string, error) {
-	schemaName, schemaDoc, err := v.loadSchemaDocument(schemaPath)
+	schemaName, schemaDoc, err := v.loadSchemaDocument(ctx, schemaPath)
 	if err != nil {
 		return nil, err
 	}
@@ -245,71 +279,60 @@ func (v *FSValidator) validateJSONSchema(instance any, schemaPath string) ([]str
 	if err := validateSchemaDialect(schemaDoc); err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", schemaName, err)
 	}
-	compiler := newSchemaCompiler(fsSchemaLoader{fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
+	compiler := newSchemaCompiler(fsSchemaLoader{ctx: ctx, fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
-	if err := compiler.AddResource(resourceURL, schemaDoc); err != nil {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if err := compiler.AddResource(resourceURL.String(), schemaDoc); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
 		return nil, fmt.Errorf("failed to register JSON schema %q: %w", schemaName, err)
 	}
-	schema, err := compiler.Compile(resourceURL)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	schema, err := compileJSONSchema(ctx, resourceURL.String(), compiler.Compile)
 	if err != nil {
 		return nil, fmt.Errorf("failed to compile JSON schema %q: %w", schemaName, err)
 	}
 
-	if err := schema.Validate(instance); err != nil {
+	return executeJSONSchema(ctx, schema, instance)
-		return []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
 }
 
-func parseJSON(body []byte) (any, error) {
+func decodeJSONValue(ctx context.Context, body []byte) (any, error) {
-	var v any
+	if err := ctx.Err(); err != nil {
-	if err := json.Unmarshal(body, &v); err != nil {
 		return nil, err
 	}
-	return v, nil
+	decoder := json.NewDecoder(bytes.NewReader(body))
-}
+	decoder.UseNumber()
 
-func (v *StandardValidator) LoadSchemaDocument(ctx context.Context, schemaPath string) (any, error) {
+	var value any
-	select {
+	decodeErr := decoder.Decode(&value)
-	case <-ctx.Done():
+	if err := ctx.Err(); err != nil {
-		return nil, ctx.Err()
-	default:
-	}
-
-	resolved, err := v.resolveSchemaPath(schemaPath)
-	if err != nil {
 		return nil, err
 	}
-
+	if decodeErr != nil {
-	raw, err := os.ReadFile(resolved)
+		return nil, decodeErr
-	if err != nil {
-		return nil, fmt.Errorf("failed to read schema file %q: %w", resolved, err)
 	}
 
-	var doc any
+	var trailing any
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	trailingErr := decoder.Decode(&trailing)
-		return nil, fmt.Errorf("failed to decode JSON schema %q: %w", resolved, err)
+	if err := ctx.Err(); err != nil {
-	}
-	if err := validateSchemaDialect(doc); err != nil {
-		return nil, fmt.Errorf("failed to decode JSON schema %q: %w", resolved, err)
-	}
-	return doc, nil
-}
-
-func (v *FSValidator) LoadSchemaDocument(ctx context.Context, schemaPath string) (any, error) {
-	select {
-	case <-ctx.Done():
-		return nil, ctx.Err()
-	default:
-	}
-
-	_, doc, err := v.loadSchemaDocument(schemaPath)
-	if err != nil {
 		return nil, err
 	}
-	return doc, nil
+	if errors.Is(trailingErr, io.EOF) {
+		return value, nil
+	} else if trailingErr != nil {
+		return nil, trailingErr
+	}
+	return nil, errors.New("multiple JSON values")
 }
 
-func (v *StandardValidator) resolveSchemaPath(schemaPath string) (string, error) {
+func (v *StandardValidator) resolveSchemaPath(ctx context.Context, schemaPath string) (string, error) {
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	if strings.TrimSpace(schemaPath) == "" {
 		return "", errors.New("schema path is required for json_schema validation")
 	}
@@ -318,13 +341,25 @@ func (v *StandardValidator) resolveSchemaPath(schemaPath string) (string, error)
 	if err != nil {
 		return "", err
 	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	resolved, err := containedFilesystemPath(root, schemaPath)
 	if err != nil {
 		return "", err
 	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	if _, err := os.Stat(resolved); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return "", contextErr
+		}
 		return "", fmt.Errorf("failed to access schema file %q: %w", resolved, err)
 	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 
 	return resolved, nil
 }
@@ -345,28 +380,39 @@ func (v *StandardValidator) schemaRoot() (string, error) {
 	return resolved, nil
 }
 
-func (v *FSValidator) loadSchemaDocument(schemaPath string) (string, any, error) {
+func (v *FSValidator) loadSchemaDocument(ctx context.Context, schemaPath string) (string, any, error) {
-	resolved, err := v.resolveSchemaPath(schemaPath)
+	resolved, err := v.resolveSchemaPath(ctx, schemaPath)
 	if err != nil {
 		return "", nil, err
 	}
 
-	raw, err := fs.ReadFile(v.fsys, resolved)
+	raw, err := readSchemaFile(ctx, func() (fs.File, error) {
+		return v.fsys.Open(resolved)
+	})
 	if err != nil {
 		return "", nil, fmt.Errorf("failed to read schema file %q: %w", resolved, err)
 	}
 
-	var doc any
+	doc, err := decodeJSONValue(ctx, raw)
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	if err != nil {
 		return "", nil, fmt.Errorf("failed to decode JSON schema %q: %w", resolved, err)
 	}
 	if err := validateSchemaDialect(doc); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return "", nil, contextErr
+		}
 		return "", nil, fmt.Errorf("failed to decode JSON schema %q: %w", resolved, err)
 	}
+	if err := ctx.Err(); err != nil {
+		return "", nil, err
+	}
 	return resolved, doc, nil
 }
 
-func (v *FSValidator) resolveSchemaPath(schemaPath string) (string, error) {
+func (v *FSValidator) resolveSchemaPath(ctx context.Context, schemaPath string) (string, error) {
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	if strings.TrimSpace(schemaPath) == "" {
 		return "", errors.New("schema path is required for json_schema validation")
 	}
@@ -377,8 +423,14 @@ func (v *FSValidator) resolveSchemaPath(schemaPath string) (string, error) {
 	cleanRoot := filecatalog.CleanFSRoot(v.root)
 	rootInfo, err := fs.Stat(v.fsys, cleanRoot)
 	if err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return "", contextErr
+		}
 		return "", fmt.Errorf("failed to access schema source %q: %w", cleanRoot, err)
 	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 
 	var resolved string
 	if rootInfo.IsDir() {
@@ -392,15 +444,21 @@ func (v *FSValidator) resolveSchemaPath(schemaPath string) (string, error) {
 		if err != nil {
 			return "", err
 		}
-		if cleanSchemaPath != path.Base(cleanRoot) {
+		if cleanSchemaPath != strings.TrimSpace(path.Base(cleanRoot)) {
 			return "", fmt.Errorf("schema path %q does not match schema file %q", cleanSchemaPath, path.Base(cleanRoot))
 		}
 		resolved = cleanRoot
 	}
 
 	if _, err := fs.Stat(v.fsys, resolved); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return "", contextErr
+		}
 		return "", fmt.Errorf("failed to access schema file %q: %w", resolved, err)
 	}
+	if err := ctx.Err(); err != nil {
+		return "", err
+	}
 	return resolved, nil
 }
 
@@ -416,8 +474,19 @@ func cleanSchemaFSPath(schemaPath string) (string, error) {
 	return cleaned, nil
 }
 
-func fsSchemaResourceURL(schemaName string) string {
+func fileSchemaResourceURL(schemaName string) *url.URL {
-	return "promptkit-schema:///" + strings.TrimPrefix(path.Clean(schemaName), "/")
+	filePath := filepath.ToSlash(schemaName)
+	if runtime.GOOS == "windows" && !strings.HasPrefix(filePath, "/") {
+		filePath = "/" + filePath
+	}
+	return &url.URL{Scheme: "file", Path: filePath}
+}
+
+func fsSchemaResourceURL(schemaName string) *url.URL {
+	return &url.URL{
+		Scheme: "promptkit-schema",
+		Path:   "/" + strings.TrimPrefix(path.Clean(schemaName), "/"),
+	}
 }
 
 func newSchemaCompiler(loader jsonschema.URLLoader) *jsonschema.Compiler {
@@ -427,6 +496,35 @@ func newSchemaCompiler(loader jsonschema.URLLoader) *jsonschema.Compiler {
 	return compiler
 }
 
+type schemaExecutor interface {
+	Validate(instance any) error
+}
+
+func compileJSONSchema(ctx context.Context, resourceURL string, compile func(string) (*jsonschema.Schema, error)) (*jsonschema.Schema, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	schema, compileErr := compile(resourceURL)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	return schema, compileErr
+}
+
+func executeJSONSchema(ctx context.Context, schema schemaExecutor, instance any) ([]string, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	validationErr := schema.Validate(instance)
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if validationErr != nil {
+		return []string{fmt.Sprintf("json schema validation failed: %v", validationErr)}, nil
+	}
+	return nil, nil
+}
+
 func validateSchemaDialect(doc any) error {
 	object, ok := doc.(map[string]any)
 	if !ok {
@@ -447,19 +545,33 @@ func validateSchemaDialect(doc any) error {
 }
 
 type standardSchemaLoader struct {
+	ctx  context.Context
 	root string
 }
 
 func (l standardSchemaLoader) Load(resourceURL string) (any, error) {
+	if err := l.ctx.Err(); err != nil {
+		return nil, err
+	}
+	parsed, err := url.Parse(resourceURL)
+	if err != nil || parsed.Scheme != "file" || parsed.Host != "" || parsed.RawQuery != "" || parsed.Opaque != "" {
+		return nil, fmt.Errorf("schema reference %q is not a contained file reference", resourceURL)
+	}
 	fileName, err := (jsonschema.FileLoader{}).ToFile(resourceURL)
 	if err != nil {
 		return nil, fmt.Errorf("schema reference %q is not a contained file reference: %w", resourceURL, err)
 	}
 	resolved, err := containedFilesystemPath(l.root, fileName)
 	if err != nil {
+		if contextErr := l.ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
 		return nil, err
 	}
-	return loadJSONSchemaFile(resolved)
+	if err := l.ctx.Err(); err != nil {
+		return nil, err
+	}
+	return loadJSONSchemaFile(l.ctx, resolved)
 }
 
 func containedFilesystemPath(root, name string) (string, error) {
@@ -485,38 +597,94 @@ func containedFilesystemPath(root, name string) (string, error) {
 	return candidate, nil
 }
 
-func loadJSONSchemaFile(name string) (any, error) {
+func loadJSONSchemaFile(ctx context.Context, name string) (any, error) {
-	raw, err := os.ReadFile(name)
+	raw, err := readSchemaFile(ctx, func() (fs.File, error) {
+		return os.Open(name)
+	})
 	if err != nil {
 		return nil, err
 	}
-	var doc any
+	doc, err := decodeJSONValue(ctx, raw)
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	if err != nil {
 		return nil, err
 	}
 	if err := validateSchemaDialect(doc); err != nil {
+		if contextErr := ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
+		return nil, err
+	}
+	if err := ctx.Err(); err != nil {
 		return nil, err
 	}
 	return doc, nil
 }
 
+func readSchemaFile(ctx context.Context, open func() (fs.File, error)) ([]byte, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	file, openErr := open()
+	if err := ctx.Err(); err != nil {
+		if file != nil {
+			_ = file.Close()
+		}
+		return nil, err
+	}
+	if openErr != nil {
+		if file != nil {
+			_ = file.Close()
+		}
+		return nil, openErr
+	}
+	if file == nil {
+		return nil, errors.New("schema source returned a nil file")
+	}
+	defer file.Close()
+
+	var contents []byte
+	chunk := make([]byte, schemaReadChunkSize)
+	for {
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		n, readErr := file.Read(chunk)
+		if n > 0 {
+			contents = append(contents, chunk[:n]...)
+		}
+		if err := ctx.Err(); err != nil {
+			return nil, err
+		}
+		if errors.Is(readErr, io.EOF) {
+			return contents, nil
+		}
+		if readErr != nil {
+			return nil, readErr
+		}
+		if n == 0 {
+			return nil, io.ErrNoProgress
+		}
+	}
+}
+
 type fsSchemaLoader struct {
+	ctx  context.Context
 	fsys fs.FS
 	root string
 }
 
 func (l fsSchemaLoader) Load(resourceURL string) (any, error) {
+	if err := l.ctx.Err(); err != nil {
+		return nil, err
+	}
 	parsed, err := url.Parse(resourceURL)
 	if err != nil {
 		return nil, fmt.Errorf("invalid schema reference %q: %w", resourceURL, err)
 	}
-	if parsed.Scheme != "promptkit-schema" || parsed.Host != "" {
+	if parsed.Scheme != "promptkit-schema" || parsed.Host != "" || parsed.RawQuery != "" || parsed.Opaque != "" {
 		return nil, fmt.Errorf("schema reference %q is not allowed", resourceURL)
 	}
-	name, err := url.PathUnescape(strings.TrimPrefix(parsed.Path, "/"))
+	name := strings.TrimPrefix(parsed.Path, "/")
-	if err != nil {
-		return nil, fmt.Errorf("invalid schema reference %q: %w", resourceURL, err)
-	}
 	name = path.Clean(name)
 	if l.root == "." {
 		if strings.HasPrefix(name, "../") || name == ".." {
@@ -528,21 +696,35 @@ func (l fsSchemaLoader) Load(resourceURL string) (any, error) {
 
 	rootInfo, err := fs.Stat(l.fsys, l.root)
 	if err != nil {
+		if contextErr := l.ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
+		return nil, err
+	}
+	if err := l.ctx.Err(); err != nil {
 		return nil, err
 	}
 	if !rootInfo.IsDir() && name != l.root {
 		return nil, fmt.Errorf("schema reference %q is outside the configured schema file", resourceURL)
 	}
 
-	raw, err := fs.ReadFile(l.fsys, name)
+	raw, err := readSchemaFile(l.ctx, func() (fs.File, error) {
+		return l.fsys.Open(name)
+	})
 	if err != nil {
 		return nil, err
 	}
-	var doc any
+	doc, err := decodeJSONValue(l.ctx, raw)
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	if err != nil {
 		return nil, err
 	}
 	if err := validateSchemaDialect(doc); err != nil {
+		if contextErr := l.ctx.Err(); contextErr != nil {
+			return nil, contextErr
+		}
+		return nil, err
+	}
+	if err := l.ctx.Err(); err != nil {
 		return nil, err
 	}
 	return doc, nil

internal/validate/standard_validator_test.go

@@ -3,9 +3,11 @@ package validate
 import (
 	"context"
 	"encoding/json"
+	"net/url"
 	"os"
 	"path/filepath"
 	"reflect"
+	"runtime"
 	"strconv"
 	"strings"
 	"testing"
@@ -62,6 +64,43 @@ func TestStandardValidatorBasicFailureEmpty(t *testing.T) {
 	}
 }
 
+func TestStandardValidatorReportsZeroRepairAttempts(t *testing.T) {
+	v := NewStandardValidator("")
+
+	for _, tc := range []struct {
+		name     string
+		body     string
+		contract domain.OutputContract
+	}{
+		{
+			name: "passed basic validation",
+			body: "answer",
+			contract: domain.OutputContract{
+				ValidationMode: domain.ValidationBasic,
+				RepairAttempts: 3,
+			},
+		},
+		{
+			name: "failed JSON validation",
+			body: `{"answer":`,
+			contract: domain.OutputContract{
+				ValidationMode: domain.ValidationJSON,
+				RepairAttempts: 3,
+			},
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			res, err := v.Validate(context.Background(), &domain.Artifact{Body: []byte(tc.body)}, tc.contract)
+			if err != nil {
+				t.Fatalf("validate: %v", err)
+			}
+			if res.RepairAttempts != 0 {
+				t.Fatalf("repair attempts = %d, want 0", res.RepairAttempts)
+			}
+		})
+	}
+}
+
 func TestStandardValidatorJSONSuccess(t *testing.T) {
 	v := NewStandardValidator("")
 
@@ -93,6 +132,51 @@ func TestStandardValidatorJSONFailure(t *testing.T) {
 	}
 }
 
+func TestJSONValidationChecksCompleteSyntaxWithoutChangingArtifact(t *testing.T) {
+	tests := []struct {
+		name      string
+		body      string
+		wantValid bool
+	}{
+		{name: "ordinary object", body: `{"count":2,"ok":true}`, wantValid: true},
+		{name: "integer at exact float boundary", body: `9007199254740992`, wantValid: true},
+		{name: "integer beyond exact float boundary", body: `9007199254740993`, wantValid: true},
+		{name: "large exponent", body: `1e400`, wantValid: true},
+		{name: "precise decimal", body: `0.123456789012345678901234567890`, wantValid: true},
+		{name: "surrounding whitespace", body: " \n [1,2,3] \t", wantValid: true},
+		{name: "malformed document", body: `{"count":`, wantValid: false},
+		{name: "trailing value", body: `1 2`, wantValid: false},
+	}
+
+	validator := NewStandardValidator("")
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			body := []byte(tc.body)
+			before := append([]byte(nil), body...)
+			artifact := &domain.Artifact{Body: body}
+
+			result, err := validator.Validate(context.Background(), artifact, domain.OutputContract{
+				ValidationMode: domain.ValidationJSON,
+			})
+			if err != nil {
+				t.Fatalf("validate JSON: %v", err)
+			}
+			if result.IsValid != tc.wantValid {
+				t.Fatalf("valid = %v, want %v; result=%+v", result.IsValid, tc.wantValid, result)
+			}
+			if tc.wantValid && result.Status != domain.ValidationPassed {
+				t.Fatalf("status = %q, want %q", result.Status, domain.ValidationPassed)
+			}
+			if !tc.wantValid && result.Status != domain.ValidationFailed {
+				t.Fatalf("status = %q, want %q", result.Status, domain.ValidationFailed)
+			}
+			if !reflect.DeepEqual(artifact.Body, before) {
+				t.Fatalf("artifact body changed: got %q, want %q", artifact.Body, before)
+			}
+		})
+	}
+}
+
 func TestStandardValidatorJSONSchemaSuccess(t *testing.T) {
 	tmp := t.TempDir()
 	schemaPath := filepath.Join(tmp, "schema.json")
@@ -285,55 +369,6 @@ func TestStandardValidatorJSONSchemaCompilationError(t *testing.T) {
 	}
 }
 
-func TestStandardValidatorLoadSchemaDocumentSuccess(t *testing.T) {
-	tmp := t.TempDir()
-	if err := os.WriteFile(filepath.Join(tmp, "schema.json"), []byte(`{
-  "type": "object",
-  "properties": {
-    "name": {"type": "string"}
-  }
-}`), 0644); err != nil {
-		t.Fatal(err)
-	}
-
-	v := NewStandardValidator(tmp)
-	loader, ok := v.(SchemaDocumentLoader)
-	if !ok {
-		t.Fatal("standard validator must implement SchemaDocumentLoader")
-	}
-
-	doc, err := loader.LoadSchemaDocument(context.Background(), "schema.json")
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-
-	obj, ok := doc.(map[string]any)
-	if !ok {
-		t.Fatalf("expected object document, got %#v", doc)
-	}
-	if obj["type"] != "object" {
-		t.Fatalf("expected schema type=object, got %#v", obj["type"])
-	}
-}
-
-func TestStandardValidatorLoadSchemaDocumentInvalidJSON(t *testing.T) {
-	tmp := t.TempDir()
-	if err := os.WriteFile(filepath.Join(tmp, "schema.json"), []byte(`{`), 0644); err != nil {
-		t.Fatal(err)
-	}
-
-	v := NewStandardValidator(tmp)
-	loader, ok := v.(SchemaDocumentLoader)
-	if !ok {
-		t.Fatal("standard validator must implement SchemaDocumentLoader")
-	}
-
-	_, err := loader.LoadSchemaDocument(context.Background(), "schema.json")
-	if err == nil {
-		t.Fatal("expected decode error")
-	}
-}
-
 func TestFSValidatorJSONSchemaSuccess(t *testing.T) {
 	v := NewFSValidator(fstest.MapFS{
 		"schemas/events.schema.json": &fstest.MapFile{Data: []byte(`{
@@ -416,17 +451,70 @@ func TestFSValidatorPreparedSchemaSurvivesSourceMutation(t *testing.T) {
 	}
 }
 
-func TestFSValidatorJSONSchemaRegistrationError(t *testing.T) {
+func TestFSValidatorEscapesSchemaResourcePath(t *testing.T) {
+	for _, schemaName := range []string{"%zz.json", "space name.json", "hash#.json", "query?.json", "rún.json"} {
+		t.Run(schemaName, func(t *testing.T) {
 			v := NewFSValidator(fstest.MapFS{
-		"schemas/%zz.json": &fstest.MapFile{Data: []byte(`{"type":"object"}`)},
+				"schemas/" + schemaName: &fstest.MapFile{Data: []byte(`{"type":"object"}`)},
 			}, "schemas")
 
 			res, err := v.Validate(context.Background(), &domain.Artifact{Body: []byte(`{}`)}, domain.OutputContract{
 				ValidationMode: domain.ValidationJSONSchema,
-		SchemaPath:     "%zz.json",
+				SchemaPath:     schemaName,
 			})
-	if err == nil || !strings.Contains(err.Error(), "failed to register JSON schema") {
+			if err != nil || !res.IsValid {
-		t.Fatalf("expected schema registration error, got result=%#v error=%v", res, err)
+				t.Fatalf("validate schema %q: result=%#v error=%v", schemaName, res, err)
+			}
+		})
+	}
+}
+
+func TestSchemaReferencesPreserveEscapedFilenames(t *testing.T) {
+	for _, name := range []string{"%2F.json", "space name.json", "hash#.json", "query?.json", "rún.json"} {
+		t.Run(name, func(t *testing.T) {
+			rootName := "root-" + name
+			childName := "child-" + name
+			childPath := "nested/" + childName
+			reference := (&url.URL{Path: childPath}).EscapedPath()
+			rootSchema := []byte(`{"$ref":` + strconv.Quote(reference) + `}`)
+			childSchema := []byte(`{"type":"integer","minimum":2}`)
+
+			t.Run("fs.FS", func(t *testing.T) {
+				validator := NewFSValidator(fstest.MapFS{
+					"schemas/" + rootName:  &fstest.MapFile{Data: rootSchema},
+					"schemas/" + childPath: &fstest.MapFile{Data: childSchema},
+				}, "schemas")
+				assertSchemaValidation(t, validator, rootName)
+			})
+
+			t.Run("operating system files", func(t *testing.T) {
+				if runtime.GOOS == "windows" && strings.ContainsAny(rootName+childName, `<>:"/\|?*`) {
+					t.Skip("filename is not legal on Windows")
+				}
+				root := t.TempDir()
+				if err := os.Mkdir(filepath.Join(root, "nested"), 0o755); err != nil {
+					t.Fatal(err)
+				}
+				if err := os.WriteFile(filepath.Join(root, rootName), rootSchema, 0o644); err != nil {
+					t.Fatal(err)
+				}
+				if err := os.WriteFile(filepath.Join(root, filepath.FromSlash(childPath)), childSchema, 0o644); err != nil {
+					t.Fatal(err)
+				}
+				assertSchemaValidation(t, NewStandardValidator(root), rootName)
+			})
+		})
+	}
+}
+
+func assertSchemaValidation(t *testing.T, validator Validator, schemaPath string) {
+	t.Helper()
+	result, err := validator.Validate(context.Background(), &domain.Artifact{Body: []byte(`2`)}, domain.OutputContract{
+		ValidationMode: domain.ValidationJSONSchema,
+		SchemaPath:     schemaPath,
+	})
+	if err != nil || !result.IsValid {
+		t.Fatalf("validate schema %q: result=%#v error=%v", schemaPath, result, err)
 	}
 }
 
@@ -516,25 +604,6 @@ func TestFSValidatorSingleSchemaFileUsesBaseName(t *testing.T) {
 	}
 }
 
-func TestFSValidatorLoadSchemaDocument(t *testing.T) {
-	v := NewFSValidator(fstest.MapFS{
-		"schemas/schema.json": &fstest.MapFile{Data: []byte(`{"type":"object"}`)},
-	}, "schemas")
-	loader, ok := v.(SchemaDocumentLoader)
-	if !ok {
-		t.Fatal("fs validator must implement SchemaDocumentLoader")
-	}
-
-	doc, err := loader.LoadSchemaDocument(context.Background(), "schema.json")
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-	obj, ok := doc.(map[string]any)
-	if !ok || obj["type"] != "object" {
-		t.Fatalf("unexpected schema document: %#v", doc)
-	}
-}
-
 func TestStandardValidatorJSONSchemaReferenceBoundaries(t *testing.T) {
 	root := t.TempDir()
 	if err := os.WriteFile(filepath.Join(root, "child.json"), []byte(`{
@@ -629,6 +698,225 @@ func TestFSValidatorJSONSchemaReferenceBoundaries(t *testing.T) {
 	}
 }
 
+func TestJSONSchemaNumericConstraintsRetainJSONPrecision(t *testing.T) {
+	tests := []struct {
+		name      string
+		schema    string
+		instance  string
+		wantValid bool
+	}{
+		{
+			name:      "const distinguishes adjacent large integers",
+			schema:    `{"const":9007199254740993}`,
+			instance:  `9007199254740993`,
+			wantValid: true,
+		},
+		{
+			name:      "const rejects adjacent large integer",
+			schema:    `{"const":9007199254740993}`,
+			instance:  `9007199254740992`,
+			wantValid: false,
+		},
+		{
+			name:      "const accepts exponent beyond float range",
+			schema:    `{"const":1e400}`,
+			instance:  `1e400`,
+			wantValid: true,
+		},
+		{
+			name:      "minimum accepts precise decimal boundary",
+			schema:    `{"type":"number","minimum":0.123456789012345678901234567890}`,
+			instance:  `0.123456789012345678901234567890`,
+			wantValid: true,
+		},
+		{
+			name:      "minimum rejects lower precise decimal",
+			schema:    `{"type":"number","minimum":0.123456789012345678901234567890}`,
+			instance:  `0.123456789012345678901234567889`,
+			wantValid: false,
+		},
+		{
+			name:      "maximum distinguishes adjacent large integers",
+			schema:    `{"type":"number","maximum":9007199254740992}`,
+			instance:  `9007199254740993`,
+			wantValid: false,
+		},
+		{
+			name:      "multiple of accepts exact decimal multiple",
+			schema:    `{"type":"number","multipleOf":0.0000000000000000001}`,
+			instance:  `0.0000000000000000003`,
+			wantValid: true,
+		},
+		{
+			name:      "multiple of rejects inexact decimal multiple",
+			schema:    `{"type":"number","multipleOf":0.0000000000000000001}`,
+			instance:  `0.00000000000000000031`,
+			wantValid: false,
+		},
+		{
+			name:      "ordinary number remains supported",
+			schema:    `{"type":"number","minimum":1,"maximum":3}`,
+			instance:  `2`,
+			wantValid: true,
+		},
+	}
+
+	for _, source := range jsonSchemaValidatorSources() {
+		t.Run(source.name, func(t *testing.T) {
+			for _, tc := range tests {
+				t.Run(tc.name, func(t *testing.T) {
+					validator := source.new(t, []byte(tc.schema))
+					result, err := validator.Validate(context.Background(), &domain.Artifact{Body: []byte(tc.instance)}, domain.OutputContract{
+						ValidationMode: domain.ValidationJSONSchema,
+						SchemaPath:     "schema.json",
+					})
+					if err != nil {
+						t.Fatalf("validate JSON Schema instance: %v", err)
+					}
+					if result.IsValid != tc.wantValid {
+						t.Fatalf("valid = %v, want %v; result=%+v", result.IsValid, tc.wantValid, result)
+					}
+				})
+			}
+		})
+	}
+}
+
+func TestJSONSchemaDecodingRequiresOneCompleteDocument(t *testing.T) {
+	for _, source := range jsonSchemaValidatorSources() {
+		t.Run(source.name, func(t *testing.T) {
+			for _, schema := range []string{`{"type":`, `{} {}`} {
+				validator := source.new(t, []byte(schema))
+				_, err := validator.Validate(context.Background(), &domain.Artifact{Body: []byte(`1`)}, domain.OutputContract{
+					ValidationMode: domain.ValidationJSONSchema,
+					SchemaPath:     "schema.json",
+				})
+				if err == nil {
+					t.Fatalf("schema %q: expected decoding error", schema)
+				}
+			}
+
+			validator := source.new(t, []byte(`{}`))
+			for _, instance := range []string{`{"value":`, `1 2`} {
+				result, err := validator.Validate(context.Background(), &domain.Artifact{Body: []byte(instance)}, domain.OutputContract{
+					ValidationMode: domain.ValidationJSONSchema,
+					SchemaPath:     "schema.json",
+				})
+				if err != nil {
+					t.Fatalf("instance %q: expected completed validation, got %v", instance, err)
+				}
+				if result.Status != domain.ValidationFailed || result.IsValid || len(result.Errors) == 0 {
+					t.Fatalf("instance %q: expected failed validation, got %+v", instance, result)
+				}
+			}
+		})
+	}
+}
+
+func TestPreparedSchemaDocumentsRetainExactNumbers(t *testing.T) {
+	const schema = `{
+  "const": 9007199254740993,
+  "minimum": 0.123456789012345678901234567890,
+  "maximum": 1e400,
+  "multipleOf": 0.0000000000000000001
+}`
+	want := map[string]string{
+		"const":      "9007199254740993",
+		"minimum":    "0.123456789012345678901234567890",
+		"maximum":    "1e400",
+		"multipleOf": "0.0000000000000000001",
+	}
+
+	for _, source := range jsonSchemaValidatorSources() {
+		t.Run(source.name, func(t *testing.T) {
+			validator := source.new(t, []byte(schema))
+			preparer, ok := validator.(ValidationPreparer)
+			if !ok {
+				t.Fatal("validator does not support validation preparation")
+			}
+			prepared, err := preparer.PrepareValidation(context.Background(), domain.OutputContract{
+				ValidationMode: domain.ValidationJSONSchema,
+				SchemaPath:     "schema.json",
+			})
+			if err != nil {
+				t.Fatalf("prepare validation: %v", err)
+			}
+			document, ok := prepared.SchemaDocument().(map[string]any)
+			if !ok {
+				t.Fatalf("schema document = %#v, want object", prepared.SchemaDocument())
+			}
+			for name, wantNumber := range want {
+				got, ok := document[name].(json.Number)
+				if !ok {
+					t.Fatalf("schema field %q = %#v, want json.Number", name, document[name])
+				}
+				if got.String() != wantNumber {
+					t.Fatalf("schema field %q = %q, want %q", name, got, wantNumber)
+				}
+			}
+		})
+	}
+}
+
+type jsonSchemaValidatorSource struct {
+	name string
+	new  func(*testing.T, []byte) Validator
+}
+
+func jsonSchemaValidatorSources() []jsonSchemaValidatorSource {
+	return []jsonSchemaValidatorSource{
+		{
+			name: "operating system files",
+			new: func(t *testing.T, schema []byte) Validator {
+				t.Helper()
+				root := t.TempDir()
+				if err := os.WriteFile(filepath.Join(root, "schema.json"), schema, 0o644); err != nil {
+					t.Fatalf("write schema: %v", err)
+				}
+				return NewStandardValidator(root)
+			},
+		},
+		{
+			name: "fs.FS",
+			new: func(t *testing.T, schema []byte) Validator {
+				t.Helper()
+				return NewFSValidator(fstest.MapFS{
+					"schema.json": &fstest.MapFile{Data: schema},
+				}, ".")
+			},
+		},
+	}
+}
+
+func BenchmarkJSONValidation(b *testing.B) {
+	largeArray := []byte(`[` + strings.Repeat(`12345678901234567890,`, 32*1024) + `0]`)
+	benchmarks := []struct {
+		name string
+		body []byte
+	}{
+		{name: "scalar", body: []byte(`1e400`)},
+		{name: "object", body: []byte(`{"name":"eris","count":9007199254740993,"enabled":true}`)},
+		{name: "large array", body: largeArray},
+	}
+	validator := NewStandardValidator("")
+	contract := domain.OutputContract{ValidationMode: domain.ValidationJSON}
+
+	for _, benchmark := range benchmarks {
+		b.Run(benchmark.name, func(b *testing.B) {
+			artifact := &domain.Artifact{Body: benchmark.body}
+			b.ReportAllocs()
+			b.SetBytes(int64(len(benchmark.body)))
+			b.ResetTimer()
+			for range b.N {
+				result, err := validator.Validate(context.Background(), artifact, contract)
+				if err != nil || !result.IsValid {
+					b.Fatalf("validate JSON: result=%+v error=%v", result, err)
+				}
+			}
+		})
+	}
+}
+
 func TestJSONSchemaDialectIsDraft2020(t *testing.T) {
 	tests := []struct {
 		name      string
@@ -673,8 +961,8 @@ func TestJSONSchemaDialectIsDraft2020(t *testing.T) {
 func assertSchemaDocument(t *testing.T, got any, expectedJSON []byte) {
 	t.Helper()
 
-	var expected any
+	expected, err := decodeJSONValue(context.Background(), expectedJSON)
-	if err := json.Unmarshal(expectedJSON, &expected); err != nil {
+	if err != nil {
 		t.Fatalf("decode expected schema document: %v", err)
 	}
 	if !reflect.DeepEqual(got, expected) {

internal/validate/validator.go

@@ -7,11 +7,16 @@ import (
 )
 
 // Validator validates the generated artifact based on the output contract.
+// Validation checks ctx around Promptkit-controlled work and synchronous
+// dependency calls. A dependency call already in progress cannot be preempted;
+// after it returns, cancellation takes precedence over its result.
 type Validator interface {
 	Validate(ctx context.Context, artifact *domain.Artifact, contract domain.OutputContract) (domain.ValidationResult, error)
 }
 
 // PreparedValidation validates artifacts against one frozen output contract.
+// Its cancellation boundary is synchronous: Validate does not detach schema
+// execution, and an observed context error prevents publication of a result.
 type PreparedValidation interface {
 	Validate(ctx context.Context, artifact *domain.Artifact) (domain.ValidationResult, error)
 	// SchemaDocument returns the root JSON Schema document used for provider
@@ -21,11 +26,9 @@ type PreparedValidation interface {
 }
 
 // ValidationPreparer freezes validation resources for one output contract.
+// Preparation reads schemas in context-checked chunks and checks ctx around
+// decoding and compilation. Filesystem and compiler calls remain synchronous,
+// so cancellation becomes authoritative when an in-progress call returns.
 type ValidationPreparer interface {
 	PrepareValidation(ctx context.Context, contract domain.OutputContract) (PreparedValidation, error)
 }
-
-// SchemaDocumentLoader loads JSON schema documents using validator path semantics.
-type SchemaDocumentLoader interface {
-	LoadSchemaDocument(ctx context.Context, schemaPath string) (any, error)
-}

json.go

@@ -2,9 +2,16 @@ package promptkit
 
 import (
 	"encoding/json"
+	"fmt"
+	"math"
 	"time"
 )
 
+const (
+	minDurationMilliseconds = int64(time.Duration(math.MinInt64) / time.Millisecond)
+	maxDurationMilliseconds = int64(time.Duration(math.MaxInt64) / time.Millisecond)
+)
+
 // MarshalJSON implements json.Marshaler for PreparedRun. It uses RFC 3339
 // timestamps, integer duration_ms, and omits zero timing values.
 func (r PreparedRun) MarshalJSON() ([]byte, error) {
@@ -21,35 +28,8 @@ func (r PreparedRun) MarshalJSON() ([]byte, error) {
 		durationMS = &r.DurationMS
 	}
 
-	return json.Marshal(struct {
+	return json.Marshal(preparedRunJSON{
-		PromptID             string                `json:"prompt_id"`
+		preparedRunJSONFields: preparedRunJSONFields(r),
-		PromptVersion        string                `json:"prompt_version,omitempty"`
-		PromptHash           string                `json:"prompt_hash,omitempty"`
-		SelectedProfileID    string                `json:"selected_profile_id"`
-		SelectedBackendID    string                `json:"selected_backend_id,omitempty"`
-		EffectiveModelParams ExecutionTarget       `json:"effective_model_params"`
-		OutputContract       OutputContract        `json:"output_contract"`
-		StructuredOutput     *StructuredOutputSpec `json:"structured_output,omitempty"`
-		InputHashes          map[string]string     `json:"input_hashes,omitempty"`
-		SessionID            string                `json:"session_id,omitempty"`
-		RenderedPromptHash   string                `json:"rendered_prompt_hash"`
-		Messages             []RenderedMessage     `json:"messages"`
-		StartTime            *time.Time            `json:"start_time,omitempty"`
-		EndTime              *time.Time            `json:"end_time,omitempty"`
-		DurationMS           *int64                `json:"duration_ms,omitempty"`
-	}{
-		PromptID:             r.PromptID,
-		PromptVersion:        r.PromptVersion,
-		PromptHash:           r.PromptHash,
-		SelectedProfileID:    r.SelectedProfileID,
-		SelectedBackendID:    r.SelectedBackendID,
-		EffectiveModelParams: r.EffectiveModelParams,
-		OutputContract:       r.OutputContract,
-		StructuredOutput:     r.StructuredOutput,
-		InputHashes:          r.InputHashes,
-		SessionID:            r.SessionID,
-		RenderedPromptHash:   r.RenderedPromptHash,
-		Messages:             r.Messages,
 		StartTime:             startTime,
 		EndTime:               endTime,
 		DurationMS:            durationMS,
@@ -74,22 +54,7 @@ func (r RunResult) MarshalJSON() ([]byte, error) {
 	}
 
 	return json.Marshal(runResultJSON{
-		RunID:                r.RunID,
+		runResultJSONFields: runResultJSONFields(r),
-		Artifact:             r.Artifact,
-		RawOutput:            r.RawOutput,
-		Validation:           r.Validation,
-		PromptID:             r.PromptID,
-		PromptVersion:        r.PromptVersion,
-		PromptHash:           r.PromptHash,
-		SessionID:            r.SessionID,
-		RenderedPromptHash:   r.RenderedPromptHash,
-		SelectedProfileID:    r.SelectedProfileID,
-		SelectedBackendID:    r.SelectedBackendID,
-		ModelName:            r.ModelName,
-		Endpoint:             r.Endpoint,
-		EffectiveModelParams: r.EffectiveModelParams,
-		InputHashes:          r.InputHashes,
-		Usage:                r.Usage,
 		StartTime:           startTime,
 		EndTime:             endTime,
 		DurationMS:          durationMS,
@@ -97,66 +62,49 @@ func (r RunResult) MarshalJSON() ([]byte, error) {
 }
 
 // UnmarshalJSON implements json.Unmarshaler for RunResult. It decodes
-// duration_ms into Duration with millisecond precision.
+// duration_ms into Duration with millisecond precision. A duration_ms outside
+// the range representable by time.Duration returns an error without changing
+// the receiver.
 func (r *RunResult) UnmarshalJSON(data []byte) error {
 	var wire runResultJSON
 	if err := json.Unmarshal(data, &wire); err != nil {
 		return err
 	}
 
-	*r = RunResult{
+	result := RunResult(wire.runResultJSONFields)
-		RunID:                wire.RunID,
+	if wire.DurationMS != nil {
-		Artifact:             wire.Artifact,
+		if *wire.DurationMS < minDurationMilliseconds || *wire.DurationMS > maxDurationMilliseconds {
-		RawOutput:            wire.RawOutput,
+			return fmt.Errorf(
-		Validation:           wire.Validation,
+				"decode RunResult duration_ms: %d cannot be represented as time.Duration",
-		PromptID:             wire.PromptID,
+				*wire.DurationMS,
-		PromptVersion:        wire.PromptVersion,
+			)
-		PromptHash:           wire.PromptHash,
+		}
-		SessionID:            wire.SessionID,
+		result.Duration = time.Duration(*wire.DurationMS) * time.Millisecond
-		RenderedPromptHash:   wire.RenderedPromptHash,
-		SelectedProfileID:    wire.SelectedProfileID,
-		SelectedBackendID:    wire.SelectedBackendID,
-		ModelName:            wire.ModelName,
-		Endpoint:             wire.Endpoint,
-		EffectiveModelParams: wire.EffectiveModelParams,
-		InputHashes:          wire.InputHashes,
-		Usage:                wire.Usage,
-		Duration:             time.Duration(valueOrZero(wire.DurationMS)) * time.Millisecond,
 	}
 	if wire.StartTime != nil {
-		r.StartTime = *wire.StartTime
+		result.StartTime = *wire.StartTime
 	}
 	if wire.EndTime != nil {
-		r.EndTime = *wire.EndTime
+		result.EndTime = *wire.EndTime
 	}
+	*r = result
 	return nil
 }
 
-type runResultJSON struct {
+type preparedRunJSONFields PreparedRun
-	RunID                string            `json:"run_id"`
+
-	Artifact             Artifact          `json:"artifact"`
+type preparedRunJSON struct {
-	RawOutput            string            `json:"raw_output"`
+	preparedRunJSONFields
-	Validation           ValidationResult  `json:"validation"`
-	PromptID             string            `json:"prompt_id"`
-	PromptVersion        string            `json:"prompt_version,omitempty"`
-	PromptHash           string            `json:"prompt_hash,omitempty"`
-	SessionID            string            `json:"session_id,omitempty"`
-	RenderedPromptHash   string            `json:"rendered_prompt_hash"`
-	SelectedProfileID    string            `json:"selected_profile_id"`
-	SelectedBackendID    string            `json:"selected_backend_id,omitempty"`
-	ModelName            string            `json:"model_name"`
-	Endpoint             string            `json:"endpoint"`
-	EffectiveModelParams ExecutionTarget   `json:"effective_model_params"`
-	InputHashes          map[string]string `json:"input_hashes,omitempty"`
-	Usage                TokenUsage        `json:"usage"`
 	StartTime  *time.Time `json:"start_time,omitempty"`
 	EndTime    *time.Time `json:"end_time,omitempty"`
 	DurationMS *int64     `json:"duration_ms,omitempty"`
 }
 
-func valueOrZero(value *int64) int64 {
+type runResultJSONFields RunResult
-	if value == nil {
+
-		return 0
+type runResultJSON struct {
-	}
+	runResultJSONFields
-	return *value
+	StartTime  *time.Time `json:"start_time,omitempty"`
+	EndTime    *time.Time `json:"end_time,omitempty"`
+	DurationMS *int64     `json:"duration_ms,omitempty"`
 }

json_contract_test.go

@@ -0,0 +1,328 @@
+package promptkit_test
+
+import (
+	"encoding/json"
+	"math"
+	"reflect"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestPreparedRunJSONContractRoundTripsAllFields(t *testing.T) {
+	value := fullyPopulatedPreparedRun()
+	payload, err := json.Marshal(value)
+	if err != nil {
+		t.Fatalf("marshal PreparedRun: %v", err)
+	}
+
+	object := decodeJSONObject(t, payload)
+	assertJSONFields(t, object,
+		"prompt_id",
+		"prompt_version",
+		"prompt_hash",
+		"selected_profile_id",
+		"selected_backend_id",
+		"effective_model_params",
+		"output_contract",
+		"structured_output",
+		"input_hashes",
+		"session_id",
+		"rendered_prompt_hash",
+		"messages",
+		"start_time",
+		"end_time",
+		"duration_ms",
+	)
+
+	var messages []map[string]json.RawMessage
+	if err := json.Unmarshal(object["messages"], &messages); err != nil {
+		t.Fatalf("decode PreparedRun messages: %v", err)
+	}
+	if len(messages) != 2 {
+		t.Fatalf("message count = %d, want 2", len(messages))
+	}
+	if _, ok := messages[0]["cache_control"]; !ok {
+		t.Fatalf("first message omitted cache_control: %s", object["messages"])
+	}
+	if _, ok := messages[1]["cache_control"]; ok {
+		t.Fatalf("second message included empty cache_control: %s", object["messages"])
+	}
+
+	var decoded promptkit.PreparedRun
+	if err := json.Unmarshal(payload, &decoded); err != nil {
+		t.Fatalf("unmarshal PreparedRun: %v", err)
+	}
+	if !reflect.DeepEqual(decoded, value) {
+		t.Fatalf("PreparedRun did not round trip:\ngot  %#v\nwant %#v", decoded, value)
+	}
+
+	zeroPayload, err := json.Marshal(promptkit.PreparedRun{})
+	if err != nil {
+		t.Fatalf("marshal zero PreparedRun: %v", err)
+	}
+	zeroObject := decodeJSONObject(t, zeroPayload)
+	for _, field := range []string{
+		"prompt_version",
+		"prompt_hash",
+		"selected_backend_id",
+		"structured_output",
+		"input_hashes",
+		"session_id",
+		"start_time",
+		"end_time",
+		"duration_ms",
+	} {
+		if _, ok := zeroObject[field]; ok {
+			t.Fatalf("zero PreparedRun included %q: %s", field, zeroPayload)
+		}
+	}
+}
+
+func TestRunResultJSONContractRoundTripsAllFields(t *testing.T) {
+	value := fullyPopulatedRunResult()
+	payload, err := json.Marshal(value)
+	if err != nil {
+		t.Fatalf("marshal RunResult: %v", err)
+	}
+
+	object := decodeJSONObject(t, payload)
+	assertJSONFields(t, object,
+		"run_id",
+		"artifact",
+		"raw_output",
+		"validation",
+		"prompt_id",
+		"prompt_version",
+		"prompt_hash",
+		"session_id",
+		"rendered_prompt_hash",
+		"selected_profile_id",
+		"selected_backend_id",
+		"model_name",
+		"endpoint",
+		"effective_model_params",
+		"input_hashes",
+		"usage",
+		"start_time",
+		"end_time",
+		"duration_ms",
+	)
+	if _, ok := object["duration"]; ok {
+		t.Fatalf("RunResult included nanosecond duration field: %s", payload)
+	}
+
+	var decoded promptkit.RunResult
+	if err := json.Unmarshal(payload, &decoded); err != nil {
+		t.Fatalf("unmarshal RunResult: %v", err)
+	}
+	if !reflect.DeepEqual(decoded, value) {
+		t.Fatalf("RunResult did not round trip:\ngot  %#v\nwant %#v", decoded, value)
+	}
+
+	zeroPayload, err := json.Marshal(promptkit.RunResult{})
+	if err != nil {
+		t.Fatalf("marshal zero RunResult: %v", err)
+	}
+	zeroObject := decodeJSONObject(t, zeroPayload)
+	for _, field := range []string{
+		"prompt_version",
+		"prompt_hash",
+		"session_id",
+		"selected_backend_id",
+		"input_hashes",
+		"start_time",
+		"end_time",
+		"duration_ms",
+		"duration",
+	} {
+		if _, ok := zeroObject[field]; ok {
+			t.Fatalf("zero RunResult included %q: %s", field, zeroPayload)
+		}
+	}
+}
+
+func TestRunResultJSONDurationMillisecondBoundaries(t *testing.T) {
+	maxMilliseconds := int64(time.Duration(math.MaxInt64) / time.Millisecond)
+	minMilliseconds := int64(time.Duration(math.MinInt64) / time.Millisecond)
+
+	for _, milliseconds := range []int64{minMilliseconds, maxMilliseconds} {
+		t.Run(strconv.FormatInt(milliseconds, 10), func(t *testing.T) {
+			var decoded promptkit.RunResult
+			if err := json.Unmarshal(durationPayload(milliseconds), &decoded); err != nil {
+				t.Fatalf("decode representable duration_ms: %v", err)
+			}
+			want := time.Duration(milliseconds) * time.Millisecond
+			if decoded.Duration != want {
+				t.Fatalf("Duration = %v, want %v", decoded.Duration, want)
+			}
+		})
+	}
+
+	for _, milliseconds := range []int64{
+		minMilliseconds - 1,
+		maxMilliseconds + 1,
+		math.MinInt64,
+		math.MaxInt64,
+	} {
+		t.Run(strconv.FormatInt(milliseconds, 10), func(t *testing.T) {
+			original := fullyPopulatedRunResult()
+			decoded := original
+			err := json.Unmarshal(durationPayload(milliseconds), &decoded)
+			if err == nil || !strings.Contains(err.Error(), "duration_ms") || !strings.Contains(err.Error(), "time.Duration") {
+				t.Fatalf("overflow error = %v", err)
+			}
+			if !reflect.DeepEqual(decoded, original) {
+				t.Fatalf("failed decode partially updated receiver:\ngot  %#v\nwant %#v", decoded, original)
+			}
+		})
+	}
+}
+
+func fullyPopulatedPreparedRun() promptkit.PreparedRun {
+	start := time.Date(2026, time.August, 11, 12, 13, 14, 150_000_000, time.UTC)
+	return promptkit.PreparedRun{
+		PromptID:             "prompt.prepared",
+		PromptVersion:        "2.1.0",
+		PromptHash:           "prompt-hash",
+		SelectedProfileID:    "profile-prepared",
+		SelectedBackendID:    "backend-prepared",
+		EffectiveModelParams: jsonContractExecutionTarget(),
+		OutputContract: promptkit.OutputContract{
+			Format:         promptkit.FormatJSON,
+			ValidationMode: promptkit.ValidationJSONSchema,
+			SchemaPath:     "schemas/prepared.json",
+			RepairAttempts: 2,
+		},
+		StructuredOutput: &promptkit.StructuredOutputSpec{
+			Type: promptkit.StructuredOutputJSONSchema,
+			JSONSchema: &promptkit.StructuredOutputJSONSpec{
+				Name:   "prepared_schema",
+				Strict: true,
+				Schema: map[string]any{
+					"type":     "object",
+					"required": []any{"value"},
+					"properties": map[string]any{
+						"value": map[string]any{"minimum": float64(1)},
+					},
+				},
+			},
+		},
+		InputHashes:        map[string]string{"first": "hash-1", "second": "hash-2"},
+		SessionID:          "session-prepared",
+		RenderedPromptHash: "rendered-hash",
+		Messages: []promptkit.RenderedMessage{
+			{
+				Role:    "system",
+				Content: "System message",
+				CacheControl: &promptkit.CacheControl{
+					Type: promptkit.CacheControlEphemeral,
+					TTL:  "1h",
+				},
+			},
+			{Role: "user", Content: "User message"},
+		},
+		StartTime:  start,
+		EndTime:    start.Add(1501 * time.Millisecond),
+		DurationMS: 1501,
+	}
+}
+
+func fullyPopulatedRunResult() promptkit.RunResult {
+	start := time.Date(2026, time.August, 11, 15, 16, 17, 250_000_000, time.UTC)
+	return promptkit.RunResult{
+		RunID: "run-id",
+		Artifact: promptkit.Artifact{
+			Name:        "result.json",
+			ContentType: "application/json",
+			Body:        []byte(`{"value":2}`),
+			URI:         "memory://result.json",
+			Size:        11,
+			Hash:        "artifact-hash",
+		},
+		RawOutput: `{"value":2}`,
+		Validation: promptkit.ValidationResult{
+			Status:         promptkit.ValidationFailed,
+			Mode:           promptkit.ValidationJSONSchema,
+			Errors:         []string{"first error", "second error"},
+			SchemaPath:     "schemas/result.json",
+			RepairAttempts: 2,
+			IsValid:        false,
+		},
+		PromptID:             "prompt.result",
+		PromptVersion:        "3.2.1",
+		PromptHash:           "result-prompt-hash",
+		SessionID:            "session-result",
+		RenderedPromptHash:   "result-rendered-hash",
+		SelectedProfileID:    "profile-result",
+		SelectedBackendID:    "backend-result",
+		ModelName:            "model-result",
+		Endpoint:             "https://result.example/v1",
+		EffectiveModelParams: jsonContractExecutionTarget(),
+		InputHashes:          map[string]string{"input": "input-hash"},
+		Usage: promptkit.TokenUsage{
+			PromptTokens:     101,
+			CompletionTokens: 202,
+			TotalTokens:      303,
+			CachedTokens:     44,
+			CacheWriteTokens: 55,
+		},
+		StartTime: start,
+		EndTime:   start.Add(1750 * time.Millisecond),
+		Duration:  1750 * time.Millisecond,
+	}
+}
+
+func jsonContractExecutionTarget() promptkit.ExecutionTarget {
+	return promptkit.ExecutionTarget{
+		BackendID:       "backend-target",
+		Endpoint:        "https://target.example/v1",
+		Model:           "model-target",
+		Temperature:     0.75,
+		MaxTokens:       321,
+		TopP:            0.875,
+		TimeoutSeconds:  43,
+		ServiceTier:     "priority",
+		ReasoningEffort: "high",
+		APIKeyEnv:       "PROMPTKIT_JSON_CONTRACT_KEY",
+		ExtraParams: map[string]any{
+			"enabled": true,
+			"weight":  float64(1.25),
+			"nested":  map[string]any{"name": "value"},
+		},
+	}
+}
+
+func durationPayload(milliseconds int64) []byte {
+	return []byte(`{"duration_ms":` + strconv.FormatInt(milliseconds, 10) + `}`)
+}
+
+func decodeJSONObject(t *testing.T, payload []byte) map[string]json.RawMessage {
+	t.Helper()
+	var object map[string]json.RawMessage
+	if err := json.Unmarshal(payload, &object); err != nil {
+		t.Fatalf("decode JSON object: %v", err)
+	}
+	return object
+}
+
+func assertJSONFields(t *testing.T, object map[string]json.RawMessage, fields ...string) {
+	t.Helper()
+	want := make(map[string]struct{}, len(fields))
+	for _, field := range fields {
+		want[field] = struct{}{}
+	}
+	for field := range object {
+		if _, ok := want[field]; !ok {
+			t.Errorf("unexpected JSON field %q", field)
+		}
+	}
+	for field := range want {
+		if _, ok := object[field]; !ok {
+			t.Errorf("missing JSON field %q", field)
+		}
+	}
+}

llm_adapter_internal_test.go

@@ -0,0 +1,96 @@
+package promptkit
+
+import (
+	"context"
+	"reflect"
+	"strconv"
+	"sync"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestPublicLLMClientAdapterGivesClientOwnedNestedValues(t *testing.T) {
+	source := adapterOwnershipRequest()
+	want := adapterOwnershipRequest()
+	client := &retainingMutatingLLMClient{}
+	adapter := publicLLMClientAdapter{client: client}
+
+	if _, err := adapter.Generate(context.Background(), source); err != nil {
+		t.Fatalf("generate: %v", err)
+	}
+	if !reflect.DeepEqual(source, want) {
+		t.Fatalf("client mutation changed prepared source:\ngot  %#v\nwant %#v", source, want)
+	}
+
+	var mutations sync.WaitGroup
+	mutations.Add(1)
+	go func() {
+		defer mutations.Done()
+		for i := 0; i < 10_000; i++ {
+			mutateGenerateRequest(&client.retained, strconv.Itoa(i))
+		}
+	}()
+	for i := 0; i < 10_000; i++ {
+		laterRequest := fromDomainGenerateRequest(source)
+		if laterRequest.Prompt.Messages[0].Content != "source-message" ||
+			laterRequest.Prompt.Messages[0].CacheControl.TTL != "source-ttl" ||
+			laterRequest.Target.ExtraParams["nested"].([]any)[0] != "source-extra" ||
+			laterRequest.StructuredOutput.JSONSchema.Schema.(map[string]any)["enum"].([]any)[0] != "source-schema" {
+			t.Fatal("retained client mutation reached a later execution request")
+		}
+	}
+	mutations.Wait()
+
+	if !reflect.DeepEqual(source, want) {
+		t.Fatalf("retained client mutation changed prepared source:\ngot  %#v\nwant %#v", source, want)
+	}
+}
+
+type retainingMutatingLLMClient struct {
+	retained GenerateRequest
+}
+
+func (c *retainingMutatingLLMClient) Generate(_ context.Context, request GenerateRequest) (*GenerateResponse, error) {
+	c.retained = request
+	mutateGenerateRequest(&c.retained, "client-mutation")
+	return &GenerateResponse{Content: "generated"}, nil
+}
+
+func adapterOwnershipRequest() domain.GenerateRequest {
+	return domain.GenerateRequest{
+		Prompt: domain.RenderedPrompt{
+			SessionID: "source-session",
+			Messages: []domain.RenderedMessage{{
+				Role:    "user",
+				Content: "source-message",
+				CacheControl: &domain.CacheControl{
+					Type: domain.CacheControlEphemeral,
+					TTL:  "source-ttl",
+				},
+			}},
+		},
+		Target: domain.ExecutionTarget{
+			Model:  "source-model",
+			APIKey: "source-api-key",
+			ExtraParams: map[string]any{
+				"nested": []any{"source-extra"},
+			},
+		},
+		StructuredOutput: &domain.StructuredOutputSpec{
+			Type: domain.StructuredOutputJSONSchema,
+			JSONSchema: &domain.StructuredOutputJSONSpec{
+				Name:   "source-schema-name",
+				Strict: true,
+				Schema: map[string]any{"enum": []any{"source-schema"}},
+			},
+		},
+	}
+}
+
+func mutateGenerateRequest(request *GenerateRequest, value string) {
+	request.Prompt.Messages[0].Content = value
+	request.Prompt.Messages[0].CacheControl.TTL = value
+	request.Target.ExtraParams["nested"].([]any)[0] = value
+	request.StructuredOutput.JSONSchema.Schema.(map[string]any)["enum"].([]any)[0] = value
+}

output_contract_contract_test.go

@@ -0,0 +1,73 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestPreparationRejectsInvalidOutputContractWithPublicError(t *testing.T) {
+	engine, err := promptkit.NewEngine(
+		promptkit.Config{},
+		promptkit.WithPromptFS(contractPromptFS("prompt", "profile", "message"), "."),
+		promptkit.WithProfiles(promptkit.Profile{
+			ID:       "profile",
+			Endpoint: "http://example.test/v1",
+			Model:    "model",
+		}),
+	)
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+
+	for _, tc := range []struct {
+		name     string
+		contract promptkit.OutputContract
+	}{
+		{
+			name: "unsupported format",
+			contract: promptkit.OutputContract{
+				Format:         promptkit.OutputFormat("binary"),
+				ValidationMode: promptkit.ValidationNone,
+			},
+		},
+		{
+			name: "repair attempts above maximum",
+			contract: promptkit.OutputContract{
+				Format:         promptkit.FormatText,
+				ValidationMode: promptkit.ValidationBasic,
+				RepairAttempts: 4,
+			},
+		},
+		{
+			name: "none validation with repair attempts",
+			contract: promptkit.OutputContract{
+				Format:         promptkit.FormatText,
+				ValidationMode: promptkit.ValidationNone,
+				RepairAttempts: 1,
+			},
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			req := promptkit.RunRequest{PromptID: "prompt", Validation: &tc.contract}
+
+			prepared, err := engine.Prepare(context.Background(), req)
+			if prepared != nil {
+				t.Fatalf("expected no partial prepared run, got %+v", prepared)
+			}
+			if !errors.Is(err, promptkit.ErrInvalidRequest) {
+				t.Fatalf("prepare error = %v, want ErrInvalidRequest", err)
+			}
+
+			preparedExecution, err := engine.PrepareExecution(context.Background(), req)
+			if preparedExecution != nil {
+				t.Fatalf("expected no partial prepared execution, got %+v", preparedExecution)
+			}
+			if !errors.Is(err, promptkit.ErrInvalidRequest) {
+				t.Fatalf("prepare execution error = %v, want ErrInvalidRequest", err)
+			}
+		})
+	}
+}

prepared_execution.go

@@ -44,12 +44,12 @@ func (p *PreparedExecution) Discard() {
 
 // String returns a constant representation that exposes no retained request,
 // rendered content, or credential data.
-func (p *PreparedExecution) String() string {
+func (p PreparedExecution) String() string {
 	return preparedExecutionString
 }
 
 // GoString returns a constant Go-syntax representation that exposes no
 // retained request, rendered content, or credential data.
-func (p *PreparedExecution) GoString() string {
+func (p PreparedExecution) GoString() string {
 	return preparedExecutionString
 }

prepared_execution_contract_test.go

@@ -5,7 +5,6 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
-	"os"
 	"reflect"
 	"strings"
 	"sync"
@@ -257,6 +256,50 @@ func TestPreparedExecutionLifecycleAndEngineBinding(t *testing.T) {
 	}
 }
 
+func TestPreparedExecutionRepairsEmptyBasicOutput(t *testing.T) {
+	client := &preparedRecordingClient{responses: []*promptkit.GenerateResponse{
+		{
+			Content: "",
+			Usage:   promptkit.TokenUsage{PromptTokens: 3, CompletionTokens: 5, TotalTokens: 8},
+		},
+		{
+			Content: "Corrected summary.",
+			Usage:   promptkit.TokenUsage{PromptTokens: 7, CompletionTokens: 11, TotalTokens: 18},
+		},
+	}}
+	engine := newPreparedContractEngine(t, client, "Summarize the source.")
+	prepared, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{
+		PromptID: "prepared",
+		Validation: &promptkit.OutputContract{
+			Format:         promptkit.FormatMarkdown,
+			ValidationMode: promptkit.ValidationBasic,
+			RepairAttempts: 1,
+		},
+	})
+	if err != nil {
+		t.Fatalf("prepare execution: %v", err)
+	}
+	details := prepared.Details()
+
+	result, err := engine.RunPrepared(context.Background(), prepared)
+	if err != nil {
+		t.Fatalf("run prepared: %v", err)
+	}
+	if result.RawOutput != "Corrected summary." || result.Validation.Status != promptkit.ValidationPassed ||
+		result.Validation.RepairAttempts != 1 || result.Usage != (promptkit.TokenUsage{PromptTokens: 10, CompletionTokens: 16, TotalTokens: 26}) {
+		t.Fatalf("repaired result = %+v", result)
+	}
+	requests := client.snapshot()
+	if len(requests) != 2 || len(requests[1].Prompt.Messages) != len(details.Messages)+1 ||
+		!reflect.DeepEqual(requests[1].Prompt.Messages[:len(details.Messages)], details.Messages) ||
+		requests[1].Prompt.Messages[len(requests[1].Prompt.Messages)-1].Role != "user" {
+		t.Fatalf("prepared repair requests = %#v", requests)
+	}
+	if _, err := engine.RunPrepared(context.Background(), prepared); !errors.Is(err, promptkit.ErrInvalidRequest) {
+		t.Fatalf("second RunPrepared error = %v, want ErrInvalidRequest", err)
+	}
+}
+
 func TestPreparedExecutionConcurrentClaimAllowsOneGeneration(t *testing.T) {
 	release := make(chan struct{})
 	client := &preparedRecordingClient{
@@ -409,14 +452,35 @@ func TestPreparedExecutionDiscardAndFormattingDoNotExposePrivateState(t *testing
 		t.Fatalf("prepare execution: %v", err)
 	}
 
-	formattedValues := []string{
+	copied := *prepared
-		fmt.Sprint(prepared),
+	zeroValue := promptkit.PreparedExecution{}
-		fmt.Sprintf("%+v", prepared),
+	var nilHandle *promptkit.PreparedExecution
-		fmt.Sprintf("%#v", prepared),
+	for name, value := range map[string]any{
-	}
+		"original pointer": prepared,
-	for _, formatted := range formattedValues {
+		"copied value":     copied,
+		"zero value":       zeroValue,
+		"zero pointer":     &zeroValue,
+	} {
+		for format, formatted := range map[string]string{
+			"String":   fmt.Sprintf("%s", value),
+			"GoString": fmt.Sprintf("%#v", value),
+			"v":        fmt.Sprintf("%v", value),
+			"+v":       fmt.Sprintf("%+v", value),
+		} {
 			if formatted != "promptkit.PreparedExecution{opaque}" {
-			t.Fatalf("unexpected opaque formatting: %q", formatted)
+				t.Fatalf("%s %s formatting = %q, want opaque representation", name, format, formatted)
+			}
+			assertPreparedPrivateValuesAbsent(t, formatted, directCredential, renderedContent)
+		}
+	}
+	for format, formatted := range map[string]string{
+		"String":   fmt.Sprintf("%s", nilHandle),
+		"GoString": fmt.Sprintf("%#v", nilHandle),
+		"v":        fmt.Sprintf("%v", nilHandle),
+		"+v":       fmt.Sprintf("%+v", nilHandle),
+	} {
+		if formatted != "<nil>" {
+			t.Fatalf("nil pointer %s formatting = %q, want <nil>", format, formatted)
 		}
 		assertPreparedPrivateValuesAbsent(t, formatted, directCredential, renderedContent)
 	}
@@ -433,27 +497,9 @@ func TestPreparedExecutionDiscardAndFormattingDoNotExposePrivateState(t *testing
 	}
 	assertPreparedPrivateValuesAbsent(t, string(detailsJSON), directCredential)
 
-	prepared.Discard()
+	executionResult, err := engine.RunPrepared(context.Background(), &copied)
-	prepared.Discard()
-	result, lifecycleErr := engine.RunPrepared(context.Background(), prepared)
-	if result != nil || !errors.Is(lifecycleErr, promptkit.ErrInvalidRequest) {
-		t.Fatalf("discarded execution result=(%+v, %v), want ErrInvalidRequest", result, lifecycleErr)
-	}
-	assertPreparedPrivateValuesAbsent(t, lifecycleErr.Error(), directCredential, renderedContent)
-	if !reflect.DeepEqual(prepared.Details(), detailsBefore) {
-		t.Fatal("details changed after discard")
-	}
-
-	executed, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{
-		PromptID: "prepared",
-		APIKey:   directCredential,
-	})
 	if err != nil {
-		t.Fatalf("prepare execution for request inspection: %v", err)
+		t.Fatalf("run copied execution after formatting: %v", err)
-	}
-	executionResult, err := engine.RunPrepared(context.Background(), executed)
-	if err != nil {
-		t.Fatalf("run execution for request inspection: %v", err)
 	}
 	requests := client.snapshot()
 	if len(requests) != 1 || requests[0].APIKey != directCredential {
@@ -478,16 +524,34 @@ func TestPreparedExecutionDiscardAndFormattingDoNotExposePrivateState(t *testing
 	}
 	assertPreparedPrivateValuesAbsent(t, string(resultJSON), directCredential)
 
-	var nilHandle *promptkit.PreparedExecution
 	nilHandle.Discard()
 	if !reflect.DeepEqual(nilHandle.Details(), promptkit.PreparedRun{}) {
 		t.Fatalf("nil handle details=%+v, want zero value", nilHandle.Details())
 	}
-	zeroHandle := &promptkit.PreparedExecution{}
+	zeroHandle := &zeroValue
 	zeroHandle.Discard()
 	if !reflect.DeepEqual(zeroHandle.Details(), promptkit.PreparedRun{}) {
 		t.Fatalf("zero handle details=%+v, want zero value", zeroHandle.Details())
 	}
+
+	discarded, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{
+		PromptID: "prepared",
+		APIKey:   directCredential,
+	})
+	if err != nil {
+		t.Fatalf("prepare execution for discard: %v", err)
+	}
+	discardedDetails := discarded.Details()
+	discarded.Discard()
+	discarded.Discard()
+	result, lifecycleErr := engine.RunPrepared(context.Background(), discarded)
+	if result != nil || !errors.Is(lifecycleErr, promptkit.ErrInvalidRequest) {
+		t.Fatalf("discarded execution result=(%+v, %v), want ErrInvalidRequest", result, lifecycleErr)
+	}
+	assertPreparedPrivateValuesAbsent(t, lifecycleErr.Error(), directCredential, renderedContent)
+	if !reflect.DeepEqual(discarded.Details(), discardedDetails) {
+		t.Fatal("details changed after discard")
+	}
 }
 
 func TestPreparedExecutionCredentialCapacityAndTimingBoundaries(t *testing.T) {
@@ -503,19 +567,25 @@ func TestPreparedExecutionCredentialCapacityAndTimingBoundaries(t *testing.T) {
 		engine, err := promptkit.NewEngine(
 			promptkit.Config{},
 			promptkit.WithPromptFS(contractPromptFS("prepared", "profile", "content"), "."),
-			promptkit.WithProfileFS(preparedCredentialProfileSource(environmentName), "."),
+			promptkit.WithProfiles(promptkit.Profile{
+				ID:             "profile",
+				Endpoint:       "http://example.test/v1",
+				Model:          "model",
+				APIKeyRequired: true,
+			}),
 			promptkit.WithLLMClient(client),
 		)
 		if err != nil {
 			t.Fatalf("construct credential engine: %v", err)
 		}
-		prepared, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{PromptID: "prepared"})
+		prepared, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{
+			PromptID:  "prepared",
+			Execution: &promptkit.ExecutionTargetOverride{APIKeyEnv: environmentName},
+		})
 		if err != nil {
 			t.Fatalf("prepare credential execution: %v", err)
 		}
-		if err := os.Unsetenv(environmentName); err != nil {
+		t.Setenv(environmentName, "")
-			t.Fatalf("unset credential environment: %v", err)
-		}
 
 		result, err := engine.RunPrepared(context.Background(), prepared)
 		if result != nil ||
@@ -647,6 +717,7 @@ func (r *mutablePreparedArtifactReader) callCount() int {
 type preparedRecordingClient struct {
 	mu        sync.Mutex
 	response  *promptkit.GenerateResponse
+	responses []*promptkit.GenerateResponse
 	err       error
 	requests  []promptkit.GenerateRequest
 	started   chan struct{}
@@ -674,6 +745,12 @@ func (c *preparedRecordingClient) Generate(
 	if c.err != nil {
 		return nil, c.err
 	}
+	if len(c.responses) > 0 {
+		if index := len(c.requests) - 1; index < len(c.responses) {
+			return c.responses[index], nil
+		}
+		return nil, fmt.Errorf("no response configured for generation %d", len(c.requests))
+	}
 	return c.response, nil
 }
 
@@ -734,16 +811,6 @@ model: ` + model + `
 	}
 }
 
-func preparedCredentialProfileSource(environmentName string) fstest.MapFS {
-	return fstest.MapFS{
-		"profile.yaml": &fstest.MapFile{Data: []byte(`id: profile
-endpoint: http://example.test/v1
-model: model
-api_key_env: ` + environmentName + `
-`)},
-	}
-}
-
 func preparedSchemaSource() fstest.MapFS {
 	return fstest.MapFS{
 		"schema.json": &fstest.MapFile{Data: []byte(`{

profile_inheritance_contract_test.go

@@ -0,0 +1,243 @@
+package promptkit_test
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"strings"
+	"sync"
+	"testing"
+	"testing/fstest"
+
+	"gitea.maximumdirect.net/eric/promptkit"
+)
+
+func TestProfileInheritanceBuiltInAliasWorkflow(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{
+		PromptDir: frameworkPromptDir,
+		SchemaDir: frameworkSchemaDir,
+	}, promptkit.WithProfiles(promptkit.Profile{
+		ID:              "weather-light",
+		BaseProfileID:   "deepseek-4-flash",
+		ReasoningEffort: "high",
+		TimeoutSeconds:  120,
+	}))
+	if err != nil {
+		t.Fatalf("construct alias engine: %v", err)
+	}
+
+	base, err := engine.InspectProfile(context.Background(), "deepseek-4-flash")
+	if err != nil {
+		t.Fatalf("inspect base: %v", err)
+	}
+	child, err := engine.InspectProfile(context.Background(), "weather-light")
+	if err != nil {
+		t.Fatalf("inspect alias: %v", err)
+	}
+	if child.ProfileID != "weather-light" ||
+		child.EffectiveModelParams.BackendID != base.EffectiveModelParams.BackendID ||
+		child.EffectiveModelParams.Model != base.EffectiveModelParams.Model ||
+		child.EffectiveModelParams.ReasoningEffort != "high" ||
+		child.EffectiveModelParams.TimeoutSeconds != 120 {
+		t.Fatalf("alias inspection = %+v, base = %+v", child, base)
+	}
+
+	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+		PromptID:  frameworkMarkdownSummaryPromptID,
+		ProfileID: "weather-light",
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline("Rin opens the gate."),
+			"glossary":   promptkit.Inline("gate: A guarded passage."),
+		},
+	})
+	if err != nil {
+		t.Fatalf("prepare alias: %v", err)
+	}
+	if prepared.SelectedProfileID != "weather-light" {
+		t.Fatalf("SelectedProfileID = %q", prepared.SelectedProfileID)
+	}
+
+	timeout := 15
+	reasoning := "low"
+	overridden, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+		PromptID:  frameworkMarkdownSummaryPromptID,
+		ProfileID: "weather-light",
+		Execution: &promptkit.ExecutionTargetOverride{
+			TimeoutSeconds:  &timeout,
+			ReasoningEffort: &reasoning,
+		},
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline("Rin opens the gate."),
+			"glossary":   promptkit.Inline("gate: A guarded passage."),
+		},
+	})
+	if err != nil {
+		t.Fatalf("prepare override: %v", err)
+	}
+	if overridden.EffectiveModelParams.TimeoutSeconds != timeout ||
+		overridden.EffectiveModelParams.ReasoningEffort != reasoning {
+		t.Fatalf("runtime override target = %+v", overridden.EffectiveModelParams)
+	}
+}
+
+func TestProfileInheritanceYAMLAliasOfBuiltIn(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{},
+		promptkit.WithPromptFS(fstest.MapFS{}, "."),
+		promptkit.WithProfileFS(fstest.MapFS{
+			"alias.yaml": &fstest.MapFile{Data: []byte("id: yaml-alias\nbase_profile: deepseek-4-flash\n")},
+		}, "."),
+	)
+	if err != nil {
+		t.Fatalf("construct YAML alias engine: %v", err)
+	}
+	inspection, err := engine.InspectProfile(context.Background(), "yaml-alias")
+	if err != nil {
+		t.Fatalf("inspect YAML alias: %v", err)
+	}
+	if inspection.ProfileID != "yaml-alias" || inspection.EffectiveModelParams.Model == "" {
+		t.Fatalf("YAML alias inspection = %+v", inspection)
+	}
+}
+
+func TestProfileInheritanceRetainsRequiredCredentialBehavior(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{},
+		promptkit.WithPromptFS(fstest.MapFS{}, "."),
+		promptkit.WithBackend(promptkit.Backend{
+			ID:        "credential-backend",
+			Endpoint:  "https://credential.example/v1",
+			APIKeyEnv: "OPTIONAL_BACKEND_KEY",
+		}),
+		promptkit.WithProfiles(
+			promptkit.Profile{
+				ID:             "credential-base",
+				BackendID:      "credential-backend",
+				Model:          "model",
+				APIKeyRequired: true,
+			},
+			promptkit.Profile{ID: "credential-child", BaseProfileID: "credential-base"},
+		),
+	)
+	if err != nil {
+		t.Fatalf("construct credential inheritance engine: %v", err)
+	}
+	inspection, err := engine.InspectProfile(context.Background(), "credential-child")
+	if err != nil {
+		t.Fatalf("inspect credential child: %v", err)
+	}
+	if !inspection.APIKeyRequired || inspection.EffectiveModelParams.APIKeyEnv != "" {
+		t.Fatalf("credential inspection = %+v", inspection)
+	}
+}
+
+func TestProfileInheritancePreservesPublicErrorIdentities(t *testing.T) {
+	newEngine := func(t *testing.T, profiles fs.FS) *promptkit.Engine {
+		t.Helper()
+		options := []promptkit.Option{promptkit.WithPromptFS(fstest.MapFS{}, ".")}
+		if profiles != nil {
+			options = append(options, promptkit.WithProfileFS(profiles, "."))
+		}
+		engine, err := promptkit.NewEngine(promptkit.Config{}, options...)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		return engine
+	}
+
+	tests := []struct {
+		name     string
+		profiles fs.FS
+		profile  string
+		contains []string
+		want     error
+		wantNot  error
+	}{
+		{
+			name:    "missing selected profile",
+			profile: "missing",
+			want:    promptkit.ErrProfileNotFound,
+			wantNot: promptkit.ErrProfileLoad,
+		},
+		{
+			name: "missing base",
+			profiles: fstest.MapFS{
+				"child.yaml": &fstest.MapFile{Data: []byte("id: child\nbase_profile: missing\n")},
+			},
+			profile:  "child",
+			contains: []string{"child", "missing"},
+			want:     promptkit.ErrProfileLoad,
+			wantNot:  promptkit.ErrProfileNotFound,
+		},
+		{
+			name: "cycle",
+			profiles: fstest.MapFS{
+				"a.yaml": &fstest.MapFile{Data: []byte("id: a\nbase_profile: b\n")},
+				"b.yaml": &fstest.MapFile{Data: []byte("id: b\nbase_profile: a\n")},
+			},
+			profile:  "a",
+			contains: []string{"a", "b"},
+			want:     promptkit.ErrProfileLoad,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			result, err := newEngine(t, tc.profiles).InspectProfile(context.Background(), tc.profile)
+			if result != nil || !errors.Is(err, tc.want) || (tc.wantNot != nil && errors.Is(err, tc.wantNot)) {
+				t.Fatalf("inspection=(%+v, %v), want %v without %v", result, err, tc.want, tc.wantNot)
+			}
+			for _, fragment := range tc.contains {
+				if !strings.Contains(err.Error(), fragment) {
+					t.Fatalf("error = %v, want %q", err, fragment)
+				}
+			}
+		})
+	}
+}
+
+func TestProfileInheritanceFreezesPreparedExecution(t *testing.T) {
+	profiles := &mutableInheritanceProfileFS{files: fstest.MapFS{
+		"child.yaml": &fstest.MapFile{Data: []byte("id: child\nbase_profile: base\n")},
+		"base.yaml":  &fstest.MapFile{Data: []byte("id: base\nendpoint: https://base.example/v1\nmodel: first-model\n")},
+	}}
+	client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "ok"}}
+	engine, err := promptkit.NewEngine(promptkit.Config{},
+		promptkit.WithPromptFS(contractPromptFS("prepared", "child", "content"), "."),
+		promptkit.WithProfileFS(profiles, "."),
+		promptkit.WithLLMClient(client),
+	)
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+
+	prepared, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{PromptID: "prepared"})
+	if err != nil {
+		t.Fatalf("prepare execution: %v", err)
+	}
+	profiles.set("base.yaml", "id: base\nendpoint: https://base.example/v1\nmodel: second-model\n")
+
+	result, err := engine.RunPrepared(context.Background(), prepared)
+	if err != nil || result == nil || len(client.requests) != 1 || client.requests[0].Target.Model != "first-model" {
+		t.Fatalf("prepared execution=(%+v, %v), requests=%+v", result, err, client.requests)
+	}
+	inspection, err := engine.InspectProfile(context.Background(), "child")
+	if err != nil || inspection.EffectiveModelParams.Model != "second-model" {
+		t.Fatalf("fresh inspection=(%+v, %v)", inspection, err)
+	}
+}
+
+type mutableInheritanceProfileFS struct {
+	mu    sync.RWMutex
+	files fstest.MapFS
+}
+
+func (f *mutableInheritanceProfileFS) Open(name string) (fs.File, error) {
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+	return f.files.Open(name)
+}
+
+func (f *mutableInheritanceProfileFS) set(name, content string) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.files[name] = &fstest.MapFile{Data: []byte(content)}
+}

profiles.go

@@ -2,17 +2,16 @@ package promptkit
 
 import (
 	"context"
-	"errors"
 	"fmt"
-	"strings"
 
 	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
 	"gitea.maximumdirect.net/eric/promptkit/internal/jsonvalue"
 	"gitea.maximumdirect.net/eric/promptkit/internal/profile"
 )
 
-// OpenAICompatibleProfile returns an ordinary in-memory Profile for an
+// OpenAICompatibleProfile returns an in-memory Profile for an OpenAI-compatible
-// OpenAI-compatible chat-completions endpoint.
+// chat-completions endpoint. A non-blank BaseProfileID permits its target
+// fields to be inherited when the profile is selected or inspected.
 //
 // It does not register global state, maintain a model catalog, or resolve
 // credentials. If APIKeyRequired is true, callers satisfy it with
@@ -25,6 +24,7 @@ import (
 func OpenAICompatibleProfile(cfg OpenAICompatibleProfileConfig) Profile {
 	return Profile{
 		ID:              cfg.ID,
+		BaseProfileID:   cfg.BaseProfileID,
 		BackendID:       cfg.BackendID,
 		Endpoint:        cfg.Endpoint,
 		Model:           cfg.Model,
@@ -87,8 +87,9 @@ func toDomainProfile(publicProfile Profile) (domain.ExecutionProfile, error) {
 		return domain.ExecutionProfile{}, err
 	}
 	prof := domain.ExecutionProfile{
-		ID:              strings.TrimSpace(publicProfile.ID),
+		ID:              publicProfile.ID,
-		BackendID:       strings.TrimSpace(publicProfile.BackendID),
+		BaseProfileID:   publicProfile.BaseProfileID,
+		BackendID:       publicProfile.BackendID,
 		Endpoint:        publicProfile.Endpoint,
 		Model:           publicProfile.Model,
 		Temperature:     publicProfile.Temperature,
@@ -100,33 +101,8 @@ func toDomainProfile(publicProfile Profile) (domain.ExecutionProfile, error) {
 		APIKeyRequired:  publicProfile.APIKeyRequired,
 		ExtraParams:     extraParams,
 	}
-	if err := validatePublicProfile(prof); err != nil {
+	if err := profile.NormalizeAndValidateDefinition(&prof); err != nil {
 		return domain.ExecutionProfile{}, err
 	}
 	return prof, nil
 }
-
-func validatePublicProfile(prof domain.ExecutionProfile) error {
-	if strings.TrimSpace(prof.ID) == "" {
-		return errors.New("id is required")
-	}
-	if strings.TrimSpace(prof.BackendID) == "" && strings.TrimSpace(prof.Endpoint) == "" {
-		return errors.New("backend or endpoint is required")
-	}
-	if strings.TrimSpace(prof.Model) == "" {
-		return errors.New("model is required")
-	}
-	if prof.Temperature < 0 || prof.Temperature > 2 {
-		return errors.New("temperature must be between 0 and 2")
-	}
-	if prof.MaxTokens < 0 {
-		return errors.New("max_tokens must be greater than or equal to 0")
-	}
-	if prof.TopP < 0 || prof.TopP > 1 {
-		return errors.New("top_p must be between 0 and 1")
-	}
-	if prof.TimeoutSeconds < 0 {
-		return errors.New("timeout_seconds must be greater than or equal to 0")
-	}
-	return nil
-}

public_contract_test.go

@@ -12,7 +12,6 @@ import (
 	"sync/atomic"
 	"testing"
 	"testing/fstest"
-	"time"
 
 	"gitea.maximumdirect.net/eric/promptkit"
 )
@@ -78,6 +77,41 @@ api_key_env: PROMPTKIT_INSPECTION_ABSENT_KEY
 	}
 }
 
+func TestFileProfileNormalizedIDMatchesInspectionAndPreparation(t *testing.T) {
+	engine, err := promptkit.NewEngine(promptkit.Config{},
+		promptkit.WithPromptFS(contractPromptFS("prompt", "normalized-profile", "message"), "."),
+		promptkit.WithProfileFS(fstest.MapFS{
+			"profile.yaml": &fstest.MapFile{Data: []byte(`
+id: "  normalized-profile  "
+endpoint: http://profile.example/v1
+model: normalized-model
+`)},
+		}, "."),
+	)
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+
+	inspection, err := engine.InspectProfile(context.Background(), " normalized-profile ")
+	if err != nil {
+		t.Fatalf("inspect normalized profile: %v", err)
+	}
+	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{
+		PromptID:  "prompt",
+		ProfileID: " normalized-profile ",
+	})
+	if err != nil {
+		t.Fatalf("prepare with normalized profile: %v", err)
+	}
+
+	if inspection.ProfileID != "normalized-profile" ||
+		prepared.SelectedProfileID != inspection.ProfileID ||
+		inspection.EffectiveModelParams.Model != "normalized-model" ||
+		prepared.EffectiveModelParams.Model != inspection.EffectiveModelParams.Model {
+		t.Fatalf("inspection=%#v prepared=%#v", inspection, prepared)
+	}
+}
+
 func TestInspectProfilePreservesPublicErrorIdentities(t *testing.T) {
 	newEngine := func(t *testing.T, options ...promptkit.Option) *promptkit.Engine {
 		t.Helper()
@@ -110,7 +144,7 @@ func TestInspectProfilePreservesPublicErrorIdentities(t *testing.T) {
 	}
 
 	malformed := newEngine(t, promptkit.WithProfileFS(fstest.MapFS{
-		"broken.yaml": &fstest.MapFile{Data: []byte("id: broken\nendpoint: http://broken.example/v1\nmodel: model\nunknown: value\n")},
+		"broken.yaml": &fstest.MapFile{Data: []byte("id: broken\nendpoint: http://broken.example/v1\nmodel: model\nextra_params:\n  invalid: .nan\n")},
 	}, "."))
 	if result, err := malformed.InspectProfile(context.Background(), "broken"); result != nil ||
 		!errors.Is(err, promptkit.ErrProfileLoad) {
@@ -391,18 +425,6 @@ output:
 	}
 }
 
-func TestPreparedRunJSONOmitsZeroTimingValues(t *testing.T) {
-	payload, err := json.Marshal(promptkit.PreparedRun{})
-	if err != nil {
-		t.Fatalf("marshal prepared run: %v", err)
-	}
-	for _, field := range []string{"start_time", "end_time", "duration_ms"} {
-		if strings.Contains(string(payload), `"`+field+`"`) {
-			t.Fatalf("expected zero %s to be omitted, got %s", field, payload)
-		}
-	}
-}
-
 func TestBackendIdentityJSONNamesAndOmission(t *testing.T) {
 	t.Run("execution target round trip", func(t *testing.T) {
 		value := promptkit.ExecutionTarget{BackendID: promptkit.BackendOpenRouter}
@@ -757,13 +779,17 @@ func TestBackendRegistrationRejectsInvalidAndDuplicateDefinitions(t *testing.T)
 		{name: "reserved extra parameter", backends: []promptkit.Backend{{ID: "custom", Endpoint: "http://example.test/v1", ExtraParams: map[string]any{"model": "override"}}}},
 		{name: "cyclic extra parameter", backends: []promptkit.Backend{{ID: "custom", Endpoint: "http://example.test/v1", ExtraParams: cycle}}},
 		{name: "malformed JSON number", backends: []promptkit.Backend{{ID: "custom", Endpoint: "http://example.test/v1", ExtraParams: map[string]any{"value": json.Number("01")}}}},
+		{name: "excessively deep extra parameter", backends: []promptkit.Backend{{ID: "custom", Endpoint: "http://example.test/v1", ExtraParams: map[string]any{"value": excessivelyDeepJSONValue()}}}},
 		{name: "duplicate consumer id", backends: []promptkit.Backend{
 			{ID: " custom ", Endpoint: "http://one.example/v1"},
 			{ID: "custom", Endpoint: "http://two.example/v1"},
 		}},
-		{name: "reserved built-in id", backends: []promptkit.Backend{{
+		{name: "reserved OpenRouter ID", backends: []promptkit.Backend{{
 			ID: promptkit.BackendOpenRouter, Endpoint: "http://replacement.example/v1",
 		}}},
+		{name: "reserved Rakestrawhome ID", backends: []promptkit.Backend{{
+			ID: promptkit.BackendRakestrawHome, Endpoint: "http://replacement.example/v1",
+		}}},
 	}
 
 	for _, tt := range tests {
@@ -818,93 +844,7 @@ func TestBackendExtraParamsAreDeeplyCopiedAtConstructionAndLookup(t *testing.T)
 	}
 }
 
-func TestPreparedRunJSONTimingRoundTrips(t *testing.T) {
+func TestEngineValidationWithZeroRepairBudgetIsSinglePass(t *testing.T) {
-	start := time.Date(2026, time.July, 29, 12, 0, 0, 0, time.UTC)
-	prepared := promptkit.PreparedRun{
-		PromptID:   "prompt",
-		StartTime:  start,
-		EndTime:    start.Add(1250 * time.Millisecond),
-		DurationMS: 1250,
-	}
-
-	payload, err := json.Marshal(prepared)
-	if err != nil {
-		t.Fatalf("marshal prepared run: %v", err)
-	}
-	var decoded promptkit.PreparedRun
-	if err := json.Unmarshal(payload, &decoded); err != nil {
-		t.Fatalf("unmarshal prepared run: %v", err)
-	}
-	if decoded.DurationMS != prepared.DurationMS ||
-		!decoded.StartTime.Equal(prepared.StartTime) ||
-		!decoded.EndTime.Equal(prepared.EndTime) {
-		t.Fatalf("timing values did not round trip: got %#v, want %#v", decoded, prepared)
-	}
-}
-
-func TestRunResultJSONUsesMillisecondsAndRoundTrips(t *testing.T) {
-	start := time.Date(2026, time.July, 29, 12, 0, 0, 0, time.UTC)
-	result := promptkit.RunResult{
-		RunID:     "opaque-run-id",
-		Artifact:  promptkit.Artifact{Name: "output", ContentType: "text/plain", Body: []byte("ok")},
-		SessionID: "session-123",
-		StartTime: start,
-		EndTime:   start.Add(1500 * time.Millisecond),
-		Duration:  1500 * time.Millisecond,
-	}
-
-	payload, err := json.Marshal(result)
-	if err != nil {
-		t.Fatalf("marshal run result: %v", err)
-	}
-	var object map[string]any
-	if err := json.Unmarshal(payload, &object); err != nil {
-		t.Fatalf("decode run result JSON: %v", err)
-	}
-	if got := object["duration_ms"]; got != float64(1500) {
-		t.Fatalf("expected duration_ms=1500, got %#v in %s", got, payload)
-	}
-	if _, exists := object["duration"]; exists {
-		t.Fatalf("unexpected nanosecond duration field in %s", payload)
-	}
-	if got := object["session_id"]; got != result.SessionID {
-		t.Fatalf("expected session_id=%q, got %#v in %s", result.SessionID, got, payload)
-	}
-	artifact, ok := object["artifact"].(map[string]any)
-	if !ok || artifact["content_type"] != "text/plain" {
-		t.Fatalf("expected stable artifact JSON fields, got %#v", object["artifact"])
-	}
-
-	var decoded promptkit.RunResult
-	if err := json.Unmarshal(payload, &decoded); err != nil {
-		t.Fatalf("unmarshal run result: %v", err)
-	}
-	if decoded.SessionID != result.SessionID ||
-		decoded.Duration != result.Duration ||
-		!decoded.StartTime.Equal(result.StartTime) ||
-		!decoded.EndTime.Equal(result.EndTime) {
-		t.Fatalf("timing values did not round trip: got %#v, want %#v", decoded, result)
-	}
-
-	payload, err = json.Marshal(promptkit.RunResult{})
-	if err != nil {
-		t.Fatalf("marshal zero run result: %v", err)
-	}
-	for _, field := range []string{"session_id", "start_time", "end_time", "duration_ms"} {
-		if strings.Contains(string(payload), `"`+field+`"`) {
-			t.Fatalf("expected zero %s to be omitted, got %s", field, payload)
-		}
-	}
-	var decodedEmpty promptkit.RunResult
-	if err := json.Unmarshal(payload, &decodedEmpty); err != nil {
-		t.Fatalf("unmarshal run result without session_id: %v", err)
-	}
-	if decodedEmpty.SessionID != "" {
-		t.Fatalf("expected absent session_id to decode empty, got %q", decodedEmpty.SessionID)
-	}
-}
-
-func TestEngineValidationIsSinglePass(t *testing.T) {
 	client := &fakeLLMClient{
 		response: &promptkit.GenerateResponse{Content: "not-json"},
 	}
@@ -919,7 +859,7 @@ func TestEngineValidationIsSinglePass(t *testing.T) {
 		Validation: &promptkit.OutputContract{
 			Format:         promptkit.FormatJSON,
 			ValidationMode: promptkit.ValidationJSON,
-			RepairAttempts: 3,
+			RepairAttempts: 0,
 		},
 	})
 	if err != nil {
@@ -934,6 +874,87 @@ func TestEngineValidationIsSinglePass(t *testing.T) {
 	}
 }
 
+func TestEngineRunRepairsJSONSchemaOutput(t *testing.T) {
+	client := &fakeLLMClient{responses: []*promptkit.GenerateResponse{
+		{
+			Content: "{}",
+			Usage:   promptkit.TokenUsage{PromptTokens: 2, CompletionTokens: 3, TotalTokens: 5, CachedTokens: 7, CacheWriteTokens: 11},
+		},
+		{
+			Content: `{"events":[{"title":"Repaired event"}]}`,
+			Usage:   promptkit.TokenUsage{PromptTokens: 13, CompletionTokens: 17, TotalTokens: 19, CachedTokens: 23, CacheWriteTokens: 29},
+		},
+	}}
+	engine := newContractEngineWithOptions(t, frameworkSchemaDir, promptkit.WithLLMClient(client))
+
+	result, err := engine.Run(context.Background(), promptkit.RunRequest{
+		PromptID:  frameworkStructuredEventsPromptID,
+		SessionID: " repair-session ",
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline("Rin opens the gate."),
+			"glossary":   promptkit.Inline("gate: A guarded passage."),
+		},
+		Validation: &promptkit.OutputContract{
+			Format:         promptkit.FormatJSON,
+			ValidationMode: promptkit.ValidationJSONSchema,
+			SchemaPath:     "structured_events.schema.json",
+			RepairAttempts: 1,
+		},
+	})
+	if err != nil {
+		t.Fatalf("run: %v", err)
+	}
+	if result.RawOutput != client.responses[1].Content || result.Validation.Status != promptkit.ValidationPassed ||
+		result.Validation.RepairAttempts != 1 {
+		t.Fatalf("repaired result = %+v", result)
+	}
+	wantUsage := promptkit.TokenUsage{PromptTokens: 15, CompletionTokens: 20, TotalTokens: 24, CachedTokens: 30, CacheWriteTokens: 40}
+	if result.Usage != wantUsage {
+		t.Fatalf("usage = %+v, want %+v", result.Usage, wantUsage)
+	}
+	if len(client.requests) != 2 {
+		t.Fatalf("generation calls = %d, want 2", len(client.requests))
+	}
+	initial, repaired := client.requests[0], client.requests[1]
+	if initial.Prompt.SessionID != "repair-session" || repaired.Prompt.SessionID != initial.Prompt.SessionID ||
+		!reflect.DeepEqual(repaired.Target, initial.Target) || repaired.TargetPresence != initial.TargetPresence ||
+		!reflect.DeepEqual(repaired.StructuredOutput, initial.StructuredOutput) {
+		t.Fatalf("generation request state drifted: initial=%+v repaired=%+v", initial, repaired)
+	}
+	if initial.StructuredOutput == nil || initial.StructuredOutput.JSONSchema == nil {
+		t.Fatalf("expected structured output on initial request: %+v", initial)
+	}
+}
+
+func TestEngineRunReturnsFinalResultAfterRepairExhaustion(t *testing.T) {
+	client := &fakeLLMClient{responses: []*promptkit.GenerateResponse{
+		{Content: "not-json", Usage: promptkit.TokenUsage{PromptTokens: 2, CompletionTokens: 3, TotalTokens: 5}},
+		{Content: "still-not-json", Usage: promptkit.TokenUsage{PromptTokens: 7, CompletionTokens: 11, TotalTokens: 18}},
+	}}
+	engine := newContractEngineWithOptions(t, frameworkSchemaDir, promptkit.WithLLMClient(client))
+
+	result, err := engine.Run(context.Background(), promptkit.RunRequest{
+		PromptID: frameworkMarkdownSummaryPromptID,
+		Inputs: map[string]promptkit.ArtifactRef{
+			"transcript": promptkit.Inline("Rin opens the gate."),
+			"glossary":   promptkit.Inline("gate: A guarded passage."),
+		},
+		Validation: &promptkit.OutputContract{
+			Format:         promptkit.FormatJSON,
+			ValidationMode: promptkit.ValidationJSON,
+			RepairAttempts: 1,
+		},
+	})
+	if err != nil || result == nil {
+		t.Fatalf("run = (%+v, %v), want exhausted result", result, err)
+	}
+	if result.RawOutput != "still-not-json" || result.Validation.Status != promptkit.ValidationFailed ||
+		result.Validation.RepairAttempts != 1 || len(result.Validation.Errors) == 0 ||
+		result.Usage != (promptkit.TokenUsage{PromptTokens: 9, CompletionTokens: 14, TotalTokens: 23}) {
+		t.Fatalf("exhausted result = %+v", result)
+	}
+}
+
 func TestRepeatedOptionsUseLastValueInEachCategory(t *testing.T) {
 	profile := promptkit.Profile{ID: "profile", Endpoint: "http://example.test/v1", Model: "model"}
 
@@ -973,6 +994,24 @@ func TestRepeatedOptionsUseLastValueInEachCategory(t *testing.T) {
 		}
 	})
 
+	t.Run("fallback profile source", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", "profile", "message"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS("profile", "first-model"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS("profile", "second-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+		if err != nil {
+			t.Fatalf("prepare from last fallback profile source: %v", err)
+		}
+		if prepared.EffectiveModelParams.Model != "second-model" {
+			t.Fatalf("expected last fallback profile source, got %q", prepared.EffectiveModelParams.Model)
+		}
+	})
+
 	t.Run("in-memory profiles", func(t *testing.T) {
 		first := profile
 		first.Model = "first-model"
@@ -1057,6 +1096,227 @@ func TestRepeatedOptionsUseLastValueInEachCategory(t *testing.T) {
 	})
 }
 
+func TestFallbackProfileSourcePrecedence(t *testing.T) {
+	const profileID = "application-profile"
+
+	prepareModel := func(t *testing.T, engine *promptkit.Engine, promptID string) string {
+		t.Helper()
+		prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: promptID})
+		if err != nil {
+			t.Fatalf("prepare: %v", err)
+		}
+		return prepared.EffectiveModelParams.Model
+	}
+
+	t.Run("in-memory profiles override ordinary and fallback profiles", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithProfiles(promptkit.Profile{ID: profileID, Endpoint: "http://example.test/v1", Model: "memory-model"}),
+			promptkit.WithProfileFS(contractProfileFS(profileID, "ordinary-model"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != "memory-model" {
+			t.Fatalf("expected in-memory profile, got %q", model)
+		}
+	})
+
+	t.Run("ordinary filesystem source overrides fallback profile", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithProfileFS(contractProfileFS(profileID, "ordinary-model"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != "ordinary-model" {
+			t.Fatalf("expected ordinary profile, got %q", model)
+		}
+	})
+
+	t.Run("ordinary option replaces configured directory", func(t *testing.T) {
+		profileDir := t.TempDir()
+		writePublicProfileFile(t, profileDir, profileID, "http://example.test/v1", "directory-model")
+		engine, err := promptkit.NewEngine(promptkit.Config{ProfileDir: profileDir},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithProfileFS(contractProfileFS(profileID, "option-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != "option-model" {
+			t.Fatalf("expected ordinary option profile, got %q", model)
+		}
+	})
+
+	t.Run("configured directory overrides fallback profile", func(t *testing.T) {
+		profileDir := t.TempDir()
+		writePublicProfileFile(t, profileDir, profileID, "http://example.test/v1", "directory-model")
+		engine, err := promptkit.NewEngine(promptkit.Config{ProfileDir: profileDir},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != "directory-model" {
+			t.Fatalf("expected configured directory profile, got %q", model)
+		}
+	})
+
+	t.Run("fallback profile overrides built-in profile", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", "mistral-small-3", "message"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS("mistral-small-3", "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != "fallback-model" {
+			t.Fatalf("expected fallback profile, got %q", model)
+		}
+	})
+
+	t.Run("missing fallback profile uses built-in profile", func(t *testing.T) {
+		t.Setenv("OPENROUTER_API_KEY", "test-key")
+		baseline, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", "mistral-small-3", "message"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct baseline engine: %v", err)
+		}
+		want := prepareModel(t, baseline, "prompt")
+
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", "mistral-small-3", "message"), "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if model := prepareModel(t, engine, "prompt"); model != want {
+			t.Fatalf("expected built-in profile model %q, got %q", want, model)
+		}
+	})
+}
+
+func TestFallbackProfileSourcePreservesLazyLoadingAndErrors(t *testing.T) {
+	const profileID = "application-profile"
+
+	t.Run("construction defers malformed fallback profiles", func(t *testing.T) {
+		_, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(fstest.MapFS{}, "."),
+			promptkit.WithFallbackProfileFS(fstest.MapFS{
+				"broken.yaml": &fstest.MapFile{Data: []byte("id: broken\nunknown: value\n")},
+			}, "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine with malformed fallback profile: %v", err)
+		}
+	})
+
+	t.Run("unrelated malformed fallback profile does not block matching definition", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithFallbackProfileFS(fstest.MapFS{
+				"broken.yaml": &fstest.MapFile{Data: []byte("id: unrelated\nunknown: value\n")},
+				"valid.yaml":  &fstest.MapFile{Data: []byte("id: application-profile\nendpoint: http://example.test/v1\nmodel: fallback-model\n")},
+			}, "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+		if err != nil {
+			t.Fatalf("prepare from valid fallback profile: %v", err)
+		}
+		if prepared.EffectiveModelParams.Model != "fallback-model" {
+			t.Fatalf("unexpected fallback profile model: %q", prepared.EffectiveModelParams.Model)
+		}
+	})
+
+	t.Run("matching malformed fallback profile does not reach built-in profile", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", "mistral-small-3", "message"), "."),
+			promptkit.WithFallbackProfileFS(fstest.MapFS{
+				"mistral-small-3.yaml": &fstest.MapFile{Data: []byte("id: mistral-small-3\nendpoint: http://example.test/v1\nmodel: fallback-model\nunknown: value\n")},
+			}, "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if _, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"}); !errors.Is(err, promptkit.ErrProfileLoad) {
+			t.Fatalf("expected ErrProfileLoad, got %v", err)
+		}
+	})
+
+	t.Run("matching malformed ordinary profile does not reach fallback profile", func(t *testing.T) {
+		engine, err := promptkit.NewEngine(promptkit.Config{},
+			promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+			promptkit.WithProfileFS(fstest.MapFS{
+				"application-profile.yaml": &fstest.MapFile{Data: []byte("id: application-profile\nendpoint: http://example.test/v1\nmodel: ordinary-model\nunknown: value\n")},
+			}, "."),
+			promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		)
+		if err != nil {
+			t.Fatalf("construct engine: %v", err)
+		}
+		if _, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"}); !errors.Is(err, promptkit.ErrProfileLoad) {
+			t.Fatalf("expected ErrProfileLoad, got %v", err)
+		}
+	})
+}
+
+func TestFallbackProfileSourceWorksAcrossWorkflows(t *testing.T) {
+	const profileID = "application-profile"
+	client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "ok"}}
+	engine, err := promptkit.NewEngine(promptkit.Config{},
+		promptkit.WithPromptFS(contractPromptFS("prompt", profileID, "message"), "."),
+		promptkit.WithFallbackProfileFS(contractProfileFS(profileID, "fallback-model"), "."),
+		promptkit.WithLLMClient(client),
+	)
+	if err != nil {
+		t.Fatalf("construct engine: %v", err)
+	}
+
+	inspection, err := engine.InspectProfile(context.Background(), profileID)
+	if err != nil {
+		t.Fatalf("inspect fallback profile: %v", err)
+	}
+	prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+	if err != nil {
+		t.Fatalf("prepare fallback profile: %v", err)
+	}
+	preparedExecution, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+	if err != nil {
+		t.Fatalf("prepare execution with fallback profile: %v", err)
+	}
+	preparedDetails := preparedExecution.Details()
+	preparedResult, err := engine.RunPrepared(context.Background(), preparedExecution)
+	if err != nil {
+		t.Fatalf("run prepared fallback profile: %v", err)
+	}
+	runResult, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
+	if err != nil {
+		t.Fatalf("run fallback profile: %v", err)
+	}
+
+	for name, model := range map[string]string{
+		"inspection":         inspection.EffectiveModelParams.Model,
+		"preparation":        prepared.EffectiveModelParams.Model,
+		"prepared execution": preparedDetails.EffectiveModelParams.Model,
+		"prepared result":    preparedResult.EffectiveModelParams.Model,
+		"run result":         runResult.EffectiveModelParams.Model,
+	} {
+		if model != "fallback-model" {
+			t.Fatalf("%s model=%q, want fallback-model", name, model)
+		}
+	}
+}
+
 func TestEngineSupportsConcurrentPrepareAndRun(t *testing.T) {
 	engine, err := promptkit.NewEngine(promptkit.Config{},
 		promptkit.WithPromptFS(contractPromptFS("prompt", "profile", "message"), "."),

types.go

@@ -82,7 +82,8 @@ const (
 // Prepare, PrepareExecution, and Run copy the request's maps, pointers, and
 // nested JSON-compatible values before using them. The caller may mutate the
 // request after any method returns. A successful PrepareExecution retains its
-// own private execution snapshot for RunPrepared.
+// own private execution snapshot for RunPrepared. Excessively deep or large
+// JSON-shaped values are rejected for safety.
 type RunRequest struct {
 	// PromptID is the required non-empty prompt identifier.
 	PromptID string
@@ -116,7 +117,8 @@ type RunRequest struct {
 	// empty maps are equivalent.
 	Vars map[string]string
 	// Execution optionally overrides individual execution settings. Nil uses
-	// the selected profile over its backend, when any, and framework defaults.
+	// the selected profile over its backend, when any, and the framework
+	// baseline.
 	Execution *ExecutionTargetOverride
 	// Validation optionally replaces the prompt's complete output contract. It
 	// does not merge individual fields. Nil uses the prompt contract.
@@ -144,9 +146,10 @@ type PreparedRun struct {
 	// SelectedBackendID equals EffectiveModelParams.BackendID. It is empty for
 	// an endpoint-only profile.
 	SelectedBackendID string `json:"selected_backend_id,omitempty"`
-	// EffectiveModelParams contains framework defaults overlaid by the selected
+	// EffectiveModelParams contains settings resolved from the framework timeout
-	// backend, profile, and then request overrides. It excludes resolved API-key
+	// baseline, selected backend, profile, and then request overrides. Unset
-	// values.
+	// optional provider controls remain zero rather than reporting a provider
+	// default. It excludes resolved API-key values.
 	EffectiveModelParams ExecutionTarget `json:"effective_model_params"`
 	// OutputContract is the complete effective output contract.
 	OutputContract OutputContract `json:"output_contract"`
@@ -240,8 +243,8 @@ type ArtifactRef struct {
 	// URI is the file path for ArtifactRefFile and optional provenance metadata
 	// for ArtifactRefInline.
 	URI string
-	// Body is the content for ArtifactRefInline and is ignored for
+	// Body is the content for ArtifactRefInline, where an empty value is valid,
-	// ArtifactRefFile.
+	// and is ignored for ArtifactRefFile.
 	Body string
 }
 
@@ -294,17 +297,22 @@ type ExecutionTarget struct {
 	// empty for endpoint-only profiles. It is supplied to injected LLMClient
 	// implementations as part of the effective target.
 	BackendID string `json:"backend_id,omitempty"`
-	// Endpoint is the model-provider base URL.
+	// Endpoint is the normalized absolute HTTP or HTTPS model-provider base URL.
+	// It has a host and no user information, query, or fragment.
 	Endpoint string `json:"endpoint"`
 	// Model is the provider model identifier.
 	Model string `json:"model"`
-	// Temperature is the effective sampling temperature from 0 through 2.
+	// Temperature is the resolved sampling temperature from 0 through 2. Zero
+	// leaves the field unspecified to compatible providers unless the
+	// corresponding ExecutionTargetPresence bit is true.
 	Temperature float64 `json:"temperature"`
-	// MaxTokens is the non-negative effective output-token limit. Zero leaves
+	// MaxTokens is the non-negative resolved output-token limit. Zero leaves
-	// the limit unspecified to compatible providers unless it was an explicit
+	// the limit unspecified to compatible providers unless the corresponding
-	// request override.
+	// ExecutionTargetPresence bit is true.
 	MaxTokens int `json:"max_tokens"`
-	// TopP is the effective nucleus-sampling value from 0 through 1.
+	// TopP is the resolved nucleus-sampling value from 0 through 1. Zero leaves
+	// the field unspecified to compatible providers unless the corresponding
+	// ExecutionTargetPresence bit is true.
 	TopP float64 `json:"top_p"`
 	// TimeoutSeconds is the non-negative per-generation deadline. Zero disables
 	// this deadline without disabling caller cancellation or the transport cap.
@@ -314,7 +322,10 @@ type ExecutionTarget struct {
 	// ReasoningEffort is the effective opaque provider-specific reasoning
 	// setting. An empty value instructs model clients to omit reasoning.
 	ReasoningEffort string `json:"reasoning_effort"`
-	// APIKeyEnv is an environment-variable name, not its credential value.
+	// APIKeyEnv is the resolved name of an optional environment lookup source,
+	// not its credential value. The built-in client omits Authorization when no
+	// usable direct or environment credential is available; injected clients may
+	// resolve this metadata differently.
 	APIKeyEnv string `json:"api_key_env"`
 	// ExtraParams contains copied JSON-compatible provider parameters.
 	ExtraParams map[string]any `json:"extra_params"`
@@ -329,9 +340,11 @@ type ExecutionTarget struct {
 type ProfileInspection struct {
 	// ProfileID is the trimmed, exact profile ID inspected by the engine.
 	ProfileID string
-	// EffectiveModelParams contains framework defaults overlaid by the selected
+	// EffectiveModelParams contains settings resolved from the framework timeout
-	// backend and then the profile, without a request override. APIKeyEnv is an
+	// baseline, selected backend, and then profile, without a request override.
-	// environment-variable name, never its credential value.
+	// Unset optional provider controls remain zero rather than reporting a
+	// provider default. APIKeyEnv is an environment-variable name, never its
+	// credential value.
 	EffectiveModelParams ExecutionTarget
 	// APIKeyRequired reports that a later execution must supply a direct API
 	// key or an explicit request environment override. It is mutually exclusive
@@ -382,19 +395,28 @@ type PromptInspection struct {
 // fields replace profile values and preserve explicit zero or empty values. A
 // non-empty ExtraParams map replaces the complete profile or backend map
 // rather than merging keys. Empty string fields, nil pointers, and a nil or
-// empty ExtraParams map inherit the selected profile over its backend, when
+// empty ExtraParams map inherit lower-precedence values. An optional provider
-// any, and framework defaults.
+// control that remains zero is unspecified; TimeoutSeconds retains its
+// framework deadline when no higher-precedence value is present.
 type ExecutionTargetOverride struct {
 	// Endpoint replaces the profile or backend endpoint when non-empty without
-	// changing the effective BackendID.
+	// changing the effective BackendID. Preparation trims it and requires an
+	// absolute HTTP or HTTPS URL with a host and no user information, query, or
+	// fragment.
 	Endpoint string
 	// Model replaces the profile model when non-empty.
 	Model string
-	// Temperature, when non-nil, must point to a value from 0 through 2.
+	// Temperature, when non-nil, must point to a value from 0 through 2. A
+	// pointed-to zero is explicitly present; nil inherits a lower-precedence
+	// value and otherwise leaves the provider control unspecified.
 	Temperature *float64
-	// MaxTokens, when non-nil, must point to a non-negative value.
+	// MaxTokens, when non-nil, must point to a non-negative value. A pointed-to
+	// zero is explicitly present; nil inherits a lower-precedence value and
+	// otherwise leaves the provider control unspecified.
 	MaxTokens *int
-	// TopP, when non-nil, must point to a value from 0 through 1.
+	// TopP, when non-nil, must point to a value from 0 through 1. A pointed-to
+	// zero is explicitly present; nil inherits a lower-precedence value and
+	// otherwise leaves the provider control unspecified.
 	TopP *float64
 	// TimeoutSeconds, when non-nil, must point to a non-negative value. A
 	// pointed-to zero disables the per-generation deadline.
@@ -407,61 +429,81 @@ type ExecutionTargetOverride struct {
 	// inherited value and disables reasoning for this run. Non-blank values
 	// are opaque and are not validated against a fixed vocabulary.
 	ReasoningEffort *string
-	// APIKeyEnv replaces the profile or backend environment-variable name when
+	// APIKeyEnv replaces the profile or backend optional environment lookup
-	// non-blank. A direct RunRequest.APIKey still takes precedence over
+	// source when non-blank. A direct RunRequest.APIKey still takes precedence.
-	// environment lookup.
+	// The built-in client omits Authorization when neither source has a usable
+	// value; injected clients may resolve this metadata differently.
 	APIKeyEnv string
 	// ExtraParams, when non-empty, replaces the complete profile or backend map.
 	// Values must be JSON-compatible: nil, booleans, finite numbers, strings,
-	// arrays or slices, and maps with non-empty string keys. Cycles are invalid.
+	// arrays or slices, and maps with non-empty string keys. Cycles and
+	// excessively deep or large values are invalid.
 	ExtraParams map[string]any
 }
 
 // Profile is an in-memory execution profile for library consumers.
 //
-// It is equivalent to a loaded profile file after validation. Raw API keys do
+// A standalone Profile is equivalent to a loaded profile file after local
-// not belong in profiles; use APIKeyRequired to require callers to provide a
+// validation. A derived profile names BaseProfileID and can inherit target
-// RunRequest.APIKey or explicit request ExecutionTargetOverride.APIKeyEnv, or
+// fields when selected or inspected. Raw API keys do not belong in profiles;
-// use profile YAML api_key_env with file and FS profile sources. Profile has no
+// use APIKeyRequired to require callers to provide a RunRequest.APIKey or
-// stable JSON representation.
+// explicit request ExecutionTargetOverride.APIKeyEnv, or use profile YAML
+// api_key_env with file and FS profile sources. Profile has no stable JSON
+// representation.
 //
-// WithProfiles validates and copies Profile values during NewEngine. Numeric
+// WithProfiles locally validates and copies Profile values during NewEngine.
-// zero, blank strings, and an empty ExtraParams map inherit framework defaults;
+// It checks base-reference existence and resolved target completeness when a
-// use ExecutionTargetOverride pointer fields to request explicit numeric zero.
+// derived profile is selected or inspected. Zero Temperature, MaxTokens, and
+// TopP values and blank ServiceTier and ReasoningEffort values leave those
+// provider controls unspecified. A zero TimeoutSeconds retains the framework
+// deadline, while an empty ExtraParams map inherits backend request defaults.
+// Use ExecutionTargetOverride pointer fields to request an explicit numeric
+// zero.
 type Profile struct {
 	// ID is the required non-blank profile identifier. WithProfiles trims it.
 	ID string
+	// BaseProfileID optionally names one base profile. WithProfiles trims it. A
+	// non-blank value permits required target fields to be inherited when the
+	// profile is selected or inspected, which is also when reference existence
+	// and resolved completeness are checked. A blank value leaves this as a
+	// standalone profile.
+	BaseProfileID string
 	// BackendID optionally selects an engine backend. WithProfiles trims it.
 	// Backend membership is checked when a request selects the profile; an
 	// unknown ID makes preparation fail with ErrProfileLoad.
 	BackendID string
-	// Endpoint is the model-provider base URL. It is required only when
+	// Endpoint is the model-provider base URL. A standalone Profile requires an
-	// BackendID is blank and otherwise overrides the backend endpoint when
+	// endpoint when BackendID is blank; a derived Profile may inherit either
-	// non-blank.
+	// field. A non-blank endpoint overrides the backend endpoint when
+	// non-blank. WithProfiles trims it and requires an absolute HTTP or HTTPS URL
+	// with a host and no user information, query, or fragment.
 	Endpoint string
-	// Model is the required non-blank provider model identifier.
+	// Model is the provider model identifier. It is required for a standalone
+	// Profile and may be inherited by a derived Profile.
 	Model string
-	// Temperature is from 0 through 2. Zero inherits the framework default.
+	// Temperature is from 0 through 2. Zero leaves the provider control
+	// unspecified.
 	Temperature float64
-	// MaxTokens is non-negative. Zero inherits the framework default.
+	// MaxTokens is non-negative. Zero leaves the provider control unspecified.
 	MaxTokens int
-	// TopP is from 0 through 1. Zero inherits the framework default rather than
+	// TopP is from 0 through 1. Zero leaves the provider control unspecified
-	// selecting an explicit zero.
+	// rather than selecting an explicit zero.
 	TopP float64
-	// TimeoutSeconds is non-negative. Zero inherits the framework default.
+	// TimeoutSeconds is non-negative. Zero retains the framework deadline.
 	TimeoutSeconds int
-	// ServiceTier is optional; a blank value inherits the framework default.
+	// ServiceTier is optional; a blank value leaves it unspecified.
 	ServiceTier string
-	// ReasoningEffort is optional; a blank value inherits the framework
+	// ReasoningEffort is optional; a blank value leaves it unspecified.
-	// default.
 	ReasoningEffort string
 	// APIKeyRequired clears a backend's inherited API-key environment name and
 	// requires a non-blank RunRequest.APIKey unless the request explicitly
-	// supplies ExecutionTargetOverride.APIKeyEnv. It does not store a credential.
+	// supplies ExecutionTargetOverride.APIKeyEnv. When false, a named
+	// environment source remains optional. It does not store a credential.
 	APIKeyRequired bool
 	// ExtraParams contains provider-specific JSON-compatible values. An empty
 	// map inherits backend request defaults, when any. WithProfiles validates
-	// and deeply copies it during NewEngine.
+	// and deeply copies it during NewEngine. Excessively deep or large values
+	// are rejected for safety.
 	ExtraParams map[string]any
 }
 
@@ -469,13 +511,16 @@ type Profile struct {
 // profile.
 //
 // It contains ordinary profile fields for OpenAI-compatible chat-completions
-// endpoints. APIKeyRequired follows Profile.APIKeyRequired. Raw API keys do not
+// endpoints. BaseProfileID and APIKeyRequired follow Profile. Raw API keys do
-// belong in this config. OpenAICompatibleProfileConfig has no stable JSON
+// not belong in this config. OpenAICompatibleProfileConfig has no stable JSON
 // representation and is not validated until its resulting Profile is supplied
 // through WithProfiles to NewEngine.
 type OpenAICompatibleProfileConfig struct {
 	// ID becomes Profile.ID.
 	ID string
+	// BaseProfileID becomes Profile.BaseProfileID. A non-blank value permits the
+	// resulting Profile to inherit target fields when it is selected or inspected.
+	BaseProfileID string
 	// BackendID becomes Profile.BackendID.
 	BackendID string
 	// Endpoint becomes Profile.Endpoint.
@@ -520,11 +565,11 @@ type ExecutionTargetPresence struct {
 // JSON representation.
 //
 // A non-nil RunRequest.Validation replaces the complete prompt contract. It
-// does not merge fields. The public Engine validates generated output once and
+// does not merge fields. The public Engine performs bounded correction after a
-// does not install an output repairer.
+// failed eligible validation when RepairAttempts is positive.
 type OutputContract struct {
-	// Format selects generated artifact metadata. An empty effective value
+	// Format selects generated artifact metadata. An empty value in a non-nil
-	// defaults to FormatText.
+	// request replacement defaults to FormatText.
 	Format OutputFormat `json:"format"`
 	// ValidationMode selects the content check. Use one of the declared
 	// ValidationMode constants.
@@ -532,9 +577,9 @@ type OutputContract struct {
 	// SchemaPath is required when ValidationMode is ValidationJSONSchema and is
 	// ignored by other modes.
 	SchemaPath string `json:"schema_path"`
-	// RepairAttempts is a requested repair limit. A non-positive value requests
+	// RepairAttempts is an additional generation-call budget from zero through
-	// no repairs. The public Engine performs no repairs even when this value is
+	// three. Zero is single-pass. A positive value is valid only with basic,
-	// positive, so its runs report zero attempts used.
+	// json, or json_schema validation.
 	RepairAttempts int `json:"repair_attempts"`
 }
 
@@ -550,8 +595,8 @@ type ValidationResult struct {
 	Errors []string `json:"errors,omitempty"`
 	// SchemaPath is the effective schema path for JSON Schema validation.
 	SchemaPath string `json:"schema_path,omitempty"`
-	// RepairAttempts is the number of repairs actually attempted. It is always
+	// RepairAttempts is the number of corrective generation calls actually
-	// zero for the public Engine.
+	// started for this result.
 	RepairAttempts int `json:"repair_attempts"`
 	// IsValid is true for ValidationPassed and ValidationSkipped and false for
 	// ValidationFailed.
@@ -640,10 +685,10 @@ type StructuredOutputJSONSpec struct {
 // and retained copies. It is responsible for the cancellation behavior of any
 // work it starts and for synchronizing access to retained or shared data.
 //
-// A returned error makes Run or RunPrepared return ErrLLMGenerate while
+// An arbitrary returned error makes Run or RunPrepared return ErrLLMGenerate
-// preserving the client error through errors.Is. A nil response with a nil
+// while preserving the client error through errors.Is rather than translating
-// error also produces ErrLLMGenerate. Promptkit copies the non-nil response
+// it. A nil response with a nil error also produces ErrLLMGenerate. Promptkit
-// before returning from either method.
+// copies the non-nil response before returning from either method.
 type LLMClient interface {
 	Generate(context.Context, GenerateRequest) (*GenerateResponse, error)
 }
@@ -669,9 +714,8 @@ type GenerateRequest struct {
 // GenerateResponse is returned by an injected LLM client and has a stable JSON
 // representation.
 type GenerateResponse struct {
-	// Content is the generated output. It must be non-empty when using the
+	// Content is the generated output. It may be explicitly empty; Promptkit
-	// built-in client; injected clients may return empty content for Promptkit
+	// applies the effective output contract to classify it.
-	// validation to classify.
 	Content string `json:"content"`
 	// Usage is the client's token accounting.
 	Usage TokenUsage `json:"usage"`
@@ -679,8 +723,12 @@ type GenerateResponse struct {
 
 // File returns a file-backed artifact reference whose URI is path.
 //
-// The default artifact reader opens path as a caller-selected operating-system
+// The default artifact reader accepts path only when it resolves to a regular
-// path without restricting it to an application root or imposing a size limit.
+// operating-system file, checking that condition before and after opening it.
+// It reads synchronously in bounded chunks and checks context cancellation
+// before opening, before and after each read, and before returning the
+// artifact; it cannot interrupt a filesystem operation already in progress.
+// It does not restrict path to an application root or impose a size limit.
 // Applications accepting untrusted paths must validate them before calling
 // Promptkit or use [WithArtifactReader] to enforce application policy.
 func File(path string) ArtifactRef {
@@ -688,13 +736,13 @@ func File(path string) ArtifactRef {
 }
 
 // Inline returns an inline artifact reference whose Body is body and whose URI
-// is empty.
+// is empty. An empty body is a valid, explicitly supplied input.
 func Inline(body string) ArtifactRef {
 	return ArtifactRef{Type: ArtifactRefInline, Body: body}
 }
 
 // InlineWithURI returns an inline artifact reference with body content and uri
-// provenance metadata.
+// provenance metadata. An empty body is a valid, explicitly supplied input.
 func InlineWithURI(uri string, body string) ArtifactRef {
 	return ArtifactRef{Type: ArtifactRefInline, URI: uri, Body: body}
 }