Finish audit remediation and prepare v0.6.0

README.md

@@ -33,6 +33,12 @@ boundary and constraints that framework work must preserve.
 
 ## Release Guidance
 
+Consumers upgrading from `v0.5.0` to `v0.6.0` should read the
+[v0.6.0 changelog and migration guide](docs/releases/v0.6.0.md).
+
+Earlier adopters 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

@@ -34,7 +34,8 @@ type Backend struct {
 	// 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

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
@@ -152,8 +182,8 @@ replace execution settings or the complete output contract.
 The [public value GoDoc](../../types.go) defines nil, empty, zero, replacement,
 copy, and credential behavior. The
 [framework format reference](../formats.md) defines how those request values
-interact with prompt definitions, file-backed profiles, built-ins, schemas,
-and framework defaults.
+interact with prompt definitions, file-backed and application fallback
+profiles, built-ins, schemas, and framework defaults.
 
 For programmatic profiles,
 [`OpenAICompatibleProfile`](../../profiles.go) converts ordinary

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
-running the documented checks before accepting changes. Run the default Go
-validation from the Promptkit repository root:
+This section is the canonical local validation workflow for Promptkit. Run
+every command from the repository root before accepting a change. The test
+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
-Markdown link and confirm its target exists. Finally, check whitespace:
+### Local Markdown Links
+
+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
-requires link validation and `git diff --check`. Run the Go validation whenever
-documentation changes commands, examples, generated output, or another
-behavior checked by the module.
+Inspect `git status --short --untracked-files=all` and the complete diff before
+accepting a change. The status may contain only the intended source changes
+during development. Reject credentials, private keys, environment files,
+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
-
-Use focused checks while iterating, then run the complete validation sequence
-before accepting the change. The root package supports:
+After committing the accepted change, require a clean candidate:
 
 ```sh
-go test .
-go vet .
-go build .
+test -z "$(git status --porcelain)"
 ```
-
-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
-definition. Definitions are selected by their YAML `id`, not their file name
-or directory.
+`.yml`. Each prompt-definition and profile file contains exactly one YAML
+document; comments and trailing whitespace are allowed. YAML decoding is
+strict: unknown fields are errors for the selected definition. Definitions are
+selected by their YAML `id`, not their file name or directory.
 
 ## 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
-the prompt file and remains within the source root. `WithPromptFile` also
-resolves it relative to that file.
+`content_file` must be a relative path. It resolves from the directory that
+contains the prompt file and must remain within the configured prompt source
+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
@@ -156,9 +164,9 @@ extra_params:
 
 | 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. |
 | `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. |
@@ -183,30 +191,33 @@ 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` |
-| `max_tokens` | `0` |
-| `top_p` | `1` |
+| `temperature` | Unspecified and omitted from compatible provider requests unless a profile or runtime override selects it. |
+| `max_tokens` | Unspecified and omitted from compatible provider requests unless a profile or runtime override selects it. |
+| `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
-replace the framework default. Numeric request overrides use pointers, so an
-explicit zero is preserved. In particular, an explicit request
-`timeout_seconds` of zero disables the per-generation deadline while leaving
-the caller context and transport timeout intact.
+Numeric zero in a file or in-memory profile does not select a numeric value.
+For `temperature`, `max_tokens`, and `top_p`, it leaves the provider control
+unspecified. For `timeout_seconds`, it retains the framework deadline. Numeric
+request overrides use pointers, so an explicit zero is retained and sent to
+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
@@ -231,14 +242,18 @@ 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
-3. embedded built-in profiles.
+2. the ordinary configured source selected by a profile file, `fs.FS`, or
+   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
-invalid matching profile is an error and does not fall back. In-memory
-`Profile` values follow the same ranges as YAML profiles. They use
-`APIKeyRequired` for request-scoped credentials instead of `api_key_env`.
-Preparation and exact profile inspection use this same source precedence.
+A profile source supplies a complete definition; definitions and their fields
+are not merged across sources. A higher-precedence source falls back only when
+the requested profile ID is absent. An invalid matching profile is an error and
+does not fall back. In-memory `Profile` values follow the same ranges as YAML
+profiles. They use `APIKeyRequired` for request-scoped credentials instead of
+`api_key_env`. Preparation and exact profile inspection use this same source
+precedence.
 
 ## Built-In Profile Catalog
 
@@ -246,8 +261,8 @@ Every built-in selects the `openrouter` backend. The engine's built-in backend
 registry supplies `https://openrouter.ai/api/v1` and the environment-variable
 name `OPENROUTER_API_KEY`, so individual profiles contain only model and
 generation settings. Built-in profile files do not repeat those connection
-values. A custom or in-memory profile with the same profile ID takes
-precedence.
+values. A configured, application fallback, or in-memory profile with the same
+profile ID takes precedence.
 
 | Provider | ID | Model |
 | --- | --- | --- |

docs/integrations/openai-compatible-chat.md

@@ -14,10 +14,16 @@ that produce these outbound settings.
 
 Generation sends an HTTP `POST` with `Content-Type: application/json`.
 Before the client is called, the engine resolves framework, backend, profile,
-and request values into one execution target. A non-empty endpoint from that
-target overrides the client's configured base URL. After trailing slashes are
-removed, `/chat/completions` is appended. Generation fails before sending when
-neither source supplies an endpoint.
+and request values into one execution target. Endpoint configuration is trimmed
+and must be an absolute HTTP or HTTPS URL with a host and without user
+information, a query, or a fragment. A non-empty endpoint from the target
+overrides the client's configured base URL. The final selected endpoint is
+validated again before transport.
+
+The completion URL is composed through parsed URL path operations. Nested base
+paths are retained, repeated trailing slashes are normalized, and the result
+has exactly one appended `/chat/completions` suffix. Generation fails before
+sending when neither source supplies a valid endpoint.
 
 The target's backend ID is routing metadata for prepared values, results, and
 injected clients. The built-in client does not derive the URL from that ID and
@@ -53,8 +59,9 @@ never also sent as a session header.
 
 The client conditionally includes:
 
-- `temperature`, `max_tokens`, and `top_p` when non-zero or explicitly
-  present;
+- `temperature`, `max_tokens`, and `top_p` only when selected by a profile or
+  runtime override, including an explicit runtime zero; they are absent when
+  unspecified;
 - non-empty `service_tier` and effective `reasoning_effort`; an explicitly
   disabled reasoning setting is empty and therefore omitted; and
 - `response_format` for JSON Schema structured output, including its name,
@@ -81,13 +88,29 @@ request fields.
 
 ## Response Handling
 
-Any 2xx response is decoded as an OpenAI-compatible chat response. The client
-returns the first choice's non-empty message content and maps prompt,
-completion, total, cached, and cache-write token counts.
+Any 2xx response body is limited to 16 MiB (16,777,216 bytes). A larger
+declared `Content-Length` is rejected before the body is read, and streamed,
+chunked, or underreported bodies are read through the same bound with at most
+one additional byte used to detect overflow. A body exactly at the limit is
+allowed. The body is closed on every outcome and an oversized stream is not
+drained.
 
-Invalid JSON, absent choices, and empty first-choice content are malformed
-responses. For a non-2xx status, the error includes the status code but never
-the provider response body.
+The bounded body must contain exactly one OpenAI-compatible JSON response
+object followed only by JSON whitespace and EOF. The client returns the first
+choice's non-empty message content 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, empty first-choice content, and size
+overflow are malformed responses and return no partial result.
+
+For a non-2xx status, the error includes the status code but never the provider
+response body. Promptkit does not yet parse provider error envelopes; bounded
+non-success parsing belongs to the
+[structured-generation-error roadmap](../roadmap/structured-generation-errors.md).
+
+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 +125,7 @@ Timeouts are layered:
   timeout when the supplied value is not positive.
 
 The earliest applicable caller, generation, or transport deadline controls the
-request. Constructing the internal client does not mutate a supplied
-`http.Client`.
+request. Caller cancellation retains `context.Canceled`; caller, generation,
+and whole-request timeout failures retain `context.DeadlineExceeded`, together
+with the request-failure identity. Constructing the internal client does not
+mutate a supplied `http.Client`.

docs/internal/capacity.md

@@ -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 +25,20 @@ 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
-`http.Client` so Promptkit can apply its timeout default without mutating the
-caller's client. Generation then:
+Construction trims and validates a nonempty configured base URL and clones any
+supplied `http.Client` so Promptkit can apply its timeout default without
+mutating the caller's client. An empty configured base remains valid because a
+resolved request target may supply the endpoint. Generation then:
 
-1. validates request-level timeout and endpoint requirements;
+1. validates shared execution-setting invariants and the final selected base
+   endpoint;
 2. maps the internal request into the OpenAI-compatible chat payload;
 3. validates and merges extra parameters;
-4. resolves authentication;
-5. performs the outbound request under the applicable deadlines; and
-6. decodes the first response choice and token usage.
+4. composes `/chat/completions` through parsed URL path operations;
+5. resolves authentication;
+6. performs the outbound request under the applicable deadlines; and
+7. decodes one strictly framed, size-bounded response object and maps its first
+   choice and token usage.
 
 `internal/llm` owns the set of reserved OpenAI-compatible request fields used
 when validating extra parameters. Backend registration consumes the same rule
@@ -65,9 +69,28 @@ 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.
 
-Caller cancellation and deadline failures during the outbound request are
-reported as request execution failures. The runner classifies these identities
-without depending on HTTP status mapping.
+Invalid nonempty configured endpoints are configuration failures. A missing or
+invalid final selected endpoint is an invalid generation request and is
+rejected before transport.
+
+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. Non-success
+responses remain status-only; bounded provider error-envelope parsing belongs
+to the
+[structured-generation-error roadmap](../roadmap/structured-generation-errors.md).
+
+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 +98,11 @@ The
 [OpenAI-compatible client tests](../../internal/llm/openai_compatible_client_test.go)
 own configuration, client cloning, deterministic deadline precedence,
 authentication, request and response mapping, malformed data, error identity,
-cancellation, and response-body suppression. The root transport contract test
-also verifies that resolved backend settings reach this client without
-serializing backend identity. All use local test servers or test transports;
-the default suite makes no live or paid provider requests.
+cancellation, endpoint selection and composition, pre-transport rejection, and
+bounded single-document response framing, closure, and response-body
+suppression. The root
+transport contract tests also verify 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 errors, public error mapping, and engine-local profile-source assembly including application fallbacks. | [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/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/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 OpenRouter. | [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/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 optional 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

@@ -20,8 +20,9 @@ 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
-documents are loaded through the validator's optional schema-loader interface.
+admitter, and a model client wrapped by the same capacity manager. Validation
+plans and provider-facing schema metadata come from the validator's preparation
+interface.
 An output repairer can be injected internally, but the ordinary runner
 constructor does not enable one.
 
@@ -69,14 +70,16 @@ 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 +90,10 @@ 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 +124,17 @@ 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
-result; an inability to perform validation is an operational error.
+and validates that artifact with the plan compiled during completion. Invalid
+generated content remains a validation result; an inability to 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
@@ -129,20 +144,27 @@ each actual generation call.
 
 When an internal repairer is present, a JSON or JSON Schema content failure can
 trigger bounded repair attempts. Repair receives the effective execution
-target and session ID, validation errors, prior output, and structured-output
-specification. The default repairer uses the same wrapped client as initial
-generation, so each repair reacquires the selected backend's active permit
-while remaining inside its original admission lease. Repair never performs a
-second bounded admission. This capability remains internal and is not a public
-option.
+target, explicit numeric-presence bits, credential, backend identity, session
+ID, validation errors, prior output, and structured-output specification. One
+request constructor supplies those common fields to initial and repair
+generation while their rendered prompts remain intentionally distinct. 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, and repaired outputs use the operation's existing validation plan.
+This capability remains internal and is not a public option.
 
 A successful result includes the output artifact and raw output, validation
 state, effective session ID, prompt and rendered-prompt hashes, selected
 profile and backend, effective settings, input hashes, token usage, a generated
 run identifier, and UTC timing. The same effective session reaches initial
 generation and any repair attempt through the rendered prompt. The same
-effective target, including backend identity, reaches generation and any
-repair attempt.
+effective target and presence metadata, including backend identity and direct
+credential during execution, reaches generation and every repair attempt.
+Result usage is the field-wise sum of all five usage values from the initial
+response and every completed repair response. Final raw output, artifact, and
+validation state still come from the last candidate. A repair error returns no
+partial run result or partial usage.
 
 ## Failure Categories
 
@@ -165,7 +187,8 @@ 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 +200,9 @@ 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,
-error categories, artifact metadata, usage, and timing. The
+bounded repair progression, initial/repair request parity, cumulative usage,
+shared initial/repair capacity, credentials and redaction, error categories,
+artifact metadata, and timing. The
 [capacity subsystem document](capacity.md) identifies the focused pool,
 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
-definitions, selects an ID and optional version, and resolves file-backed
-message content within the selected operating-system or `fs.FS` source.
+`internal/promptdef` uses one source-neutral flow for prompt selection and
+normalization. That flow scans normalized YAML ID and version metadata,
+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
@@ -28,34 +40,56 @@ duplicate detection, and source containment:
 ## Profiles And Built-Ins
 
 `internal/profile` loads and validates execution profiles from an
-operating-system filesystem or an `fs.FS`. It supports a primary repository
-with fallback only when the primary reports that a profile is absent. Strict
-YAML decoding recognizes the optional `backend` field, trims its value, and
-requires a model plus at least one non-blank backend or endpoint. Loading does
-not check registry membership because the available registry belongs to the
+operating-system filesystem or an `fs.FS`. A file contains exactly one YAML
+document and its trimmed YAML `id` is its only selection identity; filenames do
+not confer authority. Each point lookup reads discovered files once for their
+metadata and reuses the selected file's bytes for strict decoding; unrelated
+profiles are not fully decoded. Strict selected decoding recognizes the
+optional `backend` field, trims its value, and requires a model plus at least
+one non-blank backend or endpoint. 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.
 
+The root engine assembles profile repositories in precedence order: in-memory
+profiles, one ordinary configured source, an application fallback source, then
+the embedded built-in catalog. An explicit file or `fs.FS` profile source
+replaces `Config.ProfileDir` within the ordinary configured-source category.
+
 Exact profile inspection performs one point-in-time lookup through those
 profile sources and checks the resolved target without reading prompt, input,
 or schema sources. It does not retain that lookup for a later execution.
 
-`internal/profile/builtin` embeds the maintained built-in profile catalog and
-can place a caller-selected repository ahead of that catalog. Every embedded
-profile selects `openrouter` and inherits its endpoint and credential
-environment-variable name from the built-in backend registry rather than
-repeating those values. Profile behavior is owned by the
-[profile repository tests](../../internal/profile/repository_test.go), while
-catalog completeness, the backend-selection invariant, duplicate IDs, and
-overlay behavior are owned by the
+`internal/profile/builtin` embeds the maintained built-in profile catalog.
+Every embedded profile selects `openrouter` and inherits its endpoint and
+credential environment-variable name from the built-in backend registry rather
+than repeating those values. Profile 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,
-caller-selected file paths. It copies content into an artifact, records
-metadata and a content hash, applies a content-type fallback, and honors
-context cancellation.
+`internal/artifact` accepts explicitly typed inline references even when their
+body is empty. It also resolves unrestricted, caller-selected paths only when
+they identify regular operating-system files, checking that condition before
+and after opening the file. It copies content into an artifact, records
+metadata and an opaque content-equality value, and applies a content-type
+fallback.
+
+Regular files are read synchronously in bounded chunks. 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 +101,17 @@ 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
-control into the rendered prompt. The
+artifacts and variables. Within one render, each referenced artifact body is
+converted to text lazily and cached by input name for reuse across the session
+and every message; the cache is not shared across renders. Conversion uses
+bounded chunks and preserves the artifact bytes exactly.
+
+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 +121,35 @@ 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
-plan. None, basic, and JSON modes retain the effective output contract without
-source access. JSON Schema mode loads the root document, resolves and compiles
-every transitive reference during preparation, and retains the compiled
-validator. The provider-facing structured-output metadata uses that same
-captured root document.
+Every preparation operation creates one operation-local validation plan. None,
+basic, and JSON modes retain the effective output contract without source
+access. JSON Schema mode loads the root document once, resolves and compiles
+each transitive reference, and retains the compiled validator. Schema compiler
+resources use canonical escaped file or private-scheme URLs; loaders decode
+their paths once and enforce the configured source boundary. The
+provider-facing structured-output metadata uses the root document captured by
+the same plan.
 
-`PrepareExecution` also completes prompt and profile selection, artifact
-loading and hashing, session and message rendering, and target resolution.
-`RunPrepared` uses the retained source-derived state and validation plan; it
-does not reopen prompt, profile, input, or schema sources and does not rerender
-the request. By contrast, ordinary `Prepare` produces a preparation value only:
-a later `Run` performs its own source resolution and preparation.
+Schema preparation and execution remain synchronous. Promptkit checks
+cancellation before and after source resolution, JSON decoding, compilation,
+and validation, and between bounded schema-read chunks. Once cancellation is
+observed it returns the context error without publishing a partial plan or
+validation result, even when a compiler or validator has just returned a
+different error or a successful result. An `fs.FS` method or JSON Schema
+dependency call already in progress cannot be preempted; Promptkit waits for
+that call to return and then gives cancellation precedence. Validation does
+not detach dependency work into background goroutines.
+
+`Prepare` discards its validation plan after returning metadata. `Run` retains
+its plan for initial and repaired-output validation, then discards it with the
+operation. `PrepareExecution` retains the plan in its private frozen payload;
+`RunPrepared` uses that plan without reopening 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

@@ -19,8 +19,8 @@ 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
-  components;
+- `internal/domain`, which owns framework data values and source-neutral
+  invariants shared by later internal components;
 - `internal/backend`, which owns validated immutable OpenAI-compatible backend
   definitions and the built-in OpenRouter definition;
 - `internal/capacity`, which owns engine-local bounded run admission and
@@ -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,
-documentation-link, and repository-hygiene checks before accepting changes.
+Maintainers run the complete local workflow in the
+[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.
-The maintained offline consumer workflow is also run from the repository root:
-
-```sh
-go test ./...
-go test -race ./...
-go run ./examples/go-library/prepare
-```
+Maintainer validation must include ordinary and race-enabled package tests,
+static analysis, a complete build, and execution of both maintained offline
+consumer examples. The preparation example protects assembled preparation and
+inspection behavior. The execution example separately protects assembled
+`Run`, injected-client, validation, usage, and result behavior.
 
 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
-  assembled consumer workflow without contacting a model provider.
+- The maintained offline preparation and execution examples protect distinct
+  representative assembled consumer workflows without contacting a model
+  provider.
 - Fixtures should be minimal, synthetic, versioned with the behavior they
   exercise, and free of credentials or private data.
 - Golden files are appropriate only when the complete output is intentionally

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
-[development guide](development.md):
-
-```sh
-go test ./...
-go test -race ./...
-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)"
-```
+As a release prerequisite, run the complete
+[maintainer validation workflow](development.md#maintainer-validation) against
+the clean candidate. Do not substitute a partial command list: the development
+guide owns the tests, race checks, analysis, build, both offline examples,
+formatting, Markdown links, generated-output and credential review, and
+repository hygiene. Record the successful workflow result with the candidate.
 
 ## 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
-resolves any selected backend. It does not load a prompt, render content,
-reserve capacity, or contact a model.
+Inspection applies the engine's profile source precedence and resolves any
+selected backend. It does not load a prompt, render content, reserve capacity,
+or contact a model.
 
 See the
 [profile-inspection consumer guide](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work)

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/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 +38,33 @@ consumers.
 
 ## Ideas
 
-No ideas currently await selection.
+### Public bounded output repair
+
+After the codebase-audit remediations are complete, Promptkit should make its
+bounded output-repair capability available through the public engine. A
+consumer should be able to request a limited number of corrective generation
+attempts when JSON or JSON Schema output fails content validation, without
+having to reproduce Promptkit's generation, validation, capacity, and result-
+accounting orchestration.
+
+- Repair is validation recovery, not a general provider retry, failover, or
+  backoff policy. Transport failures, cancellation, and operational schema or
+  validation errors must retain their ordinary error behavior.
+- Repair must stop after the first valid result or the configured attempt
+  bound. Exhausting the bound should preserve the final invalid result and its
+  validation diagnostics rather than inventing success.
+- Initial generation and every repair attempt must use the same resolved
+  backend, effective execution settings and presence semantics, session,
+  credential boundary, structured-output contract, and backend-capacity
+  policy.
+- Results should report the number of repair attempts and cumulative usage for
+  every model call made by the run.
+- Ordinary and prepared execution should expose coherent behavior, including
+  cancellation, frozen prepared state, error identity, and capacity lifetime.
+
+Select this work only after the accepted audit findings affecting shared
+execution invariants, validation, orchestration, transport, and repair
+internals have been remediated.
 
 ## 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/structured-generation-errors.md

@@ -0,0 +1,71 @@
+# Structured Generation Errors
+
+## Purpose
+
+Promptkit should give downstream applications actionable, machine-readable
+details when the built-in OpenAI-compatible client receives a non-success HTTP
+response. Today the client reports only the status code and discards the
+provider response body. This makes ordinary configuration failures—such as an
+unsupported strict JSON Schema keyword—unnecessarily difficult to diagnose.
+
+## Target End State
+
+Failures from the built-in transport are available through a public typed error
+that works with `errors.As` while continuing to match `ErrLLMGenerate` through
+`errors.Is`. The error should expose:
+
+- the HTTP status code;
+- a normalized provider error code or type when supplied; and
+- a bounded provider message extracted from a recognized OpenAI-compatible
+  JSON error envelope.
+
+The ordinary `Error()` string should remain safe and concise: it should include
+the status and provider code or type, but not automatically include the
+provider message. Consumers that deliberately want the provider's diagnostic
+text can retrieve it from the typed error and apply their own disclosure and
+logging policy.
+
+This contract should be available for both ordinary and prepared execution.
+Errors returned by injected model clients must continue to preserve their own
+identity and should not be converted into fabricated HTTP details.
+
+## Safety And Compatibility Boundaries
+
+- Never expose the raw response body, response headers, endpoint, credentials,
+  request messages, schema document, or generated content through this API.
+- Read only a small fixed maximum response body, reject malformed or
+  unrecognized envelopes, normalize invalid UTF-8 and control characters, and
+  cap every retained diagnostic field independently.
+- Treat the extracted provider message as untrusted and potentially sensitive:
+  its GoDoc must tell consumers not to log or display it without applying their
+  own policy.
+- Preserve the existing generic behavior when a response is empty, non-JSON,
+  oversized, or does not match a recognized error envelope.
+- Do not assign retryability from an HTTP status. Promptkit supplies facts;
+  downstream applications retain retry and presentation policy.
+
+## Recommended API Direction
+
+Prefer one immutable public `GenerationError` value, constructed internally and
+carrying accessors for HTTP status, provider code or type, and provider message.
+This keeps the exact representation evolvable while giving consumers an
+idiomatic `errors.As` contract. Public Go declarations and GoDoc should own the
+final exact names and semantics.
+
+The internal OpenAI-compatible client should parse only the conventional
+top-level `error` envelope and pass normalized details through the use-case and
+public error-mapping layers. The integration documentation should continue to
+own wire behavior; the public declarations should own the consumer contract.
+
+## Acceptance Criteria
+
+- A downstream consumer can distinguish a provider HTTP 400 from other
+  generation failures and obtain a bounded provider explanation when present.
+- The typed error still satisfies `errors.Is(err, ErrLLMGenerate)`.
+- Existing cancellation, capacity, validation, and injected-client error
+  identities remain unchanged.
+- Tests cover recognized string and numeric provider codes, absent and malformed
+  envelopes, oversized bodies and fields, control characters, and error-chain
+  behavior without making live provider requests.
+- Current-state GoDoc and the OpenAI-compatible integration and internal-client
+  documents are updated only when the implementation lands.

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

@@ -98,9 +98,10 @@ 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
-	// embedded built-in profiles. An empty value selects only built-ins unless
-	// profile options are also supplied.
+	// ProfileDir is an optional ordinary configured source whose profiles take
+	// precedence over application fallback and embedded built-in profiles. An
+	// 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 +120,12 @@ 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,
-// model-client, and artifact-reader category, the last non-nil valid option
-// replaces earlier options in that category. WithBackend is the additive
-// exception: unique registrations accumulate, and a repeated backend ID is an
-// error rather than a replacement. An invalid option fails construction even
-// if a later option would replace it.
+// each prompt-source, ordinary-profile-source, fallback-profile-source,
+// in-memory-profile, schema-source, model-client, and artifact-reader
+// category, the last non-nil valid option replaces earlier options in that
+// category. WithBackend is the additive exception: unique registrations
+// accumulate, and a repeated backend ID is an error rather than a replacement.
+// An invalid option fails construction even if a later option would replace it.
 type Option interface {
 	apply(*engineOptions) error
 }
@@ -136,18 +137,20 @@ func (f optionFunc) apply(options *engineOptions) error {
 }
 
 type engineOptions struct {
-	llmClient       llm.Client
-	artifactReader  artifactadapter.Reader
-	promptDefs      promptdef.Repository
-	profiles        profile.Repository
-	memoryProfiles  profile.Repository
-	backends        []domain.Backend
-	validator       validate.Validator
-	promptSource    bool
-	profileSource   bool
-	memorySource    bool
-	validatorSource bool
-	artifactSource  bool
+	llmClient             llm.Client
+	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
 }
 
 // WithLLMClient replaces the built-in model client used by [Engine.Run] and
@@ -216,7 +219,7 @@ 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 +227,12 @@ 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
-// api_key_env for environment-based credentials; raw API keys are rejected.
-// fsys must be non-nil and root must be non-empty; otherwise NewEngine fails
-// with ErrInvalidConfig. This option replaces Config.ProfileDir and earlier
-// file or FS profile-source options, but remains below WithProfiles in
-// precedence.
+// Profiles from this ordinary configured source take precedence over
+// application fallback and built-in profiles. Profile YAML must use api_key_env
+// for environment-based credentials; raw API keys are rejected. fsys must be
+// non-nil and root must be non-empty; otherwise NewEngine fails with
+// ErrInvalidConfig. This option replaces Config.ProfileDir and earlier file or
+// FS profile-source options, but remains below WithProfiles in precedence.
 func WithProfileFS(fsys fs.FS, root string) Option {
 	return optionFunc(func(options *engineOptions) error {
 		if fsys == nil {
@@ -246,11 +249,12 @@ 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
-// environment-based credentials; raw API keys are rejected. path must name an
-// existing non-directory file when NewEngine applies the option. This option
-// replaces Config.ProfileDir and earlier file or FS profile-source options,
-// but remains below WithProfiles in precedence.
+// The profile takes precedence over application fallback and built-in profiles.
+// Profile YAML must use api_key_env for environment-based credentials; raw API
+// keys are rejected. path must name an existing non-directory file when
+// NewEngine applies the option. This option replaces Config.ProfileDir and
+// earlier file or FS profile-source options, but remains below WithProfiles in
+// precedence.
 func WithProfileFile(path string) Option {
 	return optionFunc(func(options *engineOptions) error {
 		fsys, root, err := fileSource(path)
@@ -263,8 +267,41 @@ func WithProfileFile(path string) Option {
 	})
 }
 
+// WithFallbackProfileFS supplies application-owned fallback profile
+// definitions from fsys under root.
+//
+// Profile lookup checks, in order, profiles supplied by WithProfiles; the
+// ordinary configured source selected by WithProfileFile, WithProfileFS, or
+// Config.ProfileDir; this fallback source; and Promptkit's embedded built-in
+// 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
-// configured profile files and built-in profiles.
+// ordinary configured, application fallback, and built-in profiles.
 //
 // NewEngine validates and copies every profile. IDs must be unique within one
 // call. An invalid profile, duplicate ID, or unsupported ExtraParams value
@@ -350,13 +387,7 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 		promptDefs = promptdef.NewFilesystemRepository(cfg.PromptDir)
 	}
 
-	profiles := builtin.NewRepositoryWithDirectory(cfg.ProfileDir)
-	if options.profileSource {
-		profiles = builtin.NewRepositoryWithPrimary(options.profiles)
-	}
-	if options.memorySource {
-		profiles = profile.NewOverlayRepository(options.memoryProfiles, profiles)
-	}
+	profiles := newProfileRepository(cfg.ProfileDir, options)
 
 	backendRegistry, err := backend.NewRegistry(options.backends)
 	if err != nil {
@@ -409,22 +440,41 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
 	}, 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 repository
+}
+
 func fileSource(name string) (fs.FS, string, error) {
-	cleanName := strings.TrimSpace(name)
-	if cleanName == "" {
+	if strings.TrimSpace(name) == "" {
 		return nil, "", ErrInvalidConfig
 	}
-	dir := filepath.Dir(cleanName)
-	base := filepath.Base(cleanName)
-	if base == "." || base == string(filepath.Separator) || strings.TrimSpace(base) == "" {
+	dir := filepath.Dir(name)
+	base := filepath.Base(name)
+	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 +531,10 @@ 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
-// applies framework defaults, the selected backend, and then the selected
-// profile to EffectiveModelParams without a request override. BackendID is
-// empty for an endpoint-only profile.
+// in-memory, ordinary configured-source, application fallback, and built-in
+// profile precedence. It applies the framework timeout baseline, selected
+// backend, and then selected profile to EffectiveModelParams without a request
+// override. BackendID is empty for an endpoint-only profile.
 //
 // APIKeyEnv in the returned target is an environment-variable name, never its
 // value. APIKeyRequired instead reports a direct credential requirement and is

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) {
-	file, err := os.Open(path)
+func openArtifactFile(path string) (artifactFile, error) {
+	return os.Open(path)
+}
+
+func readFileArtifact(ctx context.Context, path string, open func(string) (artifactFile, error)) (*domain.Artifact, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	info, err := os.Stat(path)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+	}
+	if !info.Mode().IsRegular() {
+		return nil, fmt.Errorf("%w: %s", ErrUnsupportedFile, path)
+	}
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+
+	file, err := open(path)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
 	}
 	defer file.Close()
 
-	data, err := io.ReadAll(file)
+	openedInfo, err := file.Stat()
 	if err != nil {
-		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+		return nil, fmt.Errorf("failed to inspect opened file %s: %w", path, err)
+	}
+	if !openedInfo.Mode().IsRegular() {
+		return nil, fmt.Errorf("%w: %s", ErrUnsupportedFile, path)
+	}
+
+	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) {
+	_, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{
+		Type: domain.ArtifactRefType("unsupported"),
+		URI:  "unsupported://bucket/key",
+	})
+	if !errors.Is(err, ErrUnsupportedRefType) {
+		t.Fatalf("expected ErrUnsupportedRefType, got %v", err)
+	}
+}
+
+func TestCompositeReaderSourceParityAndOpaqueHashes(t *testing.T) {
 	reader := NewCompositeReader()
-	ctx := context.Background()
+	hashes := make(map[string]string)
+	tests := []struct {
+		name    string
+		content string
+	}{
+		{name: "empty", content: ""},
+		{name: "ordinary", content: "same content"},
+		{name: "changed", content: "changed content"},
+	}
 
-	t.Run("inline artifact", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefInline,
-			Body: "hello world",
-		}
-		art, err := reader.Read(ctx, ref)
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if string(art.Body) != "hello world" {
-			t.Errorf("expected 'hello world', got %s", string(art.Body))
-		}
-		if art.ContentType != "text/plain" {
-			t.Errorf("expected text/plain content type, got %q", art.ContentType)
-		}
-		if art.Hash != "b94d27b9934d3e08a52e52d7da7dabfac484efe37a5380ee9088f7ace2efcde9" {
-			t.Errorf("unexpected hash: %s", art.Hash)
-		}
-		if art.Size != int64(len(ref.Body)) {
-			t.Errorf("expected size %d, got %d", len(ref.Body), art.Size)
-		}
-	})
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			filePath := filepath.Join(t.TempDir(), "artifact.txt")
+			if err := os.WriteFile(filePath, []byte(tc.content), 0o600); err != nil {
+				t.Fatal(err)
+			}
 
-	t.Run("inline artifact missing body", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefInline,
-			Body: "",
-		}
-		_, err := reader.Read(ctx, ref)
-		if !errors.Is(err, ErrMissingInlineBody) {
-			t.Errorf("expected ErrMissingInlineBody, got %v", err)
-		}
-	})
+			sources := []struct {
+				name    string
+				ref     domain.ArtifactRef
+				wantURI string
+			}{
+				{
+					name: "inline",
+					ref:  domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: tc.content},
+				},
+				{
+					name:    "inline with uri",
+					ref:     domain.ArtifactRef{Type: domain.ArtifactRefInline, URI: "memory://input", Body: tc.content},
+					wantURI: "memory://input",
+				},
+				{
+					name:    "file",
+					ref:     domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filePath},
+					wantURI: filePath,
+				},
+			}
 
-	t.Run("unsupported ref type", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefType("unsupported"),
-			URI:  "unsupported://bucket/key",
-		}
-		_, err := reader.Read(ctx, ref)
-		if !errors.Is(err, ErrUnsupportedRefType) {
-			t.Error("expected error for unsupported type")
-		}
-	})
+			var sourceHash string
+			for _, source := range sources {
+				t.Run(source.name, func(t *testing.T) {
+					first, err := reader.Read(context.Background(), source.ref)
+					if err != nil {
+						t.Fatalf("first read: %v", err)
+					}
+					second, err := reader.Read(context.Background(), source.ref)
+					if err != nil {
+						t.Fatalf("second read: %v", err)
+					}
+					if string(first.Body) != tc.content || first.Size != int64(len(tc.content)) {
+						t.Fatalf("body=%q size=%d, want %q/%d", first.Body, first.Size, tc.content, len(tc.content))
+					}
+					if first.URI != source.wantURI {
+						t.Fatalf("URI = %q, want %q", first.URI, source.wantURI)
+					}
+					if first.Hash == "" || first.Hash != second.Hash {
+						t.Fatalf("hashes are not non-empty and stable: %q/%q", first.Hash, second.Hash)
+					}
+					if sourceHash == "" {
+						sourceHash = first.Hash
+					} else if first.Hash != sourceHash {
+						t.Fatalf("equal content hashes differ: %q/%q", sourceHash, first.Hash)
+					}
+					if source.ref.Type == domain.ArtifactRefFile {
+						if first.Name != filepath.Base(filePath) || !strings.HasPrefix(first.ContentType, "text/plain") {
+							t.Fatalf("unexpected file metadata: %+v", first)
+						}
+					} else if first.ContentType != "text/plain" {
+						t.Fatalf("inline content type = %q", first.ContentType)
+					}
+				})
+			}
+			hashes[tc.name] = sourceHash
+		})
+	}
+
+	if hashes["empty"] == hashes["ordinary"] || hashes["ordinary"] == hashes["changed"] {
+		t.Fatalf("changed content did not change opaque hash: %#v", hashes)
+	}
 }
 
 func TestCompositeReaderCopiesInlineData(t *testing.T) {
@@ -87,94 +131,154 @@ func TestCompositeReaderCopiesInlineData(t *testing.T) {
 	}
 }
 
-func TestCompositeReaderHonorsCancellation(t *testing.T) {
-	ctx, cancel := context.WithCancel(context.Background())
-	cancel()
+func TestCompositeReaderHonorsPreCancellation(t *testing.T) {
+	filePath := filepath.Join(t.TempDir(), "artifact.txt")
+	if err := os.WriteFile(filePath, []byte("ignored"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	tests := []struct {
+		name string
+		ref  domain.ArtifactRef
+	}{
+		{name: "inline", ref: domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: "ignored"}},
+		{name: "file", ref: domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filePath}},
+	}
 
-	_, err := NewCompositeReader().Read(ctx, domain.ArtifactRef{
-		Type: domain.ArtifactRefInline,
-		Body: "ignored",
-	})
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("expected context cancellation, got %v", err)
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			ctx, cancel := context.WithCancel(context.Background())
+			cancel()
+
+			artifact, err := NewCompositeReader().Read(ctx, tc.ref)
+			if artifact != nil || !errors.Is(err, context.Canceled) {
+				t.Fatalf("artifact=%#v err=%v, want nil/context.Canceled", artifact, err)
+			}
+		})
 	}
 }
 
-func TestFileReader_Read(t *testing.T) {
-	content := []byte("test file content")
-	filePath := filepath.Join(t.TempDir(), "artifact.txt")
-	if err := os.WriteFile(filePath, content, 0o600); err != nil {
-		t.Fatal(err)
-	}
-
+func TestFileReaderFailuresAndMetadata(t *testing.T) {
 	reader := NewCompositeReader()
-	ctx := context.Background()
-
-	t.Run("file artifact loading", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefFile,
-			URI:  filePath,
-		}
-		art, err := reader.Read(ctx, ref)
-		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
-		}
-		if string(art.Body) != string(content) {
-			t.Errorf("expected %s, got %s", string(content), string(art.Body))
-		}
-		if art.Name != filepath.Base(filePath) {
-			t.Errorf("expected name %q, got %q", filepath.Base(filePath), art.Name)
-		}
-		if !strings.HasPrefix(art.ContentType, "text/plain") {
-			t.Errorf("expected text content type, got %q", art.ContentType)
-		}
-		if art.URI != filePath {
-			t.Errorf("expected URI %q, got %q", filePath, art.URI)
-		}
-		if art.Size != int64(len(content)) {
-			t.Errorf("expected size %d, got %d", len(content), art.Size)
-		}
-		if art.Hash != "60f5237ed4049f0382661ef009d2bc42e48c3ceb3edb6600f7024e7ab3b838f3" {
-			t.Errorf("unexpected hash: %s", art.Hash)
-		}
-	})
 
 	t.Run("missing file path", func(t *testing.T) {
-		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefFile,
-			URI:  "",
-		}
-		_, err := reader.Read(ctx, ref)
+		_, err := reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefFile})
 		if !errors.Is(err, ErrMissingFilePath) {
-			t.Errorf("expected ErrMissingFilePath, got %v", err)
+			t.Fatalf("expected ErrMissingFilePath, got %v", err)
 		}
 	})
 
 	t.Run("missing file", func(t *testing.T) {
-		ref := domain.ArtifactRef{
+		_, err := reader.Read(context.Background(), domain.ArtifactRef{
 			Type: domain.ArtifactRefFile,
 			URI:  filepath.Join(t.TempDir(), "missing.txt"),
-		}
-		if _, err := reader.Read(ctx, ref); err == nil {
+		})
+		if err == nil {
 			t.Fatal("expected missing file error")
 		}
 	})
 
+	t.Run("directory rejected before open", func(t *testing.T) {
+		artifact, err := reader.Read(context.Background(), domain.ArtifactRef{
+			Type: domain.ArtifactRefFile,
+			URI:  t.TempDir(),
+		})
+		if artifact != nil || !errors.Is(err, ErrUnsupportedFile) {
+			t.Fatalf("artifact=%#v err=%v, want nil/ErrUnsupportedFile", artifact, err)
+		}
+	})
+
+	t.Run("non-regular opened target rejected", func(t *testing.T) {
+		filePath := filepath.Join(t.TempDir(), "artifact.txt")
+		if err := os.WriteFile(filePath, []byte("content"), 0o600); err != nil {
+			t.Fatal(err)
+		}
+		directoryInfo, err := os.Stat(t.TempDir())
+		if err != nil {
+			t.Fatal(err)
+		}
+		fileReader := &fileReader{open: func(path string) (artifactFile, error) {
+			file, err := os.Open(path)
+			if err != nil {
+				return nil, err
+			}
+			return &reportedInfoFile{artifactFile: file, info: directoryInfo}, nil
+		}}
+
+		artifact, err := fileReader.Read(context.Background(), domain.ArtifactRef{
+			Type: domain.ArtifactRefFile,
+			URI:  filePath,
+		})
+		if artifact != nil || !errors.Is(err, ErrUnsupportedFile) {
+			t.Fatalf("artifact=%#v err=%v, want nil/ErrUnsupportedFile", artifact, err)
+		}
+	})
+
 	t.Run("unknown extension uses text fallback", func(t *testing.T) {
-		path := filepath.Join(t.TempDir(), "artifact.unknownextension")
-		if err := os.WriteFile(path, content, 0o600); err != nil {
+		filePath := filepath.Join(t.TempDir(), "artifact.unknownextension")
+		if err := os.WriteFile(filePath, []byte("content"), 0o600); err != nil {
 			t.Fatal(err)
 		}
 
-		art, err := reader.Read(ctx, domain.ArtifactRef{
+		artifact, err := reader.Read(context.Background(), domain.ArtifactRef{
 			Type: domain.ArtifactRefFile,
-			URI:  path,
+			URI:  filePath,
 		})
 		if err != nil {
-			t.Fatalf("unexpected error: %v", err)
+			t.Fatalf("read artifact: %v", err)
 		}
-		if art.ContentType != "text/plain" {
-			t.Errorf("expected text/plain fallback, got %q", art.ContentType)
+		if artifact.ContentType != "text/plain" {
+			t.Fatalf("content type = %q", artifact.ContentType)
 		}
 	})
 }
+
+func 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"
@@ -109,10 +108,11 @@ func (r *Registry) CapacityPolicies() map[string]domain.BackendCapacityPolicy {
 }
 
 func normalizeBackend(definition domain.Backend) (domain.Backend, error) {
-	definition.Endpoint = strings.TrimSpace(definition.Endpoint)
-	if err := validateEndpoint(definition.Endpoint); err != nil {
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(definition.Endpoint)
+	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 +182,3 @@ 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/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
+	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"`
-	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"`
-	TargetPresence       ExecutionTargetPresence `json:"-"`
-	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             string
+	PromptVersion        string
+	PromptHash           string
+	SelectedProfileID    string
+	SelectedBackendID    string
+	EffectiveModelParams ExecutionTarget
+	TargetPresence       ExecutionTargetPresence
+	OutputContract       OutputContract
+	StructuredOutput     *StructuredOutputSpec
+	InputHashes          map[string]string
+	SessionID            string
+	RenderedPromptHash   string
+	Messages             []RenderedMessage
+	StartTime            time.Time
+	EndTime              time.Time
+	DurationMS           int64
 }
 
 // ArtifactRef represents a reference to an input artifact.

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 +1,30 @@
+package domain
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+)
+
+// 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")
+	}
+	return nil
+}

internal/domain/output_contract_test.go

@@ -0,0 +1,68 @@
+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.RepairAttempts = -1 }, wantErr: "repair_attempts"},
+		{name: "zero repair attempts", change: func(c *OutputContract) { c.RepairAttempts = 0 }},
+		{name: "positive repair attempts", change: func(c *OutputContract) { c.RepairAttempts = 1 }},
+		{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

@@ -7,10 +7,10 @@ import (
 
 func TestNormalizeSessionID(t *testing.T) {
 	tests := []struct {
-		name    string
-		raw     string
-		want    string
-		wantErr bool
+		name            string
+		raw             string
+		want            string
+		wantErrContains string
 	}{
 		{
 			name: "trims surrounding Unicode whitespace",
@@ -28,21 +28,24 @@ func TestNormalizeSessionID(t *testing.T) {
 			want: strings.Repeat("界", SessionIDMaxLength),
 		},
 		{
-			name:    "one Unicode code point over maximum is rejected",
-			raw:     strings.Repeat("界", SessionIDMaxLength+1),
-			wantErr: true,
+			name:            "one Unicode code point over maximum is rejected",
+			raw:             strings.Repeat("界", SessionIDMaxLength+1),
+			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") {
-					t.Fatalf("expected useful length diagnostic, got %v", err)
+				if !strings.Contains(err.Error(), tt.wantErrContains) {
+					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 root == "" || root == "." {
+	if strings.TrimSpace(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))
-	if cleanBase == "" {
+	cleanBase := path.Clean(baseDir)
+	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 cleanUserPath == "" {
+	if strings.TrimSpace(userPath) == "" {
 		return "", "", fmt.Errorf("path is required")
 	}
-	cleanUserPath = path.Clean(cleanUserPath)
-	if path.IsAbs(cleanUserPath) {
+	if path.IsAbs(userPath) {
 		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
-// trees used by configuration, request, and prepared-state boundaries.
+// Package jsonvalue validates and defensively copies bounded JSON-compatible
+// 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 {
-		if value.IsNil() {
-			return nil, nil
-		}
-		return copyValue(value.Elem(), path, seen, allowEmptyMapKeys)
-	}
+	value = resolved
 	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 != nil || math.IsNaN(f) || math.IsInf(f, 0) {
-			return nil, fmt.Errorf("%s: invalid JSON number", path)
+		if err := state.produceNode(path); err != nil {
+			return nil, err
 		}
 		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 {
-			return nil, fmt.Errorf("%s: integer is outside the JSON-safe range", path)
+		if err := state.produceNode(path); err != nil {
+			return nil, err
 		}
 		return value.Interface(), nil
 	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
-		if value.Uint() > maxSafeJSONInteger {
-			return nil, fmt.Errorf("%s: integer is outside the JSON-safe range", path)
+		if err := state.produceNode(path); err != nil {
+			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()}
-		if _, ok := seen[current]; ok {
-			return nil, fmt.Errorf("%s: cyclic value is not supported", path)
+		nextDepth, err := state.enterContainer(path, containerDepth)
+		if err != nil {
+			return nil, err
 		}
-		seen[current] = struct{}{}
-		defer delete(seen, current)
-		return copyValue(value.Elem(), path, seen, allowEmptyMapKeys)
-	case reflect.Map:
-		return copyMapValue(value, path, seen, allowEmptyMapKeys)
+		return copyMapValue(value, path, state, allowEmptyMapKeys, nextDepth)
 	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{}{}
-	defer delete(seen, current)
+	state.active[current] = struct{}{}
+	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{}{}
-		defer delete(seen, current)
+		state.active[current] = struct{}{}
+		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) {
-	nested := map[string]int{"limit": 2}
-	sequence := []string{"one", "two"}
-	input := map[string]any{
-		"count":    int64(7),
-		"number":   json.Number("-1.25e+2"),
-		"nested":   nested,
-		"sequence": sequence,
-	}
-
-	copied, err := jsonvalue.CopyMap(input)
-	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
+type (
+	namedBool    bool
+	namedString  string
+	namedInt64   int64
+	namedUint64  uint64
+	namedFloat32 float32
+	namedFloat64 float64
+	namedKey     string
+	namedMap     map[namedKey]namedInt64
+	namedSlice   []namedString
+	namedArray   [1]map[string]int
+)
 
+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: "non-string map key", value: map[int]string{1: "one"}},
-		{name: "unsupported value", value: make(chan int)},
-		{name: "cyclic map", value: cyclicMap},
-		{name: "cyclic slice", value: cyclicSlice},
-		{name: "NaN", value: math.NaN()},
-		{name: "positive infinity", value: math.Inf(1)},
-		{name: "unsafe signed integer", value: int64(1 << 53)},
-		{name: "unsafe unsigned integer", value: uint64(1 << 53)},
+		{name: "bool", value: true},
+		{name: "named bool", value: namedBool(true)},
+		{name: "string", value: "value"},
+		{name: "named string", value: namedString("value")},
+		{name: "int", value: minInt},
+		{name: "int8", value: int8(-1 << 7)},
+		{name: "int16", value: int16(-1 << 15)},
+		{name: "int32", value: int32(-1 << 31)},
+		{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) {
-	for _, number := range []json.Number{"0", "-1", "1.25", "-1.25e+2"} {
-		t.Run("valid "+number.String(), func(t *testing.T) {
-			got, err := jsonvalue.CopyMap(map[string]any{"value": number})
+func TestCopyPreservesCompatibleCollectionsAndNilEmptyDistinctions(t *testing.T) {
+	collections := []struct {
+		name  string
+		value any
+	}{
+		{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{}},
+	}
+
+	for _, tc := range collections {
+		t.Run(tc.name, func(t *testing.T) {
+			got, err := Copy(tc.value)
 			if err != nil {
-				t.Fatalf("copy valid JSON number: %v", err)
+				t.Fatalf("copy collection: %v", err)
 			}
-			if !reflect.DeepEqual(got["value"], number) {
-				t.Fatalf("JSON number changed: got %#v want %#v", got["value"], number)
+			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"} {
-		t.Run("invalid "+number.String(), func(t *testing.T) {
-			if _, err := jsonvalue.CopyMap(map[string]any{"value": number}); err == nil {
-				t.Fatal("expected invalid JSON number error")
+	var nilMap map[string]int
+	var nilSlice []string
+	var nilPointer *namedInt64
+	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)
-	if baseURL != "" {
-		if _, err := url.ParseRequestURI(baseURL); err != nil {
+	baseURL := ""
+	if strings.TrimSpace(cfg.BaseURL) != "" {
+		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 {
-		return nil, fmt.Errorf("%w: timeout_seconds must be greater than or equal to 0", ErrInvalidRequest)
+	if err := domain.ValidateExecutionTargetSettings(req.Target); err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
 	}
 
-	endpoint := strings.TrimSpace(req.Target.Endpoint)
-	if endpoint == "" {
-		endpoint = c.baseURL
+	selectedEndpoint := req.Target.Endpoint
+	if strings.TrimSpace(selectedEndpoint) == "" {
+		selectedEndpoint = c.baseURL
 	}
-	if endpoint == "" {
-		return nil, fmt.Errorf("%w: endpoint is required", ErrInvalidRequest)
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(selectedEndpoint)
+	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 {
@@ -130,7 +150,7 @@ 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()
 
@@ -138,10 +158,13 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 		_, _ = io.Copy(io.Discard, io.LimitReader(httpResp.Body, 4096))
 		return nil, fmt.Errorf("%w: status=%d", ErrUnexpectedStatus, httpResp.StatusCode)
 	}
+	if httpResp.ContentLength > maxOpenAIChatResponseBytes {
+		return nil, openAIChatResponseTooLargeError()
+	}
 
-	var wireResp openAIChatResponse
-	if err := json.NewDecoder(httpResp.Body).Decode(&wireResp); err != nil {
-		return nil, fmt.Errorf("%w: failed to decode response: %v", ErrMalformedResponse, err)
+	wireResp, err := decodeOpenAIChatResponse(httpResp.Body)
+	if err != nil {
+		return nil, err
 	}
 
 	if len(wireResp.Choices) == 0 {
@@ -164,6 +187,46 @@ 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 == "" {

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"
 )
 
@@ -91,53 +88,3 @@ 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/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 +96,50 @@ 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)
-		idMatch := fileMatch || metadata.id == id
+		metadata, metadataErr := readProfileFileMetadata(data)
+		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
 		}
-
-		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)
-			}
+		if !idMatch {
 			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)
-		if err := validateProfile(&prof); err != nil {
+		prof.ExtraParams, err = jsonvalue.CopyMap(prof.ExtraParams)
+		if err != nil {
+			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidProfile, relPath, err)
+		}
+		if err := normalizeAndValidateProfile(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 +165,36 @@ 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 {
-		return profileFileMetadata{}
+	if err := decoder.Decode(&node); err != nil {
+		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 +209,7 @@ 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 +217,68 @@ func readProfileFileMetadata(data []byte) profileFileMetadata {
 	return metadata
 }
 
-func validateProfile(p *domain.ExecutionProfile) error {
+func (m profileFileMetadata) matchesID(id string) bool {
+	for _, candidate := range m.ids {
+		if candidate == id {
+			return true
+		}
+	}
+	return false
+}
+
+func (m *profileFileMetadata) merge(other profileFileMetadata) {
+	m.ids = append(m.ids, other.ids...)
+	m.hasRawAPIKey = m.hasRawAPIKey || other.hasRawAPIKey
+}
+
+func decodeProfile(data []byte) (*domain.ExecutionProfile, error) {
+	var prof domain.ExecutionProfile
+	decoder := yaml.NewDecoder(bytes.NewReader(data))
+	decoder.KnownFields(true)
+	if err := decoder.Decode(&prof); err != nil {
+		return nil, err
+	}
+	if err := requireYAMLStreamEnd(decoder); err != nil {
+		return nil, err
+	}
+	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")
+}
+
+func normalizeAndValidateProfile(p *domain.ExecutionProfile) error {
 	if strings.TrimSpace(p.ID) == "" {
 		return errors.New("id is required")
 	}
-	if strings.TrimSpace(p.BackendID) == "" && strings.TrimSpace(p.Endpoint) == "" {
+	p.Endpoint = strings.TrimSpace(p.Endpoint)
+	if strings.TrimSpace(p.BackendID) == "" && p.Endpoint == "" {
 		return errors.New("backend or endpoint is required")
 	}
+	if p.Endpoint != "" {
+		endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(p.Endpoint)
+		if err != nil {
+			return err
+		}
+		p.Endpoint = endpoint
+	}
 	if strings.TrimSpace(p.Model) == "" {
 		return errors.New("model is required")
 	}
 
-	if p.Temperature < 0 || p.Temperature > 2 {
-		return errors.New("temperature must be between 0 and 2")
-	}
-	if p.MaxTokens < 0 {
-		return errors.New("max_tokens must be greater than or equal to 0")
-	}
-	if p.TopP < 0 || p.TopP > 1 {
-		return errors.New("top_p must be between 0 and 1")
-	}
-	if p.TimeoutSeconds < 0 {
-		return errors.New("timeout_seconds must be greater than or equal to 0")
-	}
-
-	return nil
+	return domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{
+		Temperature:    p.Temperature,
+		MaxTokens:      p.MaxTokens,
+		TopP:           p.TopP,
+		TimeoutSeconds: p.TimeoutSeconds,
+	})
 }

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) {
-			t.Fatalf("expected ErrInvalidYAML, got %v", err)
+		if !errors.Is(err, ErrProfileNotFound) {
+			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 +351,513 @@ 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 TestOverlayRepository(t *testing.T) {
 	ctx := context.Background()
 	primaryProfile := &domain.ExecutionProfile{ID: "shared", Endpoint: "http://primary", Model: "primary"}
@@ -495,6 +947,77 @@ 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/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) {
-			art, ok := inputs[name]
-			if !ok || art == nil {
-				return "", fmt.Errorf("%w: %s", ErrUnknownInput, name)
-			}
-			return string(art.Body), nil
-		},
+		"input": resolver.resolve,
 	}
 
-	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 {
-		case <-ctx.Done():
-			return nil, ctx.Err()
-		default:
+		if err := ctx.Err(); err != nil {
+			return nil, err
 		}
 
 		if tmplMsg.Role == "" {
 			return nil, fmt.Errorf("%w: message %d", ErrInvalidMessageRole, i)
 		}
 
-		// Parse and execute template
-		tmpl, err := template.New(fmt.Sprintf("msg_%d", i)).Funcs(funcs).Option("missingkey=error").Parse(tmplMsg.Content)
-		if err != nil {
-			return nil, fmt.Errorf("%w: message %d: %v", ErrInvalidTemplate, i, err)
+		if err := ctx.Err(); err != nil {
+			return nil, 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 {
-			return nil, fmt.Errorf("%w: message %d: %w", ErrRenderFailure, i, err)
+		executeErr := tmpl.Execute(&buf, vars)
+		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) {
-	tmpl, err := template.New("session_id").Funcs(funcs).Option("missingkey=error").Parse(raw)
-	if err != nil {
-		return "", fmt.Errorf("%w: session_id: %v", ErrInvalidTemplate, err)
+type artifactTextResolver struct {
+	ctx        context.Context
+	inputs     map[string]*domain.Artifact
+	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 {
-		return "", fmt.Errorf("%w: session_id: %w", ErrRenderFailure, err)
+	executeErr := tmpl.Execute(&buf, vars)
+	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 {
-	dir string
-}
-
-type fsRepository struct {
-	fsys fs.FS
-	root string
+type sourceRepository struct {
+	source promptDefinitionSource
 }
 
 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)
 	}
@@ -94,34 +114,25 @@ func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id strin
 		default:
 		}
 
-		relPath := filecatalog.RelativePath(r.dir, fullPath)
-		fileMatch := filecatalog.Stem(filepath.Base(fullPath)) == id
-
-		raw, err := loadPromptDefinitionFile(fullPath)
+		relPath := r.source.displayPath(fullPath)
+		data, err := r.source.readDefinition(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
 		}
-
-		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 {
+		if !promptDefinitionMatches(raw, id, version) {
 			continue
 		}
 		matches = append(matches, promptDefinitionMatch{
-			def:  def,
-			path: relPath,
+			raw:        raw,
+			sourcePath: fullPath,
+			path:       relPath,
 		})
 	}
 
@@ -137,130 +148,21 @@ 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
-	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
+	raw        *promptDefinitionFile
+	sourcePath string
+	path       string
 }
 
 func decodePromptDefinition(data []byte) (*promptDefinitionFile, error) {
@@ -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"`
+		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) {
-	promptDir := filepath.Dir(sourcePath)
-	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
-		resolvedPath := strings.TrimSpace(contentFile)
-		if !filepath.IsAbs(resolvedPath) {
-			resolvedPath = filepath.Join(promptDir, resolvedPath)
-		}
-		resolvedPath = filepath.Clean(resolvedPath)
-
-		body, err := os.ReadFile(resolvedPath)
-		if err != nil {
-			return "", "", err
-		}
-		return string(body), resolvedPath, nil
-	})
+func promptDefinitionMatches(raw *promptDefinitionFile, id string, version string) bool {
+	if raw == nil {
+		return false
+	}
+	return promptSelectorMatches(raw.ID, raw.Version, id, version)
 }
 
-func normalizePromptDefinitionFromFS(raw *promptDefinitionFile, fsys fs.FS, root string, sourcePath string, rootIsDir bool) (*domain.PromptDefinition, error) {
-	promptDir := path.Dir(sourcePath)
-	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
-		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), "/")
-		}
+func promptSelectorMatches(rawID string, rawVersion string, id string, version string) bool {
+	if strings.TrimSpace(rawID) != id {
+		return false
+	}
+	return version == "" || strings.TrimSpace(rawVersion) == version
+}
 
-		body, err := fs.ReadFile(fsys, resolvedPath)
-		if err != nil {
-			return "", "", err
-		}
-		return string(body), resolvedPath, nil
+func normalizePromptDefinition(raw *promptDefinitionFile, sourceRoot contentSourceRoot, sourcePath string) (*domain.PromptDefinition, error) {
+	return normalizePromptDefinitionWithContent(raw, func(contentFile string) (string, string, error) {
+		return sourceRoot.readContentFile(sourcePath, contentFile)
 	})
 }
 
@@ -402,17 +289,14 @@ func normalizePromptDefinitionWithContent(raw *promptDefinitionFile, readContent
 		})
 	}
 
-	if !isValidOutputFormat(raw.Output.Format) {
-		return nil, fmt.Errorf("invalid output format: %q", raw.Output.Format)
+	outputContract := domain.OutputContract{
+		Format:         raw.Output.Format,
+		ValidationMode: raw.Output.ValidationMode,
+		SchemaPath:     strings.TrimSpace(raw.Output.SchemaPath),
+		RepairAttempts: raw.Output.RepairAttempts,
 	}
-	if !isValidValidationMode(raw.Output.ValidationMode) {
-		return nil, fmt.Errorf("invalid validation mode: %q", raw.Output.ValidationMode)
-	}
-	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")
+	if err := domain.ValidateOutputContract(outputContract); err != nil {
+		return nil, fmt.Errorf("output: %w", err)
 	}
 
 	defaultProfile := ""
@@ -432,12 +316,7 @@ func normalizePromptDefinitionWithContent(raw *promptDefinitionFile, readContent
 		Inputs:         inputs,
 		Templates:      templates,
 		OutputFormat:   raw.Output.Format,
-		Validation: domain.OutputContract{
-			Format:         raw.Output.Format,
-			ValidationMode: raw.Output.ValidationMode,
-			SchemaPath:     strings.TrimSpace(raw.Output.SchemaPath),
-			RepairAttempts: raw.Output.RepairAttempts,
-		},
+		Validation:     outputContract,
 	}, 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) {
-			t.Fatalf("expected resolved content_file path to be absolute, got %q", p.Templates[1].ContentFile)
+		if filepath.IsAbs(p.Templates[1].ContentFile) != contentPathsAreFull {
+			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: "missing id", id: "missing_id", targetErr: ErrInvalidPromptDefinition, errSubstrs: []string{"id 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: "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: "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: "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: "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 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: "unidentifiable invalid YAML is unrelated", id: "invalid-yaml", targetErr: ErrPromptDefinitionNotFound},
+		{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: "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: "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: "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: "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: "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: "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) {
@@ -418,64 +447,611 @@ 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) {
-	repo := NewFSRepository(fstest.MapFS{
-		"one.yaml": &fstest.MapFile{Data: []byte(`
-id: duplicate-fs-prompt
-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
-`)},
-	}, ".")
+type recordingFS struct {
+	fs.FS
+	mu     sync.Mutex
+	opened []string
+}
 
-	_, err := repo.GetPromptDefinition(context.Background(), "duplicate-fs-prompt", "")
-	if !errors.Is(err, ErrInvalidPromptDefinition) {
-		t.Fatalf("expected ErrInvalidPromptDefinition, got %v", err)
+func TestPromptRepositoryReturnsDefinitionReadFailures(t *testing.T) {
+	readErr := errors.New("definition read failed")
+	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") {
-		t.Fatalf("expected duplicate paths in error, got %v", err)
+	repo := NewFSRepository(fsys, "prompts")
+
+	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) {
-	repo := NewFSRepository(fstest.MapFS{
-		"not_named_like_id.yaml": &fstest.MapFile{Data: []byte(`
-id: strict-fs-prompt
-version: "1.0.0"
-unknown: true
+type definitionReadFailureFS struct {
+	fs.FS
+	target string
+	err    error
+}
+
+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", "")
-	if !errors.Is(err, ErrInvalidYAML) {
-		t.Fatalf("expected ErrInvalidYAML, got %v", err)
+	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: "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)
+				}
+			}
+		})
 	}
 }
 

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 +1,185 @@
+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: "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(
-		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)
+	executionSnapshot, err := clonePreparedRun(operation.run)
 	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

@@ -52,6 +52,12 @@ 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,
@@ -164,6 +170,77 @@ func TestRunnerPrepareExecutionCompletesWithoutAdmissionOrGeneration(t *testing.
 	}
 }
 
+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 TestRunnerRunPreparedRechecksEnvironmentCredentialBeforeAdmission(t *testing.T) {
 	const environmentName = "PROMPTKIT_PREPARED_EXECUTION_TEST_KEY"
 	t.Setenv(environmentName, "available-during-preparation")
@@ -271,7 +348,13 @@ 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()
@@ -282,7 +365,13 @@ func TestRunnerRunPreparedUsesFrozenValidationForInitialAndRepairOutputs(t *test
 		nil,
 		reader,
 		renderer,
-		&fakeLLM{resp: &domain.GenerateResponse{Content: `{"broken":true}`}},
+		&fakeLLM{resp: &domain.GenerateResponse{
+			Content: `{"broken":true}`,
+			Usage: domain.TokenUsage{
+				PromptTokens: 13, CompletionTokens: 17, TotalTokens: 19,
+				CachedTokens: 23, CacheWriteTokens: 29,
+			},
+		}},
 		validator,
 		repairer,
 		admitter,
@@ -309,6 +398,13 @@ func TestRunnerRunPreparedUsesFrozenValidationForInitialAndRepairOutputs(t *test
 	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)
 	}
@@ -492,7 +588,7 @@ 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 {
-	if strings.TrimSpace(target.Endpoint) == "" {
-		return errors.New("execution endpoint is required")
+func normalizeResolvedExecutionTarget(target domain.ExecutionTarget) (domain.ExecutionTarget, error) {
+	endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(target.Endpoint)
+	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

@@ -19,6 +19,7 @@ type RepairRequest struct {
 	ValidationErrors []string
 	SessionID        string
 	Target           domain.ExecutionTarget
+	TargetPresence   domain.ExecutionTargetPresence
 	StructuredOutput *domain.StructuredOutputSpec
 	Attempt          int
 	MaxAttempts      int
@@ -44,7 +45,6 @@ func (r *defaultOutputRepairer) Repair(ctx context.Context, req RepairRequest) (
 	}
 
 	prompt := domain.RenderedPrompt{
-		SessionID: req.SessionID,
 		Messages: []domain.RenderedMessage{
 			{
 				Role:    "system",
@@ -64,11 +64,13 @@ func (r *defaultOutputRepairer) Repair(ctx context.Context, req RepairRequest) (
 		},
 	}
 
-	resp, err := r.llm.Generate(ctx, domain.GenerateRequest{
-		Prompt:           prompt,
-		Target:           req.Target,
-		StructuredOutput: req.StructuredOutput,
-	})
+	resp, err := r.llm.Generate(ctx, newGenerationRequest(
+		prompt,
+		req.SessionID,
+		req.Target,
+		req.TargetPresence,
+		req.StructuredOutput,
+	))
 	if err != nil {
 		return nil, err
 	}

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{
-		Prompt:           domain.RenderedPrompt{SessionID: prepared.SessionID, Messages: prepared.Messages},
-		Target:           executionTarget,
-		TargetPresence:   prepared.TargetPresence,
-		StructuredOutput: prepared.StructuredOutput,
-	})
+	genResp, err := r.llm.Generate(ctx, newGenerationRequest(
+		domain.RenderedPrompt{Messages: prepared.Messages},
+		prepared.SessionID,
+		executionTarget,
+		prepared.TargetPresence,
+		prepared.StructuredOutput,
+	))
 	if err != nil {
 		if errors.Is(err, llm.ErrInvalidRequest) {
 			return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
 		}
 		return nil, fmt.Errorf("%w: %w", ErrLLMGenerate, err)
 	}
+	usage := genResp.Usage
 
 	outputArtifact := buildOutputArtifact(genResp.Content, prepared.OutputContract.Format)
 	validationResult, err := validateArtifact(ctx, &outputArtifact, 0)
@@ -197,6 +207,7 @@ func (r *Runner) executePreparedRun(
 				ValidationErrors: validationResult.Errors,
 				SessionID:        prepared.SessionID,
 				Target:           executionTarget,
+				TargetPresence:   prepared.TargetPresence,
 				StructuredOutput: prepared.StructuredOutput,
 				Attempt:          attemptsUsed,
 				MaxAttempts:      prepared.OutputContract.RepairAttempts,
@@ -210,6 +221,7 @@ func (r *Runner) executePreparedRun(
 			}
 
 			genResp = repairResp
+			usage = addTokenUsage(usage, repairResp.Usage)
 			outputArtifact = buildOutputArtifact(genResp.Content, prepared.OutputContract.Format)
 
 			validationResult, err = validateArtifact(ctx, &outputArtifact, attemptsUsed)
@@ -238,19 +250,33 @@ 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 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 +298,10 @@ 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 +316,16 @@ func (r *Runner) resolvePreparation(
 		return nil, err
 	}
 
-	effectiveModel, targetPresence, err := resolveExecutionTarget(selection.backend, selection.profile, req.Execution)
-	if err != nil {
-		return nil, fmt.Errorf("%w: %w", ErrInvalidRequest, err)
-	}
+	effectiveModel, targetPresence := resolveExecutionTarget(selection.backend, selection.profile, req.Execution)
 	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 +342,68 @@ func (r *Runner) completePreparation(
 	ctx context.Context,
 	req domain.RunRequest,
 	state *preparationState,
-) (*domain.PreparedRun, error) {
-	structuredOutput, err := r.resolveStructuredOutput(
-		ctx,
+) (*preparedOperation, error) {
+	validationPlan, err := r.prepareValidation(ctx, state.effectiveContract)
+	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 +477,6 @@ 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 +514,6 @@ 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
@@ -527,7 +569,7 @@ 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 +579,18 @@ 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
 	}
@@ -576,22 +606,18 @@ 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, err = mergeExecutionTargetOverride(out, *override)
-		if err != nil {
-			return domain.ExecutionTarget{}, domain.ExecutionTargetPresence{}, err
-		}
+		out, presence = mergeExecutionTargetOverride(out, *override)
 	}
-	return out, presence, nil
+	return out, presence
 }
 
 func validateAPIKey(apiKeyEnv string, apiKey string, apiKeyRequired bool) error {
@@ -658,18 +684,21 @@ 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
+		}
 	}
-	if contract.Format == "" {
-		contract.Format = domain.FormatText
+	if err := domain.ValidateOutputContract(contract); err != nil {
+		return domain.OutputContract{}, err
 	}
-	return contract
+	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) {
-	f.schemaLoads++
-	f.schemaLoadPath = schemaPath
-	if f.schemaErr != nil {
-		return nil, f.schemaErr
+func (f *fakeValidator) PrepareValidation(_ context.Context, contract domain.OutputContract) (validate.PreparedValidation, error) {
+	var schemaDocument any
+	if contract.ValidationMode == domain.ValidationJSONSchema {
+		f.schemaLoads++
+		f.schemaLoadPath = contract.SchemaPath
+		if f.schemaErr != nil {
+			return nil, f.schemaErr
+		}
+		schemaDocument = f.schemaDoc
+		if schemaDocument == nil {
+			schemaDocument = map[string]any{"type": "object"}
+		}
 	}
-	if f.schemaDoc != nil {
-		return f.schemaDoc, nil
-	}
-	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
@@ -502,6 +546,35 @@ 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 +777,23 @@ 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 +1770,6 @@ 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)}
@@ -2044,50 +2094,255 @@ func TestRunnerRunValidationStillWorks(t *testing.T) {
 	}
 }
 
-func TestRunnerRunStructuredRepairRemainsBoundedAndUsesEffectiveModelSettings(t *testing.T) {
-	repairer := &fakeRepairer{responses: []*domain.GenerateResponse{{Content: `{"broken":`}, {Content: `{"still":`}}}
-	llmClient := &fakeLLM{resp: &domain.GenerateResponse{Content: `{"initial":`}}
+func TestRunnerRepairStateMachine(t *testing.T) {
+	failed := func(mode domain.ValidationMode, diagnostic string) domain.ValidationResult {
+		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,
+	}
 
-	runner := NewRunnerWithRepairer(
-		&fakePromptRepo{def: promptDef(domain.FormatJSON, domain.ValidationJSON, 1)},
-		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
-			"exec": {ID: "exec", BackendID: "custom", Model: "profile-model", TimeoutSeconds: 55},
-		}}, fakeBackendResolver{backends: map[string]domain.Backend{
-			"custom": {ID: "custom", Endpoint: "http://backend/v1"},
-		}},
+	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:              "basic failure is ineligible despite budget",
+			mode:              domain.ValidationBasic,
+			budget:            3,
+			validationResults: []domain.ValidationResult{failed(domain.ValidationBasic, "empty output")},
+			responses:         responses(1),
+			wantStatus:        domain.ValidationFailed,
+		},
+		{
+			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: 4,
+			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,
+		},
+	}
 
-		defaultArtifactReader(),
-		defaultRenderer(),
-		llmClient,
-		validate.NewStandardValidator("."),
-		repairer, nil)
+	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: definition},
+				&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{
+					"exec": {ID: "exec", BackendID: "custom", Model: "profile-model"},
+				}},
+				fakeBackendResolver{backends: map[string]domain.Backend{
+					"custom": {ID: "custom", Endpoint: "http://backend.example/v1"},
+				}},
+				defaultArtifactReader(),
+				defaultRenderer(),
+				client,
+				validator,
+				repairer,
+				nil,
+			)
 
-	res, err := runner.Run(context.Background(), domain.RunRequest{
-		PromptID:  "p",
-		ProfileID: "exec",
-		Inputs:    singleInputRef(),
-		Execution: &domain.ExecutionTargetOverride{Endpoint: "http://override/v1", Model: "override-model", TimeoutSeconds: intPtr(22)},
-	})
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
-	if repairer.calls != 1 || res.Validation.RepairAttempts != 1 {
-		t.Fatalf("expected one bounded repair, calls=%d attempts=%d", repairer.calls, res.Validation.RepairAttempts)
-	}
-	if len(repairer.reqs) != 1 {
-		t.Fatalf("expected one repair request, got %d", len(repairer.reqs))
-	}
-	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)
-	}
-	if repairer.reqs[0].Target.TimeoutSeconds != 22 {
-		t.Fatalf("expected repair to use effective timeout, got %d", repairer.reqs[0].Target.TimeoutSeconds)
-	}
-	if llmClient.lastReq.Target.BackendID != "custom" ||
-		repairer.reqs[0].Target.BackendID != "custom" ||
-		res.SelectedBackendID != "custom" {
-		t.Fatalf("expected backend identity in generation, repair, and result: generate=%q repair=%q result=%q",
-			llmClient.lastReq.Target.BackendID, repairer.reqs[0].Target.BackendID, res.SelectedBackendID)
+			result, err := runner.Run(context.Background(), domain.RunRequest{
+				PromptID:  "p",
+				ProfileID: "exec",
+				SessionID: " repair-session ",
+				APIKey:    "direct-secret",
+				Inputs:    singleInputRef(),
+				Execution: tc.execution,
+			})
+			if err != nil {
+				t.Fatalf("run: %v", err)
+			}
+			if len(client.requests) != tc.wantRepairs+1 || len(repairer.reqs) != tc.wantRepairs {
+				t.Fatalf(
+					"generation/repair calls = (%d, %d), want (%d, %d)",
+					len(client.requests), len(repairer.reqs), tc.wantRepairs+1, tc.wantRepairs,
+				)
+			}
+			if len(plan.artifacts) != tc.wantRepairs+1 {
+				t.Fatalf("validation calls = %d, want %d", len(plan.artifacts), tc.wantRepairs+1)
+			}
+
+			initialRequest := client.requests[0]
+			if initialRequest.TargetPresence != tc.wantPresence {
+				t.Fatalf("initial target presence = %+v, want %+v", initialRequest.TargetPresence, tc.wantPresence)
+			}
+			if initialRequest.Target.APIKey != "direct-secret" || initialRequest.Target.BackendID != "custom" ||
+				initialRequest.Prompt.SessionID != "repair-session" {
+				t.Fatalf("initial common request fields = %+v", initialRequest)
+			}
+			if initialRequest.Target.Endpoint != "http://backend.example/v1" ||
+				initialRequest.Target.Model != "profile-model" ||
+				initialRequest.Target.Temperature != 0 || initialRequest.Target.MaxTokens != 0 || initialRequest.Target.TopP != 0 {
+				t.Fatalf("initial effective target = %+v", initialRequest.Target)
+			}
+			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 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)
+				}
+				if reflect.DeepEqual(generated.Prompt.Messages, initialRequest.Prompt.Messages) {
+					t.Fatalf("repair generation request %d reused the initial prompt", index)
+				}
+			}
+
+			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 +2437,6 @@ 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 +2495,7 @@ func TestResolveExecutionTargetProfileValuesPopulateAllSupportedFields(t *testin
 		},
 	}
 
-	target, presence, err := resolveExecutionTarget(nil, profileValue, nil)
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
+	target, presence := resolveExecutionTarget(nil, profileValue, nil)
 	if presence != (domain.ExecutionTargetPresence{}) {
 		t.Fatalf("expected no request override presence, got %+v", presence)
 	}
@@ -2385,10 +2547,7 @@ func TestResolveExecutionTargetRuntimeOverridesBeatProfileForAllOverrideableFiel
 		},
 	}
 
-	target, presence, err := resolveExecutionTarget(nil, profileValue, override)
-	if err != nil {
-		t.Fatalf("expected no error, got %v", err)
-	}
+	target, presence := resolveExecutionTarget(nil, profileValue, override)
 	if presence != (domain.ExecutionTargetPresence{Temperature: true, MaxTokens: true, TopP: true, TimeoutSeconds: true}) {
 		t.Fatalf("unexpected override presence: %+v", presence)
 	}
@@ -2435,12 +2594,9 @@ 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 +2725,7 @@ func TestResolveExecutionTargetUsesBackendProfileAndRequestPrecedence(t *testing
 		ExtraParams: map[string]any{"request": true},
 	}
 
-	target, _, err := resolveExecutionTarget(backendValue, profileValue, override)
-	if err != nil {
-		t.Fatalf("resolve target: %v", err)
-	}
+	target, _ := resolveExecutionTarget(backendValue, profileValue, override)
 	if target.BackendID != "custom" {
 		t.Fatalf("endpoint override changed backend identity: %+v", target)
 	}
@@ -2583,12 +2736,9 @@ 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 []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
+	return executeJSONSchema(ctx, p.schema, instance)
 }
 
 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 := compiler.AddResource(resolvedSchemaPath, schemaDocument); err != nil {
+	if err := ctx.Err(); 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)})
-	if err := compiler.AddResource(resourceURL, schemaDocument); err != nil {
+	compiler := newSchemaCompiler(fsSchemaLoader{ctx: ctx, fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
+	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 {
@@ -165,7 +189,11 @@ 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 +203,32 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 		res.IsValid = true
 		return res, nil
 	case domain.ValidationJSON:
-		_, jsonErr := parseJSON(artifact.Body)
-		if jsonErr != nil {
+		valid := json.Valid(artifact.Body)
+		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 +247,8 @@ func validateArtifact(ctx context.Context, artifact *domain.Artifact, contract d
 	}
 }
 
-func (v *StandardValidator) validateJSONSchema(instance any, schemaPath string) ([]string, error) {
-	resolvedSchemaPath, err := v.resolveSchemaPath(schemaPath)
+func (v *StandardValidator) validateJSONSchema(ctx context.Context, instance any, schemaPath string) ([]string, error) {
+	resolvedSchemaPath, err := v.resolveSchemaPath(ctx, schemaPath)
 	if err != nil {
 		return nil, err
 	}
@@ -223,20 +257,21 @@ func (v *StandardValidator) validateJSONSchema(instance any, schemaPath string)
 	if err != nil {
 		return nil, err
 	}
-	compiler := newSchemaCompiler(standardSchemaLoader{root: schemaRoot})
-	schema, err := compiler.Compile(resolvedSchemaPath)
+	if err := ctx.Err(); err != nil {
+		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 []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
+	return executeJSONSchema(ctx, schema, instance)
 }
 
-func (v *FSValidator) validateJSONSchema(instance any, schemaPath string) ([]string, error) {
-	schemaName, schemaDoc, err := v.loadSchemaDocument(schemaPath)
+func (v *FSValidator) validateJSONSchema(ctx context.Context, instance any, schemaPath string) ([]string, error) {
+	schemaName, schemaDoc, err := v.loadSchemaDocument(ctx, schemaPath)
 	if err != nil {
 		return nil, err
 	}
@@ -245,71 +280,60 @@ 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)})
-	if err := compiler.AddResource(resourceURL, schemaDoc); err != nil {
+	compiler := newSchemaCompiler(fsSchemaLoader{ctx: ctx, fsys: v.fsys, root: filecatalog.CleanFSRoot(v.root)})
+	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 []string{fmt.Sprintf("json schema validation failed: %v", err)}, nil
-	}
-	return nil, nil
+	return executeJSONSchema(ctx, schema, instance)
 }
 
-func parseJSON(body []byte) (any, error) {
-	var v any
-	if err := json.Unmarshal(body, &v); err != nil {
+func decodeJSONValue(ctx context.Context, body []byte) (any, error) {
+	if err := ctx.Err(); 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) {
-	select {
-	case <-ctx.Done():
-		return nil, ctx.Err()
-	default:
-	}
-
-	resolved, err := v.resolveSchemaPath(schemaPath)
-	if err != nil {
+	var value any
+	decodeErr := decoder.Decode(&value)
+	if err := ctx.Err(); err != nil {
 		return nil, err
 	}
-
-	raw, err := os.ReadFile(resolved)
-	if err != nil {
-		return nil, fmt.Errorf("failed to read schema file %q: %w", resolved, err)
+	if decodeErr != nil {
+		return nil, decodeErr
 	}
 
-	var doc any
-	if err := json.Unmarshal(raw, &doc); err != nil {
-		return nil, fmt.Errorf("failed to decode JSON schema %q: %w", resolved, err)
-	}
-	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 {
+	var trailing any
+	trailingErr := decoder.Decode(&trailing)
+	if err := ctx.Err(); 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 +342,25 @@ 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 +381,39 @@ func (v *StandardValidator) schemaRoot() (string, error) {
 	return resolved, nil
 }
 
-func (v *FSValidator) loadSchemaDocument(schemaPath string) (string, any, error) {
-	resolved, err := v.resolveSchemaPath(schemaPath)
+func (v *FSValidator) loadSchemaDocument(ctx context.Context, schemaPath string) (string, any, error) {
+	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
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	doc, err := decodeJSONValue(ctx, raw)
+	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 +424,14 @@ 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 +445,21 @@ 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 +475,19 @@ func cleanSchemaFSPath(schemaPath string) (string, error) {
 	return cleaned, nil
 }
 
-func fsSchemaResourceURL(schemaName string) string {
-	return "promptkit-schema:///" + strings.TrimPrefix(path.Clean(schemaName), "/")
+func fileSchemaResourceURL(schemaName string) *url.URL {
+	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 +497,35 @@ 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 +546,33 @@ 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 +598,94 @@ func containedFilesystemPath(root, name string) (string, error) {
 	return candidate, nil
 }
 
-func loadJSONSchemaFile(name string) (any, error) {
-	raw, err := os.ReadFile(name)
+func loadJSONSchemaFile(ctx context.Context, name string) (any, error) {
+	raw, err := readSchemaFile(ctx, func() (fs.File, error) {
+		return os.Open(name)
+	})
 	if err != nil {
 		return nil, err
 	}
-	var doc any
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	doc, err := decodeJSONValue(ctx, raw)
+	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, "/"))
-	if err != nil {
-		return nil, fmt.Errorf("invalid schema reference %q: %w", resourceURL, err)
-	}
+	name := strings.TrimPrefix(parsed.Path, "/")
 	name = path.Clean(name)
 	if l.root == "." {
 		if strings.HasPrefix(name, "../") || name == ".." {
@@ -528,21 +697,35 @@ 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
-	if err := json.Unmarshal(raw, &doc); err != nil {
+	doc, err := decodeJSONValue(l.ctx, raw)
+	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"
@@ -93,6 +95,51 @@ 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 +332,6 @@ 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 +414,70 @@ func TestFSValidatorPreparedSchemaSurvivesSourceMutation(t *testing.T) {
 	}
 }
 
-func TestFSValidatorJSONSchemaRegistrationError(t *testing.T) {
-	v := NewFSValidator(fstest.MapFS{
-		"schemas/%zz.json": &fstest.MapFile{Data: []byte(`{"type":"object"}`)},
-	}, "schemas")
+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/" + schemaName: &fstest.MapFile{Data: []byte(`{"type":"object"}`)},
+			}, "schemas")
 
-	res, err := v.Validate(context.Background(), &domain.Artifact{Body: []byte(`{}`)}, domain.OutputContract{
+			res, err := v.Validate(context.Background(), &domain.Artifact{Body: []byte(`{}`)}, domain.OutputContract{
+				ValidationMode: domain.ValidationJSONSchema,
+				SchemaPath:     schemaName,
+			})
+			if err != nil || !res.IsValid {
+				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:     "%zz.json",
+		SchemaPath:     schemaPath,
 	})
-	if err == nil || !strings.Contains(err.Error(), "failed to register JSON schema") {
-		t.Fatalf("expected schema registration error, got result=%#v error=%v", res, err)
+	if err != nil || !result.IsValid {
+		t.Fatalf("validate schema %q: result=%#v error=%v", schemaPath, result, err)
 	}
 }
 
@@ -516,25 +567,6 @@ 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 +661,225 @@ 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 +924,8 @@ func TestJSONSchemaDialectIsDraft2020(t *testing.T) {
 func assertSchemaDocument(t *testing.T, got any, expectedJSON []byte) {
 	t.Helper()
 
-	var expected any
-	if err := json.Unmarshal(expectedJSON, &expected); err != nil {
+	expected, err := decodeJSONValue(context.Background(), expectedJSON)
+	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) {
@@ -21,38 +28,11 @@ func (r PreparedRun) MarshalJSON() ([]byte, error) {
 		durationMS = &r.DurationMS
 	}
 
-	return json.Marshal(struct {
-		PromptID             string                `json:"prompt_id"`
-		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,
+	return json.Marshal(preparedRunJSON{
+		preparedRunJSONFields: preparedRunJSONFields(r),
+		StartTime:             startTime,
+		EndTime:               endTime,
+		DurationMS:            durationMS,
 	})
 }
 
@@ -74,89 +54,57 @@ func (r RunResult) MarshalJSON() ([]byte, error) {
 	}
 
 	return json.Marshal(runResultJSON{
-		RunID:                r.RunID,
-		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,
+		runResultJSONFields: runResultJSONFields(r),
+		StartTime:           startTime,
+		EndTime:             endTime,
+		DurationMS:          durationMS,
 	})
 }
 
 // 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{
-		RunID:                wire.RunID,
-		Artifact:             wire.Artifact,
-		RawOutput:            wire.RawOutput,
-		Validation:           wire.Validation,
-		PromptID:             wire.PromptID,
-		PromptVersion:        wire.PromptVersion,
-		PromptHash:           wire.PromptHash,
-		SessionID:            wire.SessionID,
-		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,
+	result := RunResult(wire.runResultJSONFields)
+	if wire.DurationMS != nil {
+		if *wire.DurationMS < minDurationMilliseconds || *wire.DurationMS > maxDurationMilliseconds {
+			return fmt.Errorf(
+				"decode RunResult duration_ms: %d cannot be represented as time.Duration",
+				*wire.DurationMS,
+			)
+		}
+		result.Duration = time.Duration(*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 {
-	RunID                string            `json:"run_id"`
-	Artifact             Artifact          `json:"artifact"`
-	RawOutput            string            `json:"raw_output"`
-	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"`
+type preparedRunJSONFields PreparedRun
+
+type preparedRunJSON struct {
+	preparedRunJSONFields
+	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 {
-	if value == nil {
-		return 0
-	}
-	return *value
+type runResultJSONFields RunResult
+
+type runResultJSON struct {
+	runResultJSONFields
+	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 +1,48 @@
+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)
+	}
+
+	req := promptkit.RunRequest{
+		PromptID: "prompt",
+		Validation: &promptkit.OutputContract{
+			Format:         promptkit.OutputFormat("binary"),
+			ValidationMode: promptkit.ValidationNone,
+		},
+	}
+
+	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

@@ -409,14 +409,35 @@ func TestPreparedExecutionDiscardAndFormattingDoNotExposePrivateState(t *testing
 		t.Fatalf("prepare execution: %v", err)
 	}
 
-	formattedValues := []string{
-		fmt.Sprint(prepared),
-		fmt.Sprintf("%+v", prepared),
-		fmt.Sprintf("%#v", prepared),
+	copied := *prepared
+	zeroValue := promptkit.PreparedExecution{}
+	var nilHandle *promptkit.PreparedExecution
+	for name, value := range map[string]any{
+		"original pointer": prepared,
+		"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("%s %s formatting = %q, want opaque representation", name, format, formatted)
+			}
+			assertPreparedPrivateValuesAbsent(t, formatted, directCredential, renderedContent)
+		}
 	}
-	for _, formatted := range formattedValues {
-		if formatted != "promptkit.PreparedExecution{opaque}" {
-			t.Fatalf("unexpected opaque formatting: %q", formatted)
+	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 +454,9 @@ func TestPreparedExecutionDiscardAndFormattingDoNotExposePrivateState(t *testing
 	}
 	assertPreparedPrivateValuesAbsent(t, string(detailsJSON), directCredential)
 
-	prepared.Discard()
-	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,
-	})
+	executionResult, err := engine.RunPrepared(context.Background(), &copied)
 	if err != nil {
-		t.Fatalf("prepare execution for request inspection: %v", err)
-	}
-	executionResult, err := engine.RunPrepared(context.Background(), executed)
-	if err != nil {
-		t.Fatalf("run execution for request inspection: %v", err)
+		t.Fatalf("run copied execution after formatting: %v", err)
 	}
 	requests := client.snapshot()
 	if len(requests) != 1 || requests[0].APIKey != directCredential {
@@ -478,16 +481,34 @@ 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) {

profiles.go

@@ -100,33 +100,34 @@ func toDomainProfile(publicProfile Profile) (domain.ExecutionProfile, error) {
 		APIKeyRequired:  publicProfile.APIKeyRequired,
 		ExtraParams:     extraParams,
 	}
-	if err := validatePublicProfile(prof); err != nil {
+	if err := normalizeAndValidatePublicProfile(&prof); err != nil {
 		return domain.ExecutionProfile{}, err
 	}
 	return prof, nil
 }
 
-func validatePublicProfile(prof domain.ExecutionProfile) error {
+func normalizeAndValidatePublicProfile(prof *domain.ExecutionProfile) error {
 	if strings.TrimSpace(prof.ID) == "" {
 		return errors.New("id is required")
 	}
-	if strings.TrimSpace(prof.BackendID) == "" && strings.TrimSpace(prof.Endpoint) == "" {
+	prof.Endpoint = strings.TrimSpace(prof.Endpoint)
+	if strings.TrimSpace(prof.BackendID) == "" && prof.Endpoint == "" {
 		return errors.New("backend or endpoint is required")
 	}
+	if prof.Endpoint != "" {
+		endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(prof.Endpoint)
+		if err != nil {
+			return err
+		}
+		prof.Endpoint = endpoint
+	}
 	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
+	return domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{
+		Temperature:    prof.Temperature,
+		MaxTokens:      prof.MaxTokens,
+		TopP:           prof.TopP,
+		TimeoutSeconds: prof.TimeoutSeconds,
+	})
 }

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}
@@ -757,6 +779,7 @@ 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"},
@@ -818,92 +841,6 @@ func TestBackendExtraParamsAreDeeplyCopiedAtConstructionAndLookup(t *testing.T)
 	}
 }
 
-func TestPreparedRunJSONTimingRoundTrips(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"},
@@ -973,6 +910,24 @@ 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 +1012,227 @@ 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
-	// backend, profile, and then request overrides. It excludes resolved API-key
-	// values.
+	// EffectiveModelParams contains settings resolved from the framework timeout
+	// baseline, selected backend, profile, and then request overrides. Unset
+	// 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
-	// ArtifactRefFile.
+	// Body is the content for ArtifactRefInline, where an empty value is valid,
+	// 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
-	// the limit unspecified to compatible providers unless it was an explicit
-	// request override.
+	// MaxTokens is the non-negative resolved output-token limit. Zero leaves
+	// the limit unspecified to compatible providers unless the corresponding
+	// 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.
@@ -329,9 +337,11 @@ 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
-	// backend and then the profile, without a request override. APIKeyEnv is an
-	// environment-variable name, never its credential value.
+	// EffectiveModelParams contains settings resolved from the framework timeout
+	// baseline, selected backend, and then profile, without a request override.
+	// 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 +392,28 @@ 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
-// any, and framework defaults.
+// empty ExtraParams map inherit lower-precedence values. An optional provider
+// 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.
@@ -413,7 +432,8 @@ type ExecutionTargetOverride struct {
 	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
 }
 
@@ -425,9 +445,12 @@ type ExecutionTargetOverride struct {
 // 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
-// zero, blank strings, and an empty ExtraParams map inherit framework defaults;
-// use ExecutionTargetOverride pointer fields to request explicit numeric zero.
+// WithProfiles validates and copies Profile values during NewEngine. 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
@@ -437,23 +460,24 @@ type Profile struct {
 	BackendID string
 	// Endpoint is the model-provider base URL. It is required only when
 	// BackendID is blank and otherwise overrides the backend endpoint when
-	// non-blank.
+	// 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 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
-	// selecting an explicit zero.
+	// TopP is from 0 through 1. Zero leaves the provider control unspecified
+	// 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
-	// default.
+	// ReasoningEffort is optional; a blank value leaves it unspecified.
 	ReasoningEffort string
 	// APIKeyRequired clears a backend's inherited API-key environment name and
 	// requires a non-blank RunRequest.APIKey unless the request explicitly
@@ -461,7 +485,8 @@ type Profile struct {
 	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
 }
 
@@ -523,8 +548,8 @@ type ExecutionTargetPresence struct {
 // does not merge fields. The public Engine validates generated output once and
 // does not install an output repairer.
 type OutputContract struct {
-	// Format selects generated artifact metadata. An empty effective value
-	// defaults to FormatText.
+	// Format selects generated artifact metadata. An empty value in a non-nil
+	// request replacement defaults to FormatText.
 	Format OutputFormat `json:"format"`
 	// ValidationMode selects the content check. Use one of the declared
 	// ValidationMode constants.
@@ -532,8 +557,8 @@ 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
-	// no repairs. The public Engine performs no repairs even when this value is
+	// RepairAttempts is a non-negative requested repair limit. Zero requests no
+	// repairs. The public Engine performs no repairs even when this value is
 	// positive, so its runs report zero attempts used.
 	RepairAttempts int `json:"repair_attempts"`
 }
@@ -679,8 +704,12 @@ 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
-// path without restricting it to an application root or imposing a size limit.
+// The default artifact reader accepts path only when it resolves to a regular
+// 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 +717,13 @@ 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}
 }