Finish cleanup roadmap follow-through

docs/internal/adapters.md

@@ -105,6 +105,7 @@ Validator:
 HTTP error mapping:
 
 - maps domain/use-case errors to stable HTTP code + error code/message.
+- distinguishes missing profile selection and missing `api_key_env` variable using stable use-case sentinel errors.
 - avoids returning internal wrapped-cause details in response payload.
 
 ## CLI Adapter Semantics
@@ -140,4 +141,4 @@ Behavior highlights:
 - Adapter packages do not own runner decision logic.
 - External request/response strictness is part of contract stability.
 - Prepared-render output never includes resolved API key values.
-- Outbound OpenAI-compatible request includes only currently serialized fields (`model`, `messages`, optional `temperature`, `max_tokens`, `top_p`, optional `response_format`).
+- Outbound OpenAI-compatible request includes only currently serialized fields (`model`, `messages`, optional `temperature`, `max_tokens`, `top_p`, optional `service_tier`, optional `response_format`).

docs/internal/runner.md

@@ -66,14 +66,21 @@ It receives fully constructed repositories/readers/validators from adapters. Eff
 
 ## Failure Behavior
 
-Key error classes surfaced from `Runner`:
+Primary runner error classes:
 
-- `ErrInvalidRequest`: invalid prompt/profile/request/runtime/API-key-env prerequisites.
+- `ErrInvalidRequest`: invalid run request envelope.
-- `ErrProfileLoad`: prompt or profile load failures.
+- `ErrProfileRequired`: specific invalid-request reason when neither request `profile_id` nor prompt `default_profile` is available.
+- `ErrAPIKeyEnvMissing`: specific invalid-request reason when `api_key_env` is set but the named environment variable is unset/empty.
+- `ErrProfileLoad`: prompt/profile repository load failures.
 - `ErrArtifactLoad`: artifact read failures.
 - `ErrPromptRender`: template render failures.
-- `ErrLLMGenerate`: model request failures.
+- `ErrLLMGenerate`: outbound model request failures.
-- `ErrValidation`: validation runtime failures (including schema load/compile failures).
+- `ErrValidation`: validation runtime failures (including structured-output schema load/compile failures).
+
+Reason sentinel behavior:
+
+- `ErrProfileRequired` and `ErrAPIKeyEnvMissing` are wrapped with `ErrInvalidRequest`.
+- Adapters can use `errors.Is` for stable reason mapping without matching runner prose.
 
 Validation content failures are not run errors:
 
@@ -90,13 +97,15 @@ Validation content failures are not run errors:
 3. select profile ID:
    - explicit request profile ID
    - prompt `default_profile`
-   - otherwise request error
+   - otherwise return an invalid request with `ErrProfileRequired`
 4. load execution profile.
 5. merge effective runtime target:
    - built-in execution defaults
    - selected profile values
    - request overrides
-6. verify required `api_key_env` environment variable (name only; value is not returned).
+6. verify required `api_key_env` environment variable:
+   - missing/empty env value returns an invalid request with `ErrAPIKeyEnvMissing`
+   - only the environment-variable name is retained; secret value is never returned
 7. resolve output contract and structured-output schema payload when `json_schema` mode is active.
 8. read input artifacts.
 9. render prompt messages.

docs/roadmap/audit.md

@@ -1,519 +0,0 @@
-# Code Quality And Deduplication Audit
-
-## Executive Summary
-
-Overall code quality is solid for the current project size. The repository follows the documented shape: thin CLI/HTTP adapters, a central `Runner`, strict config/prompt/profile loading, narrow filesystem and LLM adapters, and package tests around the important public behaviors.
-
-The top three refactoring targets are:
-
-1. centralize execution-target field mapping across profiles, CLI overrides, HTTP DTOs, prepared output, and LLM request serialization;
-2. share filesystem repository scanning helpers between prompt and profile repositories;
-3. reduce CLI command wiring duplication for app settings, runner construction, and common preflight behavior.
-
-The codebase appears ready for a limited cleanup pass. No major architectural rewrite is warranted. The main risk before a release is maintenance drift when adding fields or policy to execution settings, repositories, or command setup.
-
-## Repository Map Reviewed
-
-Main areas inspected:
-
-- `cmd/scriptorium`: process entrypoint path, via repository layout and command docs.
-- `internal/adapter/cli`: `run`, `render`, `serve` parsing, config loading, app wiring, stdout/stderr behavior, summary output, exit-code policy.
-- `internal/adapter/http`: request DTOs, strict JSON decoding, domain mapping, response metadata mapping, error mapping.
-- `internal/config`: app config loading, defaults, strict YAML decoding, CLI override precedence.
-- `internal/defaults`: built-in runtime/config/content-type defaults.
-- `internal/domain`: domain request/result/profile/target/artifact/validation types.
-- `internal/usecase`: `Runner.Prepare`, `Runner.Run`, execution target merge, schema structured-output setup, validation and repair policy.
-- `internal/promptdef`: recursive prompt YAML loading, prompt normalization, content-file resolution, prompt output-contract validation.
-- `internal/profile`: recursive profile YAML loading, strict decoding, duplicate ID handling, profile validation, raw API-key rejection.
-- `internal/artifact`: `inline` and `file` artifact resolution.
-- `internal/prompt`: Go template rendering and required input checks.
-- `internal/llm`: OpenAI-compatible request construction, timeout/API-key handling, response parsing.
-- `internal/validate`: basic/JSON/JSON Schema validation and schema document loading.
-- `internal/format`: prepared-run text/JSON output.
-- `examples/`: maintained example config, prompts, profiles, schemas, fixtures, HTTP/render examples.
-- Package tests under `internal/**`.
-
-Major execution paths reviewed:
-
-- CLI `run`: parse flags/config, build `RunRequest`, construct runner, call LLM, write artifact, print summary, choose exit code.
-- CLI `render`: parse flags/config, build `RunRequest`, construct runner without LLM, prepare only, format prepared run.
-- CLI `serve`: parse flags/config, construct runner, expose HTTP `POST /v1/runs`.
-- HTTP `POST /v1/runs`: strict decode DTO, map to `RunRequest`, map `RunResult` to JSON response.
-- Runner `Prepare` and `Run`: prompt/profile/artifact/schema resolution, render, LLM generate, validation, repair hook.
-
-Areas not deeply inspected:
-
-- No `internal/app`, `internal/stage`, `internal/modules`, `internal/storage`, `internal/manifest`, `pkg`, or external test-suite directories exist in the current repository.
-- Full test execution was not run because this is a report-only pass; inspection was static.
-
-## High-Confidence Deduplication Opportunities
-
-### 1. Execution Target Field Mapping Is Repeated Across Boundaries
-
-Affected files/packages:
-
-- `internal/domain/domain.go`
-- `internal/usecase/runner.go`
-- `internal/adapter/cli/run.go`
-- `internal/adapter/http/dto.go`
-- `internal/adapter/http/handler.go`
-- `internal/llm/openai_compatible_client.go`
-- `internal/format/prepared_run.go`
-- tests in `internal/usecase`, `internal/adapter/http`, `internal/adapter/cli`, `internal/llm`, and `internal/format`
-
-Duplicated or near-duplicated behavior:
-
-- Execution settings fields are listed and copied in several places: profile-to-target conversion, target merge, CLI runtime override construction, HTTP request mapping, HTTP metadata mapping, prepared-run text output, and OpenAI-compatible request construction.
-- Adding `service_tier` required coordinated edits across many of these sites, which is a strong signal that field-level mapping is too scattered.
-
-Why it matters:
-
-- A future profile/runtime field can easily be parsed but not sent, displayed but not merged, accepted over HTTP but not included in metadata, or tested in one command but not another.
-- This is public-interface drift risk because CLI, HTTP, render output, and provider requests all expose different slices of the same effective execution target.
-
-Recommended refactor:
-
-- Keep transport DTOs local to adapters, but add small mapping helpers at each boundary:
-  - a use-case/domain helper for profile-to-target copy and execution-target merge;
-  - an HTTP helper such as `executionTargetFromModelOverrideDTO` and `modelParamsDTOFromExecutionTarget`;
-  - an LLM helper such as `openAIChatRequestFromGenerateRequest` for outbound serialization.
-- Add a table-driven test that constructs an `ExecutionTarget` with every supported field and verifies the HTTP metadata DTO and LLM request payload retain the intended fields.
-- Avoid reflection-based generic copying; explicit mapping is still clearer here.
-
-Suggested tests:
-
-- Expand runner merge tests to use an all-fields target.
-- Add HTTP DTO mapping tests that fail when a domain execution field is omitted from request or metadata mapping.
-- Add LLM request serialization tests for all serialized fields and explicit omission of non-serialized fields.
-
-Risk level: high. The refactor itself is low-to-medium implementation risk, but the duplicated behavior has high drift risk.
-
-### 2. Prompt And Profile Filesystem Repository Scanning Is Duplicated
-
-Affected files/packages:
-
-- `internal/promptdef/filesystem_repository.go`
-- `internal/profile/filesystem_repository.go`
-- repository tests in `internal/promptdef` and `internal/profile`
-
-Duplicated or near-duplicated behavior:
-
-- Both repositories recursively walk configured directories, filter `.yaml`/`.yml`, sort paths, compute relative paths, derive likely IDs from filenames, use strict YAML decoding, partially decode IDs to decide whether to surface malformed likely-target files, detect duplicate IDs, and include relative paths in errors.
-
-Why it matters:
-
-- Prompt/profile repository policy should remain aligned: nested scanning, stable ordering, duplicate ID errors, likely-target malformed file behavior, and relative-path diagnostics.
-- Future changes to repository scanning or extension rules would need to be made in two packages.
-
-Recommended refactor:
-
-- Introduce a narrow internal helper for filesystem catalog behavior, for example `internal/filecatalog` or another small package whose scope is only:
-  - recursive YAML file discovery with context cancellation;
-  - stable sorting;
-  - relative path formatting;
-  - YAML extension checks;
-  - filename stem extraction.
-- Keep prompt/profile-specific normalization and validation in their current packages.
-- Do not make a generic repository framework.
-
-Suggested tests:
-
-- Add helper-level tests for recursive YAML discovery, ordering, extension filtering, and relative path output.
-- Keep existing prompt/profile behavior tests unchanged to confirm public error behavior survives.
-
-Risk level: high for drift prevention; low implementation risk if the helper stays small.
-
-### 3. CLI Command Setup And Runner Wiring Are Repeated
-
-Affected files/packages:
-
-- `internal/adapter/cli/run.go`
-- `internal/adapter/cli/run_test.go`
-
-Duplicated or near-duplicated behavior:
-
-- `run`, `render`, and `serve` each construct similar runner dependencies.
-- `run` and `serve` both create an OpenAI-compatible client with default timeout.
-- `run`, `render`, and `serve` all resolve app settings, clean dirs, validate required prompt/profile dirs, and create filesystem repositories/validator/renderer/readers.
-- `run` and `render` share most execution request flags and request construction, with command-specific differences around `--schema-dir`, `--format`, and LLM use.
-
-Why it matters:
-
-- Adding or changing a dependency, default, or preflight check can drift between commands.
-- The current duplication is still readable, but it is large enough that future command additions or option changes will be error-prone.
-
-Recommended refactor:
-
-- Add a small CLI-local wiring helper, such as `newRunnerFromDirs(promptDir, profileDir, schemaDir string, llmClient llm.Client) *usecase.Runner`.
-- Add a small CLI-local resolved settings struct for common `prompt_dir`, `profile_dir`, `schema_dir`, and render-format handling.
-- Preserve the existing command-specific parse functions and flag surfaces; do not hide command behavior behind a generic command framework.
-
-Suggested tests:
-
-- Keep parser tests for each command.
-- Add one test that verifies `run`, `render`, and `serve` use the same configured prompt/profile/schema dirs by exercising config-derived dirs.
-- Keep command-level success tests for `run` and `render`.
-
-Risk level: medium. Behavior is public, but a small CLI-local helper can be behavior-preserving.
-
-### 4. HTTP Error Mapping Depends On Error Message Substrings
-
-Affected files/packages:
-
-- `internal/usecase/runner.go`
-- `internal/adapter/http/handler.go`
-- `internal/adapter/http/handler_test.go`
-
-Duplicated or near-duplicated behavior:
-
-- Runner creates `ErrInvalidRequest` with human-readable details for cases such as missing profile selection and missing API-key env.
-- HTTP maps some specific invalid-request cases by checking `strings.Contains(err.Error(), ...)`.
-
-Why it matters:
-
-- User-facing HTTP error codes can drift if a runner error message is clarified.
-- This crosses package boundaries in a brittle way: HTTP should depend on stable error identity, not exact prose from `internal/usecase`.
-
-Recommended refactor:
-
-- Add narrow sentinel errors or typed invalid-request reasons in `internal/usecase`, for example profile-required and API-key-env-missing.
-- Keep HTTP response messages stable and adapter-owned.
-- Do not expose HTTP-specific error codes from the runner.
-
-Suggested tests:
-
-- HTTP error mapping tests should assert `profile_required` and `api_key_env_missing` via sentinel wrapping, not via message matching.
-- Runner tests should assert `errors.Is` for the new reason errors.
-
-Risk level: high for public API stability; low-to-medium implementation risk.
-
-## Medium-Confidence Opportunities
-
-### 1. Strict YAML Decode Setup Is Repeated
-
-Affected files/packages:
-
-- `internal/config/config.go`
-- `internal/promptdef/filesystem_repository.go`
-- `internal/profile/filesystem_repository.go`
-
-Duplicated or near-duplicated behavior:
-
-- Each package constructs a YAML decoder and enables `KnownFields(true)`.
-
-Semantic differences that may be intentional:
-
-- Prompt/profile loaders need likely-target behavior and raw `api_key` handling.
-- Config loading has its own explicit/implicit file search policy.
-
-Why it matters:
-
-- Strict decoding is an architectural invariant. A future YAML loader could forget to enable it.
-
-Recommended refactor:
-
-- Consider a tiny helper for strict YAML decode from bytes.
-- Keep package-specific error wrapping and partial ID decode logic local.
-
-Suggested tests:
-
-- Existing unknown-field tests in config, promptdef, and profile should remain.
-- Add any new YAML-consuming package with an unknown-field test.
-
-Risk level: medium.
-
-### 2. Schema Path Resolution Is Centralized, But Schema Loading Happens Twice
-
-Affected files/packages:
-
-- `internal/usecase/runner.go`
-- `internal/validate/standard_validator.go`
-
-Duplicated or near-duplicated behavior:
-
-- For `json_schema`, `Runner.Prepare` asks the validator to load the schema document for provider-level structured output.
-- Later validation resolves and compiles the same schema path again.
-
-Semantic differences that may be intentional:
-
-- Provider structured-output payload needs the raw JSON schema document.
-- Runtime validation needs a compiled schema.
-
-Why it matters:
-
-- The current behavior is correct, but schema file errors can surface at different phases and the same file is read more than once during `Run`.
-- Future caching or schema behavior changes should have one clear owner.
-
-Recommended refactor:
-
-- Keep path resolution inside `internal/validate`.
-- Consider a schema service/loader interface that can return both raw document and compiled schema from one path, only if schema-related work grows.
-- Do not add caching unless repeated schema loads become a measured cost.
-
-Suggested tests:
-
-- Keep existing nested schema path tests.
-- Add a test that `Prepare` fails before LLM call when structured-output schema cannot load.
-
-Risk level: medium.
-
-### 3. Artifact Hashing And Output Artifact Construction Are Split
-
-Affected files/packages:
-
-- `internal/artifact/reader.go`
-- `internal/usecase/runner.go`
-
-Duplicated or near-duplicated behavior:
-
-- Input artifacts and generated output artifacts both compute SHA-256 hashes and fill size/body/content-type fields.
-
-Semantic differences that may be intentional:
-
-- Input artifacts derive content type from source extension or inline default.
-- Output artifacts derive content type from prompt output format and always use the default output artifact name.
-
-Why it matters:
-
-- Hash algorithm and artifact metadata policy should remain consistent.
-
-Recommended refactor:
-
-- Consider an `artifact.Build` or `artifact.HashBody` helper only for shared hash/size construction.
-- Keep source-specific content type and naming decisions local.
-
-Suggested tests:
-
-- Existing artifact reader tests and runner output content-type tests should cover behavior.
-- Add a small helper test if a shared hash function is introduced.
-
-Risk level: medium-low.
-
-### 4. Domain Contains An Unsupported `s3` Artifact Reference Constant
-
-Affected files/packages:
-
-- `internal/domain/domain.go`
-- `internal/artifact/reader.go`
-- docs that state only `inline` and `file` are implemented
-
-Duplicated or near-duplicated behavior:
-
-- Not a duplication issue. This is a cleanup issue: `domain.ArtifactRefS3` exists, but no reader supports it and docs do not document it as implemented.
-
-Semantic differences that may be intentional:
-
-- The constant may be a placeholder for a future adapter.
-
-Why it matters:
-
-- Future-facing code outside `docs/roadmap/` can confuse maintainers and tests. It also weakens the otherwise clean "implemented behavior only" policy.
-
-Recommended refactor:
-
-- Remove the constant if no near-term S3 implementation is planned.
-- If kept, add a code comment that it is intentionally unsupported today and ensure docs continue to state only `inline` and `file` are supported.
-
-Suggested tests:
-
-- Existing unsupported artifact type tests should continue to pass.
-
-Risk level: medium-low.
-
-### 5. Test Fixture Setup Is Repeated In CLI And Use-Case Tests
-
-Affected files/packages:
-
-- `internal/adapter/cli/run_test.go`
-- `internal/usecase/runner_test.go`
-- `internal/usecase/integration_test.go`
-
-Duplicated or near-duplicated behavior:
-
-- Tests repeatedly construct temp prompt/profile dirs, write prompt/profile YAML, set up fake LLMs/readers/renderers/validators, and build runners.
-
-Semantic differences that may be intentional:
-
-- Package-local tests avoid exporting test helpers and keep each package independent.
-
-Why it matters:
-
-- Repeated setup makes behavior-preserving refactors noisier and can obscure the exact behavior under test.
-
-Recommended refactor:
-
-- Add package-local helper builders where duplication is highest, especially in CLI command tests.
-- Avoid a cross-package test utility package unless multiple packages need the same public fixture contract.
-
-Suggested tests:
-
-- This is test cleanup only; existing test assertions should remain equivalent.
-
-Risk level: low.
-
-## Boundary And Responsibility Concerns
-
-- HTTP error mapping currently depends on runner error message text. Stable reason identity should live in `internal/usecase`; HTTP status/code/message mapping should remain in `internal/adapter/http`.
-- Execution-target merge policy lives in `internal/usecase`, which fits the architecture. The concern is not placement but incomplete centralization of field mapping around that policy.
-- CLI app wiring currently constructs concrete repositories/readers/renderers/validators inline in each command. That is acceptable adapter responsibility, but repeated wiring should be centralized within the CLI adapter package.
-- Prompt/profile filesystem traversal is duplicated in two repository packages. A narrow filesystem catalog helper would fit the architecture because it would not own prompt/profile policy.
-- LLM provider request shape is properly isolated in `internal/llm`. It should remain there; do not move OpenAI/OpenRouter request-field policy into runner or prompt definitions.
-
-## Path, Key, And Naming Construction Review
-
-Path and naming construction is mostly explicit and low-risk:
-
-- Config paths are cleaned in `internal/config` and again in CLI finalization.
-- Prompt `content_file` paths are resolved relative to the prompt YAML file in `internal/promptdef`.
-- Schema paths are resolved through `internal/validate.StandardValidator`.
-- OpenAI-compatible endpoint paths use `defaults.OpenAIChatCompletionsPath`.
-- Output artifact name comes from `defaults.OutputArtifactName`.
-
-Areas needing cleanup:
-
-- Prompt/profile recursive YAML discovery and relative-path formatting should share one helper.
-- CLI path cleaning after config resolution is repeated and should be consolidated with common command settings finalization.
-- Structured schema names are derived in `internal/usecase`; this is currently one place and should stay there unless structured-output support expands.
-
-No remote keys, cache paths, manifest paths, lock files, or generated report paths exist in the current implementation.
-
-## Resolution And Catalog Review
-
-Named concept resolution is mostly consistent:
-
-- Prompt resolution uses YAML `id` and optional `version`, not file path.
-- Profile resolution uses YAML `id`, not file path.
-- Prompt/profile subdirectories are organizational only.
-- Duplicate prompt/profile IDs fail instead of choosing first match.
-- Schema resolution is explicit path-based relative to `schema_dir`; no basename search.
-- Input artifacts are resolved by `artifact.Reader`, with `inline` and `file` as implemented types.
-- Prompt required-input checks happen in the renderer.
-
-Recommended centralization:
-
-- Share only filesystem catalog mechanics between prompt/profile repositories.
-- Keep prompt ID/version, profile ID, schema path, and input resolution policies in their current owning packages.
-
-## Config And Command-Loading Review
-
-Config precedence is consistent with policy:
-
-1. built-in defaults;
-2. config file values;
-3. CLI overrides.
-
-Intentional differences:
-
-- `render` supports `--format`; `run` does not.
-- `serve` supports `--addr`; `run`/`render` do not.
-- `serve` rejects runtime model override flags.
-- `render` does not expose `--schema-dir`, even though it can prepare `json_schema` prompts through config/default schema settings.
-- CLI runtime model override flags are narrower than HTTP model override fields; this is currently documented by omission in CLI docs.
-
-Likely cleanup areas:
-
-- Common command settings finalization can be made smaller and less repetitive.
-- Runner dependency construction should be a CLI-local helper.
-- The `runConfig` flag-set booleans work, but adding more runtime override fields will continue to require touching several fields and the override condition.
-
-## State, Manifest, Or Progress Handling Review
-
-Scriptorium has no durable state, manifests, checkpoints, progress records, cache state, or resume behavior. This matches `docs/policy/architecture.md` and `docs/operations.md`.
-
-There is no drift affecting resume, retry, force, dry-run, or audit behavior because those features do not exist. Validation repair hooks exist in the runner but are not wired by CLI/HTTP today; that boundary is documented.
-
-## Refactors To Avoid
-
-- Do not introduce a generic workflow/stage engine. Scriptorium intentionally executes one prompt request.
-- Do not build a plugin architecture for prompt/profile/schema/artifact backends before another backend exists.
-- Do not replace explicit CLI parse functions with a broad command framework.
-- Do not merge CLI and HTTP adapters. Their public interfaces and failure semantics differ.
-- Do not create a generic reflection-based mapper for domain/DTO/request structs.
-- Do not centralize prompt/profile normalization into one generic YAML repository; their validation and error semantics are different.
-- Do not add schema caching or a manifest/state store as part of deduplication.
-- Do not document or implement unsupported artifact backends as cleanup.
-
-## Recommended Implementation Sequence
-
-1. Execution-target mapping cleanup
-   - Goal: reduce drift when adding profile/runtime/provider fields.
-   - Files to update: `internal/usecase`, `internal/adapter/http`, `internal/llm`, targeted tests.
-   - Acceptance criteria: all existing field mapping tests pass; a new all-fields test fails if a mapped execution field is omitted.
-   - Suggested validation: `go test ./internal/usecase ./internal/adapter/http ./internal/llm ./internal/format`.
-   - Small enough for one implementation prompt: yes.
-
-2. Prompt/profile filesystem catalog helper
-   - Goal: centralize recursive YAML discovery, path sorting, extension filtering, and relative path formatting.
-   - Files to update: prompt/profile repositories plus new narrow helper package.
-   - Acceptance criteria: existing repository tests pass unchanged; new helper tests cover nested discovery and stable ordering.
-   - Suggested validation: `go test ./internal/promptdef ./internal/profile`.
-   - Small enough for one implementation prompt: yes.
-
-3. CLI wiring and config finalization helper
-   - Goal: centralize common command settings resolution and runner construction.
-   - Files to update: `internal/adapter/cli/run.go` and CLI tests.
-   - Acceptance criteria: no flag behavior changes; run/render/serve config precedence tests pass.
-   - Suggested validation: `go test ./internal/adapter/cli`.
-   - Small enough for one implementation prompt: yes.
-
-4. Stable use-case invalid-request reasons
-   - Goal: remove HTTP dependency on runner error message substrings.
-   - Files to update: `internal/usecase/runner.go`, `internal/adapter/http/handler.go`, tests.
-   - Acceptance criteria: HTTP error codes/messages remain unchanged; runner exposes stable `errors.Is` reason errors.
-   - Suggested validation: `go test ./internal/usecase ./internal/adapter/http`.
-   - Small enough for one implementation prompt: yes.
-
-5. Test helper cleanup
-   - Goal: reduce repeated fixture construction after behavior-preserving refactors are complete.
-   - Files to update: primarily `internal/adapter/cli/run_test.go`, optionally `internal/usecase/runner_test.go`.
-   - Acceptance criteria: test intent remains clear; no cross-package helper package unless strongly justified.
-   - Suggested validation: `go test ./internal/adapter/cli ./internal/usecase`.
-   - Small enough for one implementation prompt: yes.
-
-6. Dead-code/legacy sweep
-   - Goal: remove or clearly mark unsupported placeholders such as `ArtifactRefS3`.
-   - Files to update: domain/artifact tests/docs only if needed.
-   - Acceptance criteria: docs continue to describe only implemented behavior outside roadmap.
-   - Suggested validation: `go test ./internal/domain ./internal/artifact`.
-   - Small enough for one implementation prompt: yes.
-
-## Test Strategy
-
-Tests to add before or during cleanup:
-
-- Execution target all-fields mapping tests:
-  - runner profile/default/request merge;
-  - HTTP request DTO to domain target;
-  - HTTP domain result to metadata DTO;
-  - LLM domain target to outbound JSON payload.
-- Repository catalog helper tests:
-  - recursive scan;
-  - `.yaml` and `.yml` filtering;
-  - stable sorted paths;
-  - relative clean path formatting;
-  - context cancellation if the helper preserves current behavior.
-- HTTP error mapping tests:
-  - profile-required and API-key-env-missing should rely on sentinel errors, not message text.
-- CLI wiring regression tests:
-  - `run`, `render`, and `serve` preserve config precedence and required directory checks.
-- Schema behavior tests:
-  - `Prepare` fails before LLM generation when `json_schema` structured-output schema cannot be loaded.
-
-Validation commands for cleanup work:
-
-- `go test ./internal/usecase ./internal/adapter/http ./internal/llm ./internal/format`
-- `go test ./internal/promptdef ./internal/profile`
-- `go test ./internal/adapter/cli`
-- `go test ./...` before merging broader cleanup.
-
-No automated docs or link checker is currently present.
-
-## Appendix: Findings Not Worth Acting On
-
-- Repeated `select { case <-ctx.Done(): ... }` checks are acceptable. They are local, simple, and appear at IO/loop boundaries where behavior is easy to read.
-- CLI and HTTP request validation should remain separate. Their external contracts differ, and centralizing all request validation would blur adapter responsibilities.
-- Prompt and profile validation should not be merged. They both use YAML and IDs, but their schemas, normalization rules, and error policies differ.
-- Prepared-run text formatting is verbose but intentionally presentation-specific. Avoid abstracting it until another formatter needs the same layout policy.
-- Hashing appears in artifact loading and runner metadata, but not all hashes represent the same thing. A tiny body-hash helper may be useful later; a generic hashing subsystem is not justified.
-- HTTP DTOs duplicate domain field names by design. They should remain transport-owned so JSON compatibility can evolve deliberately.
-- Config `applyConfig` and `ApplyCLIOverrides` look similar, but they apply different source labels and validation contexts. A broad merge abstraction would likely reduce clarity.

docs/roadmap/cleanup.md

@@ -1,504 +0,0 @@
-# Cleanup Implementation Roadmap
-
-## Purpose
-
-This roadmap turns the findings in [audit.md](audit.md) into a staged, decision-complete cleanup plan for Scriptorium.
-
-Audience: LLM coding agents implementing the cleanup in order.
-
-Controlling policies:
-
-- [Documentation policy](../policy/documentation.md)
-- [Architecture policy](../policy/architecture.md)
-- [Development guide](../policy/development.md)
-
-## Global Implementation Rules
-
-- Implement stages in order.
-- Keep public CLI flags, HTTP request/response shapes, config precedence, prompt/profile ID semantics, and validation behavior stable unless a stage explicitly says otherwise.
-- Keep adapters thin and use-case policy in `internal/usecase`.
-- Prefer explicit helpers over reflection, generic workflow abstractions, or broad framework-style rewrites.
-- Update non-roadmap docs only when implemented behavior changes or when a stale implemented-behavior statement is found during a stage.
-- Do not document future behavior outside `docs/roadmap/`.
-- Run the stage-specific tests before moving to the next stage.
-- Run `go test ./...` after the final stage.
-
-## Stage 1: Execution Target Mapping Cleanup
-
-### Goal
-
-Reduce drift when adding or changing execution/runtime fields such as `service_tier`, `api_key_env`, `reasoning_effort`, or future provider request keys.
-
-### Scope
-
-Update only explicit execution-target mapping and serialization paths. Do not add new CLI flags or new provider features.
-
-### Implementation
-
-In `internal/usecase`:
-
-- Keep execution-target merge policy in `internal/usecase`.
-- Add focused helper coverage around `resolveExecutionTarget`, `mergeExecutionTarget`, and profile-to-target conversion.
-- Keep the existing semantics:
-  - built-in defaults first;
-  - profile values override defaults;
-  - request overrides override profile values;
-  - zero numeric values do not override;
-  - empty/whitespace string values do not override;
-  - non-empty `ExtraParams` replaces the previous map with a copy.
-
-In `internal/adapter/http`:
-
-- Add local helper functions:
-  - `executionTargetFromModelOverrideDTO(*modelOverrideRequestDTO) *domain.ExecutionTarget`
-  - `modelParamsDTOFromExecutionTarget(domain.ExecutionTarget) modelParamsDTO`
-- Use those helpers in `handler.go`.
-- Keep DTO types unexported and transport-owned.
-- Keep HTTP response field names and omission behavior unchanged.
-
-In `internal/llm`:
-
-- Add a local helper such as `openAIChatRequestFromGenerateRequest(req domain.GenerateRequest, defaultModel string) (openAIChatRequest, error)` or an equivalent small function.
-- Keep endpoint construction, HTTP client timeout handling, API-key environment lookup, and response parsing in `Generate`.
-- Keep outbound serialization behavior unchanged:
-  - send `model` and `messages`;
-  - send `temperature`, `max_tokens`, `top_p`, and `service_tier` only when currently sent;
-  - send `response_format` only when structured output is present;
-  - do not serialize `reasoning_effort` or `extra_params`.
-
-Do not:
-
-- use reflection to copy fields;
-- move HTTP DTOs into `internal/domain`;
-- add generic mapper packages;
-- change prepared-run JSON tags.
-
-### Tests
-
-Add or update tests so an all-fields `domain.ExecutionTarget` catches omissions.
-
-Required tests:
-
-- Runner merge/profile conversion:
-  - profile values populate all supported execution fields;
-  - runtime overrides beat profile values for all overrideable fields;
-  - empty string overrides do not erase profile values;
-  - empty `ExtraParams` does not erase profile values.
-- HTTP adapter:
-  - request `model` object maps every supported field into `RunRequest.Execution`;
-  - response `metadata.model_params` includes every supported field according to current DTO tags.
-- LLM client:
-  - outbound JSON includes every serialized execution field;
-  - outbound JSON omits `service_tier` when empty;
-  - outbound JSON still omits `reasoning_effort` and `extra_params`.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/usecase ./internal/adapter/http ./internal/llm ./internal/format
-```
-
-### Acceptance Criteria
-
-- No public behavior changes.
-- Adding a new execution target field later has obvious mapping/test locations.
-- Existing HTTP and LLM behavior remains stable.
-- Stage is small enough for one implementation prompt.
-
-## Stage 2: Prompt/Profile Filesystem Catalog Helper
-
-### Goal
-
-Centralize shared recursive YAML discovery mechanics while preserving prompt/profile-specific validation and error behavior.
-
-### Scope
-
-Create a narrow helper package for filesystem catalog mechanics only.
-
-Recommended package:
-
-- `internal/filecatalog`
-
-### Implementation
-
-Add helper functions with explicit, small responsibilities:
-
-- recursively find YAML files under a root directory;
-- honor context cancellation during walking;
-- accept `.yaml` and `.yml`;
-- return stable sorted full paths;
-- compute clean relative paths from a root;
-- return filename stems with `.yaml`/`.yml` stripped.
-
-Use the helper in:
-
-- `internal/promptdef/filesystem_repository.go`
-- `internal/profile/filesystem_repository.go`
-
-Preserve existing behavior:
-
-- prompt/profile lookup uses YAML `id`, not file path;
-- subdirectories are organizational only;
-- duplicate prompt/profile IDs are invalid;
-- malformed likely-target files still surface errors;
-- relative nested paths still appear in errors;
-- prompt `content_file` resolution remains relative to the prompt YAML file;
-- prompt/profile strict YAML and validation stay in their existing packages.
-
-Do not:
-
-- create a generic repository framework;
-- merge prompt and profile normalization;
-- move prompt/profile domain policy into the helper;
-- change error messages except for unavoidable wording caused by helper extraction.
-
-### Tests
-
-Add tests for `internal/filecatalog`:
-
-- nested YAML discovery;
-- `.yaml` and `.yml` accepted;
-- non-YAML files ignored;
-- returned paths sorted deterministically;
-- relative path formatting works for nested files;
-- filename stem stripping handles both extensions.
-
-Keep existing prompt/profile repository tests passing.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/filecatalog ./internal/promptdef ./internal/profile
-```
-
-### Acceptance Criteria
-
-- Prompt/profile repository tests pass without behavior expectation changes.
-- Shared filesystem scanning logic exists in one place.
-- Prompt/profile packages still own their own validation and normalization.
-- Stage is small enough for one implementation prompt.
-
-## Stage 3: CLI Wiring And Settings Finalization Cleanup
-
-### Goal
-
-Reduce duplicated command setup while preserving each command's public flag surface and behavior.
-
-### Scope
-
-Clean up `internal/adapter/cli` only, except for tests.
-
-### Implementation
-
-Add CLI-local helpers. Recommended helpers:
-
-- `commonCommandSettings` or similar struct containing resolved `promptDir`, `profileDir`, `schemaDir`, `serverAddr`, and `defaultRenderFormat` where applicable.
-- `resolveCommonSettings(fs *flag.FlagSet, configPath string, overrides appconfig.CLIOverrides) (commonCommandSettings, error)`.
-- `validateRequiredLibraryDirs(promptDir, profileDir string) error`.
-- `newRunner(promptDir, profileDir, schemaDir string, llmClient llm.Client) *usecase.Runner`.
-- Optionally `newOpenAIClient() (*llm.OpenAICompatibleClient, error)` if it removes exact duplication without obscuring command behavior.
-
-Preserve command differences:
-
-- `run` exposes runtime model override flags and `--schema-dir`;
-- `render` exposes runtime model override flags and `--format`, but not `--schema-dir`;
-- `serve` exposes `--addr` and `--schema-dir`, but no runtime model override flags;
-- `render` default format comes from `defaults.render_format` unless `--format` is set;
-- deprecated `--prompt-id` and `--profile-id` aliases remain accepted.
-
-Keep existing parse functions:
-
-- `parseRunArgs`
-- `parseRenderArgs`
-- `parseServeArgs`
-
-Do not:
-
-- replace the standard library `flag` package;
-- introduce a command framework;
-- make `serve` accept runtime model override flags;
-- change error prefixes such as `run parse error`, `render parse error`, or `serve parse error`;
-- change CLI output behavior.
-
-### Tests
-
-Required regression tests:
-
-- `run`, `render`, and `serve` still apply config precedence correctly.
-- Missing effective `prompt_dir` and `profile_dir` still return the same guidance.
-- `render` still uses config default render format and explicit `--format` override.
-- `serve` still rejects runtime model override flags.
-- `run` and `render` still build equivalent runtime override requests for shared flags.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/adapter/cli
-```
-
-### Acceptance Criteria
-
-- No CLI flag, output, exit-code, or precedence changes.
-- Runner dependency construction is centralized inside the CLI adapter.
-- Command-specific behavior remains easy to read.
-- Stage is small enough for one implementation prompt.
-
-## Stage 4: Stable Use-Case Error Reasons For HTTP Mapping
-
-### Goal
-
-Remove HTTP error mapping's dependency on runner error message substrings.
-
-### Scope
-
-Change error identity, not public HTTP error responses.
-
-### Implementation
-
-In `internal/usecase`:
-
-- Add stable sentinel errors for invalid-request reasons that HTTP currently distinguishes by message text.
-- Required sentinels:
-  - missing profile selection, for the case where neither request profile nor prompt `default_profile` is available;
-  - missing API-key environment value, for the case where `api_key_env` is set but the named environment variable is unset or empty.
-- Wrap these sentinels with `ErrInvalidRequest` so existing broad invalid-request checks keep working.
-- Preserve clear human-readable runner errors.
-
-In `internal/adapter/http`:
-
-- Replace `strings.Contains(err.Error(), ...)` checks for these cases with `errors.Is`.
-- Keep current HTTP status codes, error codes, and response messages:
-  - `400 profile_required`;
-  - `400 api_key_env_missing`.
-
-Do not:
-
-- expose HTTP-specific error codes from `internal/usecase`;
-- change the HTTP JSON error body shape;
-- remove broad fallback handling for `usecase.ErrInvalidRequest`.
-
-### Tests
-
-Required tests:
-
-- Runner tests assert `errors.Is(err, usecase.ErrProfileRequired)` or the chosen sentinel name for missing profile selection.
-- Runner tests assert `errors.Is(err, usecase.ErrAPIKeyEnvMissing)` or the chosen sentinel name for missing API-key environment value.
-- HTTP handler tests still assert unchanged status/code/message for both cases.
-- HTTP handler tests should not construct errors by relying on exact runner prose for these two cases.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/usecase ./internal/adapter/http
-```
-
-### Acceptance Criteria
-
-- HTTP mapping no longer depends on runner message substrings for the two distinguished invalid-request cases.
-- Public HTTP behavior is unchanged.
-- Runner errors remain clear in CLI output.
-- Stage is small enough for one implementation prompt.
-
-## Stage 5: Schema Failure Regression Coverage
-
-### Goal
-
-Protect the current structured-output invariant before future schema cleanup: `Prepare` must load a `json_schema` document before any LLM call.
-
-### Scope
-
-Add regression coverage only. Do not add schema caching or change schema loading architecture in this stage.
-
-### Implementation
-
-In `internal/usecase/runner_test.go` or an appropriate package test:
-
-- Add a test where a prompt uses `validation_mode: json_schema` with a missing or failing schema document.
-- Assert `Runner.Prepare` fails with `ErrValidation`.
-- Assert no LLM call is made for `Runner.Run` when structured-output schema loading fails.
-
-If existing tests already cover part of this behavior, consolidate assertions without making the test suite harder to read.
-
-Do not:
-
-- cache compiled schemas;
-- change `validate.StandardValidator` behavior;
-- introduce a schema service abstraction.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/usecase ./internal/validate
-```
-
-### Acceptance Criteria
-
-- Missing structured-output schema fails before generation.
-- Existing JSON Schema validation behavior is unchanged.
-- Stage is small enough for one implementation prompt.
-
-## Stage 6: Test Fixture Cleanup
-
-### Goal
-
-Reduce repeated test setup after behavior-preserving production refactors are complete.
-
-### Scope
-
-Prefer package-local test helpers. Avoid cross-package test utility packages unless a helper is needed by more than two packages and represents a stable public fixture contract.
-
-### Implementation
-
-In `internal/adapter/cli/run_test.go`:
-
-- Consolidate repeated temp prompt/profile/input setup into local helper functions.
-- Keep helper names behavior-focused, for example:
-  - `newCLITestLibrary`
-  - `writePromptFileWithDefaultProfile`
-  - `writeProfileFile`
-  - `runCLICommand`
-- Do not hide assertions inside helpers unless the assertion is truly setup validation.
-
-In `internal/usecase/runner_test.go`:
-
-- Keep existing fake interfaces package-local.
-- Remove only high-volume duplication that obscures test intent.
-
-Do not:
-
-- move package-private fake types into production code;
-- create a broad `internal/testutil` package unless a later cleanup stage proves it necessary;
-- rewrite tests into table-driven form when cases have meaningfully different setup.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/adapter/cli ./internal/usecase
-```
-
-### Acceptance Criteria
-
-- Test intent is at least as clear as before.
-- No production behavior changes.
-- Test fixture setup has less repeated boilerplate in CLI tests.
-- Stage is small enough for one implementation prompt.
-
-## Stage 7: Unsupported Placeholder Sweep
-
-### Goal
-
-Remove code that suggests unimplemented artifact behavior outside roadmap documentation.
-
-### Scope
-
-Remove unsupported placeholders only when they are not needed by current tests or public docs.
-
-### Implementation
-
-Remove `domain.ArtifactRefS3` from `internal/domain/domain.go` unless new evidence shows it is intentionally needed by implemented code.
-
-Preserve current behavior:
-
-- supported artifact reference types remain `inline` and `file`;
-- unsupported artifact reference types still return `artifact.ErrUnsupportedRefType`;
-- docs continue to describe only `inline` and `file` outside roadmap files.
-
-Update tests only if they reference the removed constant. Prefer testing unsupported artifact behavior with a literal custom type such as `domain.ArtifactRefType("s3")` or `domain.ArtifactRefType("unsupported")`.
-
-Do not:
-
-- add S3 support;
-- document S3 as implemented;
-- add future-backend placeholders elsewhere.
-
-### Validation
-
-Run:
-
-```bash
-go test ./internal/domain ./internal/artifact
-```
-
-### Acceptance Criteria
-
-- Unsupported placeholder constant is removed.
-- Unsupported artifact-type behavior remains covered.
-- No non-roadmap doc claims unimplemented artifact support.
-- Stage is small enough for one implementation prompt.
-
-## Stage 8: Final Verification And Documentation Alignment
-
-### Goal
-
-Confirm the cleanup sequence preserved behavior and documentation accuracy.
-
-### Implementation
-
-Run the full test suite:
-
-```bash
-go test ./...
-```
-
-Run the maintained render smoke command:
-
-```bash
-go run ./cmd/scriptorium render --config ./examples/config.yml --prompt generic.markdown_summary --input transcript=./examples/fixtures/transcript.md --input glossary=./examples/fixtures/glossary.yml --format json
-```
-
-Search for stale or unsupported terms:
-
-```bash
-rg -n "ArtifactRefS3|s3|strings\\.Contains\\(err\\.Error\\(\\)|TODO|future|planned" internal docs README.md examples
-```
-
-Review results manually:
-
-- `s3` should not appear as an implemented artifact type.
-- `strings.Contains(err.Error())` should not be used for stable use-case reason mapping.
-- Any `future` or `planned` wording outside `docs/roadmap/` must describe current boundaries, not aspirational behavior.
-
-Update docs only if cleanup changed implemented behavior or if the search reveals stale implemented-behavior docs.
-
-### Acceptance Criteria
-
-- Full test suite passes.
-- Maintained render smoke command succeeds.
-- No stale unsupported artifact placeholder remains.
-- Non-roadmap docs describe implemented behavior only.
-- Working tree contains only intentional cleanup changes.
-
-## Deferred Work
-
-Do not implement these during the staged cleanup unless a later audit makes them high-confidence:
-
-- schema caching or a combined raw/compiled schema service;
-- artifact hash/build helper beyond a small helper introduced opportunistically during touched code;
-- generic YAML repository framework;
-- generic CLI command framework;
-- plugin architecture for future prompt/profile/schema/artifact backends;
-- durable state, manifests, checkpoints, or resume behavior.
-
-## Completion Criteria
-
-The cleanup roadmap is complete when all stages have been implemented in order, the final verification passes, and the resulting code still satisfies:
-
-- `Runner.Run` reuses `Runner.Prepare`;
-- CLI and HTTP adapters instantiate `Runner` without a repairer;
-- unknown config/prompt/profile YAML and HTTP JSON fields are rejected;
-- raw API key values are not accepted or emitted;
-- prompt/profile subdirectories remain organizational only;
-- schema paths remain explicit and relative to `schema_dir` when not absolute;
-- public CLI and HTTP behavior remains stable.

internal/adapter/cli/run.go

@@ -81,6 +81,14 @@ type serveConfig struct {
 	schemaDir  string
 }
 
+type commonCommandSettings struct {
+	promptDir           string
+	profileDir          string
+	schemaDir           string
+	serverAddr          string
+	defaultRenderFormat renderformat.PreparedRunOutputFormat
+}
+
 type listFlag []string
 
 func (l *listFlag) String() string {
@@ -125,22 +133,13 @@ func runCommand(args []string, stdout, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	llmClient, err := llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
+	llmClient, err := newOpenAIClient()
-		Timeout: defaults.LLMRequestTimeoutDefault,
-	})
 	if err != nil {
 		fmt.Fprintf(stderr, "llm client error: %v\n", err)
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
+	runner := newRunner(cfg.promptDir, cfg.profileDir, cfg.schemaDir, llmClient)
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		llmClient,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
 
 	res, runErr := runner.Run(context.Background(), req)
 	if runErr != nil {
@@ -170,14 +169,7 @@ func renderCommand(args []string, stdout, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
+	runner := newRunner(cfg.promptDir, cfg.profileDir, cfg.schemaDir, nil)
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		nil,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
 
 	prepared, prepErr := runner.Prepare(context.Background(), req)
 	if prepErr != nil {
@@ -205,22 +197,13 @@ func serveCommand(args []string, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	llmClient, err := llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
+	llmClient, err := newOpenAIClient()
-		Timeout: defaults.LLMRequestTimeoutDefault,
-	})
 	if err != nil {
 		fmt.Fprintf(stderr, "llm client error: %v\n", err)
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
+	runner := newRunner(cfg.promptDir, cfg.profileDir, cfg.schemaDir, llmClient)
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		llmClient,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
 
 	h := httpadapter.NewHandler(runner)
 	srv := &http.Server{
@@ -307,7 +290,7 @@ func parseServeArgs(args []string) (*serveConfig, error) {
 		return nil, fmt.Errorf("unexpected positional args: %v", fs.Args())
 	}
 
-	settings, err := resolveAppSettings(fs, cfg.configPath, appconfig.CLIOverrides{
+	settings, err := resolveCommonSettings(fs, cfg.configPath, appconfig.CLIOverrides{
 		PromptDir:  cfg.promptDirIfSet(fs),
 		ProfileDir: cfg.profileDirIfSet(fs),
 		SchemaDir:  cfg.schemaDirIfSet(fs),
@@ -317,16 +300,13 @@ func parseServeArgs(args []string) (*serveConfig, error) {
 		return nil, err
 	}
 
-	cfg.promptDir = settings.PromptDir
+	cfg.promptDir = settings.promptDir
-	cfg.profileDir = settings.ProfileDir
+	cfg.profileDir = settings.profileDir
-	cfg.schemaDir = settings.SchemaDir
+	cfg.schemaDir = settings.schemaDir
-	cfg.addr = settings.ServerAddr
+	cfg.addr = settings.serverAddr
 
-	if strings.TrimSpace(cfg.promptDir) == "" {
+	if err := validateRequiredLibraryDirs(cfg.promptDir, cfg.profileDir); err != nil {
-		return nil, errors.New(errPromptDirRequired)
+		return nil, err
-	}
-	if strings.TrimSpace(cfg.profileDir) == "" {
-		return nil, errors.New(errProfileDirRequired)
 	}
 
 	cfg.promptDir = filepath.Clean(cfg.promptDir)
@@ -359,7 +339,7 @@ func finalizeExecutionRequestConfig(fs *flag.FlagSet, cfg *runConfig) error {
 		return fmt.Errorf("unexpected positional args: %v", fs.Args())
 	}
 
-	settings, err := resolveAppSettings(fs, cfg.configPath, appconfig.CLIOverrides{
+	settings, err := resolveCommonSettings(fs, cfg.configPath, appconfig.CLIOverrides{
 		PromptDir:  cfg.promptDirIfSet(fs),
 		ProfileDir: cfg.profileDirIfSet(fs),
 		SchemaDir:  cfg.schemaDirIfSet(fs),
@@ -368,16 +348,13 @@ func finalizeExecutionRequestConfig(fs *flag.FlagSet, cfg *runConfig) error {
 		return err
 	}
 
-	cfg.promptDir = settings.PromptDir
+	cfg.promptDir = settings.promptDir
-	cfg.profileDir = settings.ProfileDir
+	cfg.profileDir = settings.profileDir
-	cfg.schemaDir = settings.SchemaDir
+	cfg.schemaDir = settings.schemaDir
-	cfg.defaultRenderFormat = settings.DefaultRenderFormat
+	cfg.defaultRenderFormat = settings.defaultRenderFormat
 
-	if strings.TrimSpace(cfg.promptDir) == "" {
+	if err := validateRequiredLibraryDirs(cfg.promptDir, cfg.profileDir); err != nil {
-		return errors.New(errPromptDirRequired)
+		return err
-	}
-	if strings.TrimSpace(cfg.profileDir) == "" {
-		return errors.New(errProfileDirRequired)
 	}
 	if strings.TrimSpace(cfg.promptID) == "" {
 		return errors.New("--prompt is required")
@@ -476,6 +453,47 @@ func resolveAppSettings(fs *flag.FlagSet, configPath string, overrides appconfig
 	return merged, nil
 }
 
+func resolveCommonSettings(fs *flag.FlagSet, configPath string, overrides appconfig.CLIOverrides) (commonCommandSettings, error) {
+	settings, err := resolveAppSettings(fs, configPath, overrides)
+	if err != nil {
+		return commonCommandSettings{}, err
+	}
+	return commonCommandSettings{
+		promptDir:           settings.PromptDir,
+		profileDir:          settings.ProfileDir,
+		schemaDir:           settings.SchemaDir,
+		serverAddr:          settings.ServerAddr,
+		defaultRenderFormat: settings.DefaultRenderFormat,
+	}, nil
+}
+
+func validateRequiredLibraryDirs(promptDir, profileDir string) error {
+	if strings.TrimSpace(promptDir) == "" {
+		return errors.New(errPromptDirRequired)
+	}
+	if strings.TrimSpace(profileDir) == "" {
+		return errors.New(errProfileDirRequired)
+	}
+	return nil
+}
+
+func newRunner(promptDir, profileDir, schemaDir string, llmClient llm.Client) *usecase.Runner {
+	return usecase.NewRunner(
+		promptdef.NewFilesystemRepository(promptDir),
+		profile.NewFilesystemRepository(profileDir),
+		artifactadapter.NewCompositeReader(),
+		prompt.NewGoRenderer(),
+		llmClient,
+		validate.NewStandardValidator(schemaDir),
+	)
+}
+
+func newOpenAIClient() (*llm.OpenAICompatibleClient, error) {
+	return llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
+		Timeout: defaults.LLMRequestTimeoutDefault,
+	})
+}
+
 func buildRunRequestFromConfig(cfg *runConfig) (domain.RunRequest, error) {
 	inputMappings, err := parseMappings(cfg.inputRaw, false)
 	if err != nil {

internal/adapter/cli/run_test.go

@@ -11,6 +11,7 @@ import (
 	"net/http/httptest"
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"sync/atomic"
 	"testing"
@@ -410,6 +411,28 @@ defaults:
 	}
 }
 
+func TestParseRenderArgsExplicitFormatOverridesConfigDefaultFormat(t *testing.T) {
+	configPath := writeAppConfigFile(t, `
+prompt_dir: ./from-config/prompts
+profile_dir: ./from-config/profiles
+defaults:
+  render_format: json
+`)
+
+	cfg, err := parseRenderArgs([]string{
+		"--config", configPath,
+		"--prompt", "p",
+		"--input", "a=b",
+		"--format", "text",
+	})
+	if err != nil {
+		t.Fatalf("expected valid args, got %v", err)
+	}
+	if cfg.outputFormat != renderformat.PreparedRunFormatText {
+		t.Fatalf("expected explicit --format text to override config default, got %q", cfg.outputFormat)
+	}
+}
+
 func TestParseServeArgsWithExplicitConfigLoadsSettingsAndCLIAddrOverrides(t *testing.T) {
 	configPath := writeAppConfigFile(t, `
 prompt_dir: ./from-config/prompts
@@ -471,6 +494,59 @@ server:
 	}
 }
 
+func TestRunAndRenderBuildEquivalentRuntimeOverrideRequestsForSharedFlags(t *testing.T) {
+	runCfg, err := parseRunArgs([]string{
+		"--prompt-dir", "./prompts",
+		"--profile-dir", "./profiles",
+		"--prompt", "prompt-1",
+		"--profile", "profile-1",
+		"--input", "transcript=./transcript.md",
+		"--var", "session_date=2026-05-01",
+		"--llm-base-url", "http://localhost:8000/v1",
+		"--model", "model-x",
+		"--temperature", "0.8",
+		"--max-tokens", "123",
+		"--top-p", "0.6",
+		"--timeout", "90s",
+		"--api-key-env", "SCRIPTORIUM_API_KEY",
+	})
+	if err != nil {
+		t.Fatalf("expected valid run args, got %v", err)
+	}
+
+	renderCfg, err := parseRenderArgs([]string{
+		"--prompt-dir", "./prompts",
+		"--profile-dir", "./profiles",
+		"--prompt", "prompt-1",
+		"--profile", "profile-1",
+		"--input", "transcript=./transcript.md",
+		"--var", "session_date=2026-05-01",
+		"--llm-base-url", "http://localhost:8000/v1",
+		"--model", "model-x",
+		"--temperature", "0.8",
+		"--max-tokens", "123",
+		"--top-p", "0.6",
+		"--timeout", "90s",
+		"--api-key-env", "SCRIPTORIUM_API_KEY",
+	})
+	if err != nil {
+		t.Fatalf("expected valid render args, got %v", err)
+	}
+
+	runReq, err := buildRunRequestFromConfig(runCfg)
+	if err != nil {
+		t.Fatalf("expected run request build success, got %v", err)
+	}
+	renderReq, err := buildRunRequestFromConfig(&renderCfg.runConfig)
+	if err != nil {
+		t.Fatalf("expected render request build success, got %v", err)
+	}
+
+	if !reflect.DeepEqual(runReq, renderReq) {
+		t.Fatalf("expected run/render shared flag requests to match.\nrun=%#v\nrender=%#v", runReq, renderReq)
+	}
+}
+
 func TestParseRunArgsFailsClearlyWhenNoEffectivePromptDir(t *testing.T) {
 	configPath := writeAppConfigFile(t, `
 profile_dir: ./profiles
@@ -586,42 +662,29 @@ func TestRunCommandVarsOptional(t *testing.T) {
 }
 
 func TestRunCommandSucceedsWithPromptAndProfileDirsFromConfig(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
 	ts := newTestLLMServer("from-config-dirs", nil)
 	defer ts.Close()
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", ts.URL+"/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", ts.URL+"/v1", "profile-model")
 	configPath := writeAppConfigFile(t, fmt.Sprintf(`
 prompt_dir: %s
 profile_dir: %s
-`, promptDir, profileDir))
+`, lib.promptDir, lib.profileDir))
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, runCommand, []string{
-	var stderr bytes.Buffer
-	code := runCommand([]string{
 		"--config", configPath,
 		"--prompt", "prompt.default",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if stdout.String() != "from-config-dirs" {
+	if stdout != "from-config-dirs" {
-		t.Fatalf("unexpected stdout output: %q", stdout.String())
+		t.Fatalf("unexpected stdout output: %q", stdout)
 	}
 }
 
@@ -630,28 +693,15 @@ func TestRenderCommandDefaultFormatTextIncludesPreparedDetailsAndNoSecrets(t *te
 	const secret = "super-secret-render-key"
 	t.Setenv(envName, secret)
 
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello transcript")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello transcript"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFileWithTemplate(t, promptDir, "prompt.render", "local-default", "Date {{.session_date}} - Summarize: {{input \"transcript\"}}")
+	writePromptFileWithTemplate(t, lib.promptDir, "prompt.render", "local-default", "Date {{.session_date}} - Summarize: {{input \"transcript\"}}")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.render",
 		"--profile", "local-default",
 		"--input", "transcript=" + inputPath,
@@ -663,15 +713,15 @@ func TestRenderCommandDefaultFormatTextIncludesPreparedDetailsAndNoSecrets(t *te
 		"--top-p", "0.2",
 		"--timeout", "20s",
 		"--api-key-env", envName,
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if stderr.Len() != 0 {
+	if stderr != "" {
-		t.Fatalf("expected empty stderr on success, got %q", stderr.String())
+		t.Fatalf("expected empty stderr on success, got %q", stderr)
 	}
 
-	out := stdout.String()
+	out := stdout
 	for _, want := range []string{
 		"prompt: prompt.render",
 		"selected_profile_id: local-default",
@@ -698,111 +748,72 @@ func TestRenderCommandDefaultFormatTextIncludesPreparedDetailsAndNoSecrets(t *te
 }
 
 func TestRenderCommandSucceedsWithPromptAndProfileDirsFromConfig(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello transcript")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello transcript"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFile(t, promptDir, "prompt.render", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.render", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
 	configPath := writeAppConfigFile(t, fmt.Sprintf(`
 prompt_dir: %s
 profile_dir: %s
-`, promptDir, profileDir))
+`, lib.promptDir, lib.profileDir))
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
-	code := renderCommand([]string{
 		"--config", configPath,
 		"--prompt", "prompt.render",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if !strings.Contains(stdout.String(), "prompt: prompt.render") {
+	if !strings.Contains(stdout, "prompt: prompt.render") {
-		t.Fatalf("expected rendered output, got %q", stdout.String())
+		t.Fatalf("expected rendered output, got %q", stdout)
 	}
 }
 
 func TestRenderCommandExplicitTextFormatWorks(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello transcript")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello transcript"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFile(t, promptDir, "prompt.render", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.render", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.render",
 		"--input", "transcript=" + inputPath,
 		"--format", "text",
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if !strings.Contains(stdout.String(), "prompt: prompt.render") {
+	if !strings.Contains(stdout, "prompt: prompt.render") {
-		t.Fatalf("expected text output for explicit --format text, got %q", stdout.String())
+		t.Fatalf("expected text output for explicit --format text, got %q", stdout)
 	}
 }
 
 func TestRenderCommandExplicitJSONFormatOutputsValidJSON(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello transcript")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello transcript"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFile(t, promptDir, "prompt.render", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.render", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.render",
 		"--input", "transcript=" + inputPath,
 		"--format", "json",
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
 
 	var payload map[string]any
-	if err := json.Unmarshal(stdout.Bytes(), &payload); err != nil {
+	if err := json.Unmarshal([]byte(stdout), &payload); err != nil {
-		t.Fatalf("expected valid json output, got %v\nbody=%s", err, stdout.String())
+		t.Fatalf("expected valid json output, got %v\nbody=%s", err, stdout)
 	}
 	if payload["prompt_id"] != "prompt.render" {
 		t.Fatalf("expected prompt_id, got %#v", payload["prompt_id"])
@@ -834,38 +845,25 @@ func TestRenderCommandUnknownFormatFailsClearly(t *testing.T) {
 }
 
 func TestRenderCommandOutWritesToFile(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello transcript")
-	profileDir := filepath.Join(tmp, "profiles")
+	outPath := filepath.Join(lib.rootDir, "render.txt")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello transcript"), 0o644); err != nil {
-		t.Fatal(err)
-	}
-	outPath := filepath.Join(tmp, "render.txt")
 
-	writePromptFile(t, promptDir, "prompt.render", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.render", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.render",
 		"--input", "transcript=" + inputPath,
 		"--out", outPath,
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if stdout.Len() != 0 {
+	if stdout != "" {
-		t.Fatalf("expected empty stdout when --out is set, got %q", stdout.String())
+		t.Fatalf("expected empty stdout when --out is set, got %q", stdout)
 	}
 	out, err := os.ReadFile(outPath)
 	if err != nil {
@@ -877,36 +875,23 @@ func TestRenderCommandOutWritesToFile(t *testing.T) {
 }
 
 func TestRenderCommandPromptDefaultProfileWorksThroughCLIPath(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "default-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "default-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.default",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	out := stdout.String()
+	out := stdout
 	if !strings.Contains(out, "selected_profile_id: local-default") {
 		t.Fatalf("expected prompt default profile in output, got %q", out)
 	}
@@ -916,38 +901,25 @@ func TestRenderCommandPromptDefaultProfileWorksThroughCLIPath(t *testing.T) {
 }
 
 func TestRenderCommandExplicitProfileOverridesPromptDefault(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", "http://127.0.0.1:1/v1", "default-model")
+	writeProfileFile(t, lib.profileDir, "local-default", "http://127.0.0.1:1/v1", "default-model")
-	writeProfileFile(t, profileDir, "quality", "http://127.0.0.1:1/v1", "quality-model")
+	writeProfileFile(t, lib.profileDir, "quality", "http://127.0.0.1:1/v1", "quality-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, renderCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := renderCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.default",
 		"--profile", "quality",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	out := stdout.String()
+	out := stdout
 	if !strings.Contains(out, "selected_profile_id: quality") {
 		t.Fatalf("expected explicit profile in output, got %q", out)
 	}
@@ -957,104 +929,67 @@ func TestRenderCommandExplicitProfileOverridesPromptDefault(t *testing.T) {
 }
 
 func TestRunCommandPromptDefaultProfileWorksThroughCLIPath(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
 	ts := newTestLLMServer("default-output", nil)
 	defer ts.Close()
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", ts.URL+"/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", ts.URL+"/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, runCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := runCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.default",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if stdout.String() != "default-output" {
+	if stdout != "default-output" {
-		t.Fatalf("unexpected stdout output: %q", stdout.String())
+		t.Fatalf("unexpected stdout output: %q", stdout)
 	}
-	if !strings.Contains(stderr.String(), "selected_profile=local-default") {
+	if !strings.Contains(stderr, "selected_profile=local-default") {
-		t.Fatalf("expected selected profile in summary, got %q", stderr.String())
+		t.Fatalf("expected selected profile in summary, got %q", stderr)
 	}
 }
 
 func TestRunCommandExplicitProfileOverridesPromptDefault(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
 	defaultServer := newTestLLMServer("from-default", nil)
 	defer defaultServer.Close()
 	overrideServer := newTestLLMServer("from-override", nil)
 	defer overrideServer.Close()
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", defaultServer.URL+"/v1", "default-model")
+	writeProfileFile(t, lib.profileDir, "local-default", defaultServer.URL+"/v1", "default-model")
-	writeProfileFile(t, profileDir, "quality", overrideServer.URL+"/v1", "quality-model")
+	writeProfileFile(t, lib.profileDir, "quality", overrideServer.URL+"/v1", "quality-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, runCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := runCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.default",
 		"--profile", "quality",
 		"--input", "transcript=" + inputPath,
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
-	if stdout.String() != "from-override" {
+	if stdout != "from-override" {
-		t.Fatalf("expected explicit profile output, got %q", stdout.String())
+		t.Fatalf("expected explicit profile output, got %q", stdout)
 	}
-	if !strings.Contains(stderr.String(), "selected_profile=quality") {
+	if !strings.Contains(stderr, "selected_profile=quality") {
-		t.Fatalf("expected selected profile quality, got %q", stderr.String())
+		t.Fatalf("expected selected profile quality, got %q", stderr)
 	}
 }
 
 func TestRunCommandRuntimeFlagsOverrideSelectedProfileValues(t *testing.T) {
-	tmp := t.TempDir()
+	lib := newCLITestLibrary(t)
-	promptDir := filepath.Join(tmp, "prompts")
+	inputPath := lib.writeInputFile(t, "transcript.md", "hello")
-	profileDir := filepath.Join(tmp, "profiles")
-	if err := os.MkdirAll(promptDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.MkdirAll(profileDir, 0o755); err != nil {
-		t.Fatal(err)
-	}
-	inputPath := filepath.Join(tmp, "transcript.md")
-	if err := os.WriteFile(inputPath, []byte("hello"), 0o644); err != nil {
-		t.Fatal(err)
-	}
 
 	var baseHits int32
 	baseServer := newTestLLMServer("base", &baseHits)
@@ -1071,14 +1006,12 @@ func TestRunCommandRuntimeFlagsOverrideSelectedProfileValues(t *testing.T) {
 	}))
 	defer overrideServer.Close()
 
-	writePromptFile(t, promptDir, "prompt.default", "local-default")
+	writePromptFile(t, lib.promptDir, "prompt.default", "local-default")
-	writeProfileFile(t, profileDir, "local-default", baseServer.URL+"/v1", "profile-model")
+	writeProfileFile(t, lib.profileDir, "local-default", baseServer.URL+"/v1", "profile-model")
 
-	var stdout bytes.Buffer
+	code, stdout, stderr := runCLICommand(t, runCommand, []string{
-	var stderr bytes.Buffer
+		"--prompt-dir", lib.promptDir,
-	code := runCommand([]string{
+		"--profile-dir", lib.profileDir,
-		"--prompt-dir", promptDir,
-		"--profile-dir", profileDir,
 		"--prompt", "prompt.default",
 		"--input", "transcript=" + inputPath,
 		"--llm-base-url", overrideServer.URL + "/v1",
@@ -1087,9 +1020,9 @@ func TestRunCommandRuntimeFlagsOverrideSelectedProfileValues(t *testing.T) {
 		"--max-tokens", "55",
 		"--top-p", "0.2",
 		"--timeout", "20s",
-	}, &stdout, &stderr)
+	})
 	if code != ExitOK {
-		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr.String())
+		t.Fatalf("expected ExitOK, got %d stderr=%q", code, stderr)
 	}
 	if atomic.LoadInt32(&baseHits) != 0 {
 		t.Fatalf("expected base profile endpoint not to be hit, got %d", baseHits)
@@ -1097,8 +1030,8 @@ func TestRunCommandRuntimeFlagsOverrideSelectedProfileValues(t *testing.T) {
 	if atomic.LoadInt32(&overrideHits) != 1 {
 		t.Fatalf("expected override endpoint to be hit once, got %d", overrideHits)
 	}
-	if stdout.String() != "override" {
+	if stdout != "override" {
-		t.Fatalf("unexpected stdout output: %q", stdout.String())
+		t.Fatalf("unexpected stdout output: %q", stdout)
 	}
 	if !strings.Contains(observedBody, `"model":"override-model"`) {
 		t.Fatalf("expected override model in request body, got %s", observedBody)
@@ -1133,6 +1066,49 @@ func TestWriteOutputAndSummaryUseSeparateWriters(t *testing.T) {
 	}
 }
 
+type cliTestLibrary struct {
+	rootDir    string
+	promptDir  string
+	profileDir string
+}
+
+func newCLITestLibrary(t *testing.T) *cliTestLibrary {
+	t.Helper()
+	root := t.TempDir()
+	lib := &cliTestLibrary{
+		rootDir:    root,
+		promptDir:  filepath.Join(root, "prompts"),
+		profileDir: filepath.Join(root, "profiles"),
+	}
+	if err := os.MkdirAll(lib.promptDir, 0o755); err != nil {
+		t.Fatalf("failed to create prompt fixture directory: %v", err)
+	}
+	if err := os.MkdirAll(lib.profileDir, 0o755); err != nil {
+		t.Fatalf("failed to create profile fixture directory: %v", err)
+	}
+	return lib
+}
+
+func (l *cliTestLibrary) writeInputFile(t *testing.T, name, body string) string {
+	t.Helper()
+	path := filepath.Join(l.rootDir, name)
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatalf("failed to create input fixture directory: %v", err)
+	}
+	if err := os.WriteFile(path, []byte(body), 0o644); err != nil {
+		t.Fatalf("failed to write input fixture: %v", err)
+	}
+	return path
+}
+
+func runCLICommand(t *testing.T, command func([]string, io.Writer, io.Writer) int, args []string) (int, string, string) {
+	t.Helper()
+	var stdout bytes.Buffer
+	var stderr bytes.Buffer
+	code := command(args, &stdout, &stderr)
+	return code, stdout.String(), stderr.String()
+}
+
 func writePromptFile(t *testing.T, dir, id, defaultProfile string) {
 	t.Helper()
 	writePromptFileWithTemplate(t, dir, id, defaultProfile, "Summarize: {{input \"transcript\"}}")

internal/adapter/http/handler.go

@@ -63,18 +63,7 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 
 	var model *domain.ExecutionTarget
 	if req.Model != nil {
-		model = &domain.ExecutionTarget{
+		model = executionTargetFromModelOverrideDTO(req.Model)
-			Endpoint:        req.Model.Endpoint,
-			Model:           req.Model.Model,
-			Temperature:     req.Model.Temperature,
-			MaxTokens:       req.Model.MaxTokens,
-			TopP:            req.Model.TopP,
-			TimeoutSeconds:  req.Model.TimeoutSeconds,
-			ServiceTier:     req.Model.ServiceTier,
-			ReasoningEffort: req.Model.ReasoningEffort,
-			APIKeyEnv:       req.Model.APIKeyEnv,
-			ExtraParams:     req.Model.ExtraParams,
-		}
 	}
 
 	res, err := h.runner.Run(r.Context(), domain.RunRequest{
@@ -110,19 +99,8 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 			SelectedProfileID:  res.SelectedProfileID,
 			ModelName:          res.ModelName,
 			Endpoint:           res.Endpoint,
-			ModelParams: modelParamsDTO{
+			ModelParams:        modelParamsDTOFromExecutionTarget(res.EffectiveModelParams),
-				Endpoint:        res.EffectiveModelParams.Endpoint,
+			InputHashes:        res.InputHashes,
-				Model:           res.EffectiveModelParams.Model,
-				Temperature:     res.EffectiveModelParams.Temperature,
-				MaxTokens:       res.EffectiveModelParams.MaxTokens,
-				TopP:            res.EffectiveModelParams.TopP,
-				TimeoutSeconds:  res.EffectiveModelParams.TimeoutSeconds,
-				ServiceTier:     res.EffectiveModelParams.ServiceTier,
-				ReasoningEffort: res.EffectiveModelParams.ReasoningEffort,
-				APIKeyEnv:       res.EffectiveModelParams.APIKeyEnv,
-				ExtraParams:     res.EffectiveModelParams.ExtraParams,
-			},
-			InputHashes: res.InputHashes,
 			Usage: tokenUsageDTO{
 				PromptTokens:     res.Usage.PromptTokens,
 				CompletionTokens: res.Usage.CompletionTokens,
@@ -143,6 +121,39 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, resp)
 }
 
+func executionTargetFromModelOverrideDTO(dto *modelOverrideRequestDTO) *domain.ExecutionTarget {
+	if dto == nil {
+		return nil
+	}
+	return &domain.ExecutionTarget{
+		Endpoint:        dto.Endpoint,
+		Model:           dto.Model,
+		Temperature:     dto.Temperature,
+		MaxTokens:       dto.MaxTokens,
+		TopP:            dto.TopP,
+		TimeoutSeconds:  dto.TimeoutSeconds,
+		ServiceTier:     dto.ServiceTier,
+		ReasoningEffort: dto.ReasoningEffort,
+		APIKeyEnv:       dto.APIKeyEnv,
+		ExtraParams:     dto.ExtraParams,
+	}
+}
+
+func modelParamsDTOFromExecutionTarget(target domain.ExecutionTarget) modelParamsDTO {
+	return modelParamsDTO{
+		Endpoint:        target.Endpoint,
+		Model:           target.Model,
+		Temperature:     target.Temperature,
+		MaxTokens:       target.MaxTokens,
+		TopP:            target.TopP,
+		TimeoutSeconds:  target.TimeoutSeconds,
+		ServiceTier:     target.ServiceTier,
+		ReasoningEffort: target.ReasoningEffort,
+		APIKeyEnv:       target.APIKeyEnv,
+		ExtraParams:     target.ExtraParams,
+	}
+}
+
 func mapValidation(v domain.ValidationResult) validationDTO {
 	return validationDTO{
 		Status:         string(v.Status),
@@ -164,9 +175,9 @@ func mapRunError(err error) (int, string, string) {
 		return http.StatusBadRequest, "prompt_load_failed", "failed to load prompt definition"
 	case errors.Is(err, profile.ErrInvalidYAML), errors.Is(err, profile.ErrInvalidProfile):
 		return http.StatusBadRequest, "profile_load_failed", "failed to load execution profile"
-	case errors.Is(err, usecase.ErrInvalidRequest) && strings.Contains(err.Error(), "profile id is required either in request or prompt default_profile"):
+	case errors.Is(err, usecase.ErrProfileRequired):
 		return http.StatusBadRequest, "profile_required", "profile_id is required when prompt default_profile is not set"
-	case errors.Is(err, usecase.ErrInvalidRequest) && strings.Contains(err.Error(), "api key environment variable"):
+	case errors.Is(err, usecase.ErrAPIKeyEnvMissing):
 		return http.StatusBadRequest, "api_key_env_missing", "api_key_env is set but the environment variable is missing"
 	case errors.Is(err, usecase.ErrInvalidRequest):
 		return http.StatusBadRequest, "invalid_request", "invalid run request"

internal/adapter/http/handler_test.go

@@ -4,10 +4,10 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"net/http"
 	"net/http/httptest"
+	"reflect"
 	"strings"
 	"testing"
 	"time"
@@ -173,6 +173,136 @@ func TestHandlerPostRunsSuccessUsingPromptDefaultProfile(t *testing.T) {
 	}
 }
 
+func TestHandlerModelOverrideMapsAllSupportedExecutionFields(t *testing.T) {
+	r := &fakeRunner{result: &domain.RunResult{
+		Artifact:             domain.Artifact{Body: []byte("ok")},
+		Validation:           domain.ValidationResult{Status: domain.ValidationPassed, Mode: domain.ValidationBasic, IsValid: true},
+		EffectiveModelParams: domain.ExecutionTarget{Endpoint: "http://llm/v1", Model: "m1"},
+	}}
+	h := NewHandler(r)
+
+	reqBody := `{
+		"prompt_id": "prompt-1",
+		"inputs": {"transcript": {"type": "file", "uri": "./t.md"}},
+		"model": {
+			"endpoint": "http://override/v1",
+			"model": "override-model",
+			"temperature": 0.6,
+			"max_tokens": 250,
+			"top_p": 0.85,
+			"timeout_seconds": 33,
+			"service_tier": "flex",
+			"reasoning_effort": "medium",
+			"api_key_env": "SCRIPTORIUM_API_KEY",
+			"extra_params": {"provider_option":"on"}
+		}
+	}`
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(reqBody))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d body=%s", w.Code, w.Body.String())
+	}
+	if r.last.Execution == nil {
+		t.Fatalf("expected execution override in run request")
+	}
+	got := r.last.Execution
+	if got.Endpoint != "http://override/v1" ||
+		got.Model != "override-model" ||
+		got.Temperature != 0.6 ||
+		got.MaxTokens != 250 ||
+		got.TopP != 0.85 ||
+		got.TimeoutSeconds != 33 ||
+		got.ServiceTier != "flex" ||
+		got.ReasoningEffort != "medium" ||
+		got.APIKeyEnv != "SCRIPTORIUM_API_KEY" {
+		t.Fatalf("unexpected mapped execution target: %+v", got)
+	}
+	if !reflect.DeepEqual(got.ExtraParams, map[string]string{"provider_option": "on"}) {
+		t.Fatalf("unexpected mapped extra_params: %#v", got.ExtraParams)
+	}
+}
+
+func TestHandlerResponseMetadataModelParamsIncludesAllSupportedFields(t *testing.T) {
+	r := &fakeRunner{result: &domain.RunResult{
+		Artifact: domain.Artifact{
+			Name:        "output",
+			ContentType: "text/plain",
+			Body:        []byte("ok"),
+			Size:        2,
+			Hash:        "abc",
+		},
+		Validation: domain.ValidationResult{Status: domain.ValidationPassed, Mode: domain.ValidationBasic, IsValid: true},
+		EffectiveModelParams: domain.ExecutionTarget{
+			Endpoint:        "http://llm/v1",
+			Model:           "gpt-test",
+			Temperature:     0.4,
+			MaxTokens:       321,
+			TopP:            0.7,
+			TimeoutSeconds:  45,
+			ServiceTier:     "priority",
+			ReasoningEffort: "high",
+			APIKeyEnv:       "SCRIPTORIUM_API_KEY",
+			ExtraParams: map[string]string{
+				"provider_option": "on",
+			},
+		},
+	}}
+	h := NewHandler(r)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{"prompt_id":"p","inputs":{"x":{"type":"file","uri":"a"}}}`))
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected 200, got %d body=%s", w.Code, w.Body.String())
+	}
+
+	var resp map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
+		t.Fatalf("invalid JSON response: %v", err)
+	}
+	metadata := resp["metadata"].(map[string]any)
+	params := metadata["model_params"].(map[string]any)
+
+	if params["endpoint"] != "http://llm/v1" {
+		t.Fatalf("unexpected endpoint: %#v", params["endpoint"])
+	}
+	if params["model"] != "gpt-test" {
+		t.Fatalf("unexpected model: %#v", params["model"])
+	}
+	if params["temperature"] != 0.4 {
+		t.Fatalf("unexpected temperature: %#v", params["temperature"])
+	}
+	if params["max_tokens"] != float64(321) {
+		t.Fatalf("unexpected max_tokens: %#v", params["max_tokens"])
+	}
+	if params["top_p"] != 0.7 {
+		t.Fatalf("unexpected top_p: %#v", params["top_p"])
+	}
+	if params["timeout_seconds"] != float64(45) {
+		t.Fatalf("unexpected timeout_seconds: %#v", params["timeout_seconds"])
+	}
+	if params["service_tier"] != "priority" {
+		t.Fatalf("unexpected service_tier: %#v", params["service_tier"])
+	}
+	if params["reasoning_effort"] != "high" {
+		t.Fatalf("unexpected reasoning_effort: %#v", params["reasoning_effort"])
+	}
+	if params["api_key_env"] != "SCRIPTORIUM_API_KEY" {
+		t.Fatalf("unexpected api_key_env: %#v", params["api_key_env"])
+	}
+	extraParams, ok := params["extra_params"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected extra_params object, got %#v", params["extra_params"])
+	}
+	if extraParams["provider_option"] != "on" {
+		t.Fatalf("unexpected extra_params.provider_option: %#v", extraParams["provider_option"])
+	}
+}
+
 func TestHandlerInvalidJSON(t *testing.T) {
 	h := NewHandler(&fakeRunner{})
 	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString("{"))
@@ -216,10 +346,10 @@ func TestHandlerUsecaseErrorMapping(t *testing.T) {
 	}{
 		{name: "prompt not found", err: wrap(usecase.ErrProfileLoad, promptdef.ErrPromptDefinitionNotFound), status: http.StatusNotFound, code: "prompt_not_found", message: "prompt definition not found"},
 		{name: "prompt load invalid", err: wrap(usecase.ErrProfileLoad, promptdef.ErrInvalidPromptDefinition), status: http.StatusBadRequest, code: "prompt_load_failed", message: "failed to load prompt definition"},
-		{name: "missing profile/default", err: wrap(usecase.ErrInvalidRequest, errors.New("profile id is required either in request or prompt default_profile")), status: http.StatusBadRequest, code: "profile_required", message: "profile_id is required when prompt default_profile is not set"},
+		{name: "missing profile/default", err: wrap(usecase.ErrInvalidRequest, usecase.ErrProfileRequired), status: http.StatusBadRequest, code: "profile_required", message: "profile_id is required when prompt default_profile is not set"},
 		{name: "profile not found", err: wrap(usecase.ErrProfileLoad, profile.ErrProfileNotFound), status: http.StatusNotFound, code: "profile_not_found", message: "execution profile not found"},
 		{name: "profile invalid", err: wrap(usecase.ErrProfileLoad, profile.ErrInvalidProfile), status: http.StatusBadRequest, code: "profile_load_failed", message: "failed to load execution profile"},
-		{name: "api key env missing", err: wrap(usecase.ErrInvalidRequest, errors.New(`api key environment variable "SCRIPTORIUM_API_KEY" is not set`)), status: http.StatusBadRequest, code: "api_key_env_missing", message: "api_key_env is set but the environment variable is missing"},
+		{name: "api key env missing", err: wrap(usecase.ErrInvalidRequest, usecase.ErrAPIKeyEnvMissing), status: http.StatusBadRequest, code: "api_key_env_missing", message: "api_key_env is set but the environment variable is missing"},
 		{name: "artifact", err: wrap(usecase.ErrArtifactLoad, fmt.Errorf("read failed")), status: http.StatusBadRequest, code: "artifact_read_failed", message: "failed to read input artifact", avoidCause: "read failed"},
 		{name: "prompt render", err: wrap(usecase.ErrPromptRender, fmt.Errorf("render failed")), status: http.StatusBadRequest, code: "prompt_render_failed", message: "failed to render prompt", avoidCause: "render failed"},
 		{name: "llm", err: wrap(usecase.ErrLLMGenerate, fmt.Errorf("llm failed")), status: http.StatusBadGateway, code: "llm_failed", message: "model generation request failed", avoidCause: "llm failed"},

internal/artifact/reader_test.go

@@ -46,8 +46,8 @@ func TestCompositeReader_Read(t *testing.T) {
 
 	t.Run("unsupported ref type", func(t *testing.T) {
 		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefS3,
+			Type: domain.ArtifactRefType("unsupported"),
-			URI:  "s3://bucket/key",
+			URI:  "unsupported://bucket/key",
 		}
 		_, err := reader.Read(ctx, ref)
 		if !errors.Is(err, ErrUnsupportedRefType) {

internal/domain/domain.go

@@ -10,7 +10,6 @@ type ArtifactRefType string
 const (
 	ArtifactRefInline ArtifactRefType = "inline"
 	ArtifactRefFile   ArtifactRefType = "file"
-	ArtifactRefS3     ArtifactRefType = "s3"
 )
 
 // OutputFormat defines the desired format of the generated artifact.

internal/filecatalog/catalog.go

@@ -0,0 +1,54 @@
+package filecatalog
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+// FindYAMLFiles returns sorted full paths for .yaml and .yml files under root.
+func FindYAMLFiles(ctx context.Context, root string) ([]string, error) {
+	var files []string
+	err := filepath.WalkDir(root, func(path string, d os.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+		if d.IsDir() {
+			return nil
+		}
+		if !isYAMLFile(d.Name()) {
+			return nil
+		}
+		files = append(files, path)
+		return nil
+	})
+	sort.Strings(files)
+	return files, err
+}
+
+// RelativePath computes a clean relative path from root to path.
+func RelativePath(root string, path string) string {
+	rel, err := filepath.Rel(root, path)
+	if err != nil {
+		return filepath.Clean(path)
+	}
+	return filepath.Clean(rel)
+}
+
+// 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

@@ -0,0 +1,84 @@
+package filecatalog
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"testing"
+)
+
+func TestFindYAMLFilesNestedSortedAndFiltered(t *testing.T) {
+	root := t.TempDir()
+	mustWriteFile(t, filepath.Join(root, "z", "prompt.yml"), "id: z")
+	mustWriteFile(t, filepath.Join(root, "a", "profile.yaml"), "id: a")
+	mustWriteFile(t, filepath.Join(root, "a", "ignore.txt"), "not yaml")
+	mustWriteFile(t, filepath.Join(root, "b", "ignore.yaml.bak"), "not yaml")
+
+	got, err := FindYAMLFiles(context.Background(), root)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+
+	want := []string{
+		filepath.Join(root, "a", "profile.yaml"),
+		filepath.Join(root, "z", "prompt.yml"),
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("expected sorted YAML files %v, got %v", want, got)
+	}
+}
+
+func TestFindYAMLFilesHonorsContextCancellation(t *testing.T) {
+	root := t.TempDir()
+	mustWriteFile(t, filepath.Join(root, "one.yaml"), "id: one")
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	_, err := FindYAMLFiles(ctx, root)
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got %v", err)
+	}
+}
+
+func TestRelativePathNested(t *testing.T) {
+	root := t.TempDir()
+	path := filepath.Join(root, "nested", "profiles", "local.yaml")
+	got := RelativePath(root, path)
+	want := filepath.Join("nested", "profiles", "local.yaml")
+	if got != want {
+		t.Fatalf("expected relative path %q, got %q", want, got)
+	}
+}
+
+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 mustWriteFile(t *testing.T, path string, content string) {
+	t.Helper()
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		t.Fatalf("failed to create directory: %v", err)
+	}
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatalf("failed to write file %q: %v", path, err)
+	}
+}

internal/llm/openai_compatible_client.go

@@ -75,14 +75,6 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 		return nil, fmt.Errorf("%w: timeout_seconds must be greater than or equal to 0", ErrInvalidRequest)
 	}
 
-	model := strings.TrimSpace(req.Target.Model)
-	if model == "" {
-		model = strings.TrimSpace(c.defaultModel)
-	}
-	if model == "" {
-		return nil, fmt.Errorf("%w: model is required", ErrInvalidRequest)
-	}
-
 	endpoint := strings.TrimSpace(req.Target.Endpoint)
 	if endpoint == "" {
 		endpoint = c.baseURL
@@ -92,36 +84,9 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 	}
 	endpoint = strings.TrimRight(endpoint, "/") + defaults.OpenAIChatCompletionsPath
 
-	wireReq := openAIChatRequest{
+	wireReq, err := openAIChatRequestFromGenerateRequest(req, c.defaultModel)
-		Model: model,
+	if err != nil {
-	}
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
-
-	wireReq.Messages = make([]openAIChatMessage, 0, len(req.Prompt.Messages))
-	for _, msg := range req.Prompt.Messages {
-		wireReq.Messages = append(wireReq.Messages, openAIChatMessage{
-			Role:    msg.Role,
-			Content: msg.Content,
-		})
-	}
-
-	if req.Target.Temperature != 0 {
-		wireReq.Temperature = &req.Target.Temperature
-	}
-	if req.Target.MaxTokens != 0 {
-		wireReq.MaxTokens = &req.Target.MaxTokens
-	}
-	if req.Target.TopP != 0 {
-		wireReq.TopP = &req.Target.TopP
-	}
-	if strings.TrimSpace(req.Target.ServiceTier) != "" {
-		wireReq.ServiceTier = req.Target.ServiceTier
-	}
-	if req.StructuredOutput != nil {
-		responseFormat, err := toOpenAIResponseFormat(req.StructuredOutput)
-		if err != nil {
-			return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
-		}
-		wireReq.ResponseFormat = responseFormat
 	}
 
 	payload, err := json.Marshal(wireReq)
@@ -190,6 +155,50 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 	}, nil
 }
 
+func openAIChatRequestFromGenerateRequest(req domain.GenerateRequest, defaultModel string) (openAIChatRequest, error) {
+	model := strings.TrimSpace(req.Target.Model)
+	if model == "" {
+		model = strings.TrimSpace(defaultModel)
+	}
+	if model == "" {
+		return openAIChatRequest{}, errors.New("model is required")
+	}
+
+	wireReq := openAIChatRequest{
+		Model: model,
+	}
+
+	wireReq.Messages = make([]openAIChatMessage, 0, len(req.Prompt.Messages))
+	for _, msg := range req.Prompt.Messages {
+		wireReq.Messages = append(wireReq.Messages, openAIChatMessage{
+			Role:    msg.Role,
+			Content: msg.Content,
+		})
+	}
+
+	if req.Target.Temperature != 0 {
+		wireReq.Temperature = &req.Target.Temperature
+	}
+	if req.Target.MaxTokens != 0 {
+		wireReq.MaxTokens = &req.Target.MaxTokens
+	}
+	if req.Target.TopP != 0 {
+		wireReq.TopP = &req.Target.TopP
+	}
+	if strings.TrimSpace(req.Target.ServiceTier) != "" {
+		wireReq.ServiceTier = req.Target.ServiceTier
+	}
+	if req.StructuredOutput != nil {
+		responseFormat, err := toOpenAIResponseFormat(req.StructuredOutput)
+		if err != nil {
+			return openAIChatRequest{}, err
+		}
+		wireReq.ResponseFormat = responseFormat
+	}
+
+	return wireReq, nil
+}
+
 type openAIChatRequest struct {
 	Model          string                `json:"model"`
 	Messages       []openAIChatMessage   `json:"messages"`

internal/llm/openai_compatible_client_test.go

@@ -96,6 +96,15 @@ func TestOpenAICompatibleClientGenerateSuccess(t *testing.T) {
 	if got, ok := obs.Body["model"].(string); !ok || got != "gpt-test" {
 		t.Fatalf("unexpected model payload: %#v", obs.Body["model"])
 	}
+	if got, ok := obs.Body["temperature"].(float64); !ok || got != 0.4 {
+		t.Fatalf("unexpected temperature payload: %#v", obs.Body["temperature"])
+	}
+	if got, ok := obs.Body["max_tokens"].(float64); !ok || got != 123 {
+		t.Fatalf("unexpected max_tokens payload: %#v", obs.Body["max_tokens"])
+	}
+	if got, ok := obs.Body["top_p"].(float64); !ok || got != 0.7 {
+		t.Fatalf("unexpected top_p payload: %#v", obs.Body["top_p"])
+	}
 	if got, ok := obs.Body["service_tier"].(string); !ok || got != "priority" {
 		t.Fatalf("unexpected service_tier payload: %#v", obs.Body["service_tier"])
 	}
@@ -166,6 +175,43 @@ func TestOpenAICompatibleClientOmitsResponseFormatWhenNoStructuredOutput(t *test
 	}
 }
 
+func TestOpenAICompatibleClientOmitsReasoningEffortAndExtraParams(t *testing.T) {
+	var observedBody map[string]any
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		defer r.Body.Close()
+		if err := json.NewDecoder(r.Body).Decode(&observedBody); err != nil {
+			t.Fatalf("failed to decode request body: %v", err)
+		}
+		_, _ = w.Write([]byte(`{"choices":[{"message":{"content":"ok"}}]}`))
+	}))
+	defer ts.Close()
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{BaseURL: ts.URL + "/v1"})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	_, err = client.Generate(context.Background(), domain.GenerateRequest{
+		Prompt: domain.RenderedPrompt{Messages: []domain.RenderedMessage{{Role: "user", Content: "hi"}}},
+		Target: domain.ExecutionTarget{
+			Model:           "model",
+			ReasoningEffort: "high",
+			ExtraParams: map[string]string{
+				"provider_option": "on",
+			},
+		},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if _, exists := observedBody["reasoning_effort"]; exists {
+		t.Fatalf("expected reasoning_effort omitted, got %#v", observedBody["reasoning_effort"])
+	}
+	if _, exists := observedBody["extra_params"]; exists {
+		t.Fatalf("expected extra_params omitted, got %#v", observedBody["extra_params"])
+	}
+}
+
 func TestOpenAICompatibleClientNoAuthorizationHeaderWhenNoAPIKey(t *testing.T) {
 	hadAuth := false
 	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {

internal/profile/filesystem_repository.go

@@ -7,10 +7,10 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"sort"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/filecatalog"
 	"gopkg.in/yaml.v3"
 )
 
@@ -34,7 +34,7 @@ func (r *filesystemRepository) GetProfile(ctx context.Context, id string) (*doma
 		return nil, fmt.Errorf("%w: profile id is required", ErrInvalidProfile)
 	}
 
-	files, err := r.yamlFiles(ctx)
+	files, err := filecatalog.FindYAMLFiles(ctx, r.dir)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read profile directory: %w", err)
 	}
@@ -47,24 +47,25 @@ func (r *filesystemRepository) GetProfile(ctx context.Context, id string) (*doma
 		default:
 		}
 
-		relPath := r.relativePath(fullPath)
+		relPath := filecatalog.RelativePath(r.dir, fullPath)
-		fileMatch := profileIDFromFileName(filepath.Base(fullPath)) == id
+		fileMatch := filecatalog.Stem(filepath.Base(fullPath)) == id
 		data, err := os.ReadFile(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
+		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 {
-			idMatch := fileMatch || profileFileHasID(data, id)
-			if strings.Contains(err.Error(), "field api_key not found") {
-				if idMatch {
-					return nil, fmt.Errorf("%w: %s", ErrRawAPIKeyNotAllowed, relPath)
-				}
-				continue
-			}
 			if idMatch {
 				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
 			}
@@ -106,56 +107,36 @@ type profileMatch struct {
 	path    string
 }
 
-func (r *filesystemRepository) yamlFiles(ctx context.Context) ([]string, error) {
+type profileFileMetadata struct {
-	var files []string
+	id           string
-	err := filepath.WalkDir(r.dir, func(path string, d os.DirEntry, err error) error {
+	hasRawAPIKey bool
-		if err != nil {
-			return err
-		}
-		select {
-		case <-ctx.Done():
-			return ctx.Err()
-		default:
-		}
-		if d.IsDir() {
-			return nil
-		}
-		if !isYAMLFile(d.Name()) {
-			return nil
-		}
-		files = append(files, path)
-		return nil
-	})
-	sort.Strings(files)
-	return files, err
 }
 
-func (r *filesystemRepository) relativePath(path string) string {
+func readProfileFileMetadata(data []byte) profileFileMetadata {
-	rel, err := filepath.Rel(r.dir, path)
+	var node yaml.Node
-	if err != nil {
+	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&node); err != nil {
-		return filepath.Clean(path)
+		return profileFileMetadata{}
 	}
-	return filepath.Clean(rel)
+	if node.Kind != yaml.DocumentNode || len(node.Content) == 0 {
-}
+		return profileFileMetadata{}
-
-func isYAMLFile(name string) bool {
-	return strings.HasSuffix(name, ".yaml") || strings.HasSuffix(name, ".yml")
-}
-
-func profileIDFromFileName(name string) string {
-	name = strings.TrimSuffix(name, ".yaml")
-	name = strings.TrimSuffix(name, ".yml")
-	return name
-}
-
-func profileFileHasID(data []byte, id string) bool {
-	var raw struct {
-		ID string `yaml:"id"`
 	}
-	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&raw); err != nil {
+	mapping := node.Content[0]
-		return false
+	if mapping.Kind != yaml.MappingNode {
+		return profileFileMetadata{}
 	}
-	return strings.TrimSpace(raw.ID) == id
+
+	var metadata profileFileMetadata
+	for i := 0; i+1 < len(mapping.Content); i += 2 {
+		key := mapping.Content[i]
+		value := mapping.Content[i+1]
+		switch key.Value {
+		case "id":
+			metadata.id = strings.TrimSpace(value.Value)
+		case "api_key":
+			metadata.hasRawAPIKey = true
+		}
+	}
+	return metadata
 }
 
 func validateProfile(p *domain.ExecutionProfile) error {

internal/profile/repository_test.go

@@ -133,6 +133,20 @@ api_key: secret
 		}
 	})
 
+	t.Run("raw api_key in non-target profile is ignored", func(t *testing.T) {
+		writeProfileTestFile(t, filepath.Join(tmpDir, "raw-api-key-non-target.yaml"), `
+id: raw-api-key-non-target
+endpoint: http://localhost:8000/v1
+model: m
+api_key: secret
+`)
+
+		_, err := repo.GetProfile(ctx, "does-not-exist-with-raw-key-nearby")
+		if !errors.Is(err, ErrProfileNotFound) {
+			t.Fatalf("expected ErrProfileNotFound for non-target raw api_key file, got %v", err)
+		}
+	})
+
 	t.Run("invalid yaml", func(t *testing.T) {
 		_, err := repo.GetProfile(ctx, "invalid_yaml")
 		if !errors.Is(err, ErrInvalidYAML) {

internal/promptdef/filesystem_repository.go

@@ -7,10 +7,10 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
-	"sort"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/filecatalog"
 	"gopkg.in/yaml.v3"
 )
 
@@ -63,7 +63,7 @@ func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id strin
 		return nil, fmt.Errorf("%w: prompt id is required", ErrInvalidPromptDefinition)
 	}
 
-	files, err := r.yamlFiles(ctx)
+	files, err := filecatalog.FindYAMLFiles(ctx, r.dir)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read prompt definition directory: %w", err)
 	}
@@ -76,8 +76,8 @@ func (r *filesystemRepository) GetPromptDefinition(ctx context.Context, id strin
 		default:
 		}
 
-		relPath := r.relativePath(fullPath)
+		relPath := filecatalog.RelativePath(r.dir, fullPath)
-		fileMatch := promptIDFromFileName(filepath.Base(fullPath)) == id
+		fileMatch := filecatalog.Stem(filepath.Base(fullPath)) == id
 
 		raw, err := loadPromptDefinitionFile(fullPath)
 		if err != nil {
@@ -130,38 +130,6 @@ type promptDefinitionMatch struct {
 	path string
 }
 
-func (r *filesystemRepository) yamlFiles(ctx context.Context) ([]string, error) {
-	var files []string
-	err := filepath.WalkDir(r.dir, func(path string, d os.DirEntry, err error) error {
-		if err != nil {
-			return err
-		}
-		select {
-		case <-ctx.Done():
-			return ctx.Err()
-		default:
-		}
-		if d.IsDir() {
-			return nil
-		}
-		if !isYAMLFile(d.Name()) {
-			return nil
-		}
-		files = append(files, path)
-		return nil
-	})
-	sort.Strings(files)
-	return files, err
-}
-
-func (r *filesystemRepository) relativePath(path string) string {
-	rel, err := filepath.Rel(r.dir, path)
-	if err != nil {
-		return filepath.Clean(path)
-	}
-	return filepath.Clean(rel)
-}
-
 func loadPromptDefinitionFile(path string) (*promptDefinitionFile, error) {
 	data, err := os.ReadFile(path)
 	if err != nil {
@@ -306,16 +274,6 @@ func normalizePromptDefinition(raw *promptDefinitionFile, sourcePath string) (*d
 	}, nil
 }
 
-func isYAMLFile(name string) bool {
-	return strings.HasSuffix(name, ".yaml") || strings.HasSuffix(name, ".yml")
-}
-
-func promptIDFromFileName(name string) string {
-	name = strings.TrimSuffix(name, ".yaml")
-	name = strings.TrimSuffix(name, ".yml")
-	return name
-}
-
 func isValidOutputFormat(f domain.OutputFormat) bool {
 	switch f {
 	case domain.FormatText, domain.FormatMarkdown, domain.FormatJSON:

internal/usecase/runner.go

@@ -24,12 +24,14 @@ import (
 )
 
 var (
-	ErrInvalidRequest = errors.New("invalid run request")
+	ErrInvalidRequest   = errors.New("invalid run request")
-	ErrProfileLoad    = errors.New("failed to load prompt definition")
+	ErrProfileRequired  = errors.New("profile selection is required")
-	ErrArtifactLoad   = errors.New("failed to load artifact")
+	ErrAPIKeyEnvMissing = errors.New("api_key_env points to an unset environment variable")
-	ErrPromptRender   = errors.New("failed to render prompt")
+	ErrProfileLoad      = errors.New("failed to load prompt definition")
-	ErrLLMGenerate    = errors.New("failed to generate output")
+	ErrArtifactLoad     = errors.New("failed to load artifact")
-	ErrValidation     = errors.New("failed to validate output")
+	ErrPromptRender     = errors.New("failed to render prompt")
+	ErrLLMGenerate      = errors.New("failed to generate output")
+	ErrValidation       = errors.New("failed to validate output")
 )
 
 // Runner executes the Scriptorium core use case.
@@ -177,7 +179,7 @@ func (r *Runner) Prepare(ctx context.Context, req domain.RunRequest) (*domain.Pr
 		selectedProfileID = strings.TrimSpace(def.DefaultProfile)
 	}
 	if selectedProfileID == "" {
-		return nil, fmt.Errorf("%w: profile id is required either in request or prompt default_profile", ErrInvalidRequest)
+		return nil, fmt.Errorf("%w: %w: profile id is required either in request or prompt default_profile", ErrInvalidRequest, ErrProfileRequired)
 	}
 
 	execProfile, err := r.profiles.GetProfile(ctx, selectedProfileID)
@@ -376,7 +378,7 @@ func validateAPIKeyEnv(apiKeyEnv string) error {
 		return nil
 	}
 	if strings.TrimSpace(os.Getenv(envName)) == "" {
-		return fmt.Errorf("api key environment variable %q is not set", envName)
+		return fmt.Errorf("%w: api key environment variable %q is not set", ErrAPIKeyEnvMissing, envName)
 	}
 	return nil
 }

internal/usecase/runner_test.go

@@ -234,6 +234,9 @@ func TestRunnerPrepareMissingExplicitProfileAndMissingDefaultProfileFails(t *tes
 	if !errors.Is(err, ErrInvalidRequest) {
 		t.Fatalf("expected ErrInvalidRequest, got %v", err)
 	}
+	if !errors.Is(err, ErrProfileRequired) {
+		t.Fatalf("expected ErrProfileRequired, got %v", err)
+	}
 }
 
 func TestRunnerPrepareSelectedProfileDoesNotExistFails(t *testing.T) {
@@ -487,6 +490,35 @@ func TestRunnerPrepareJSONSchemaBuildsStructuredOutputSpec(t *testing.T) {
 	}
 }
 
+func TestRunnerPrepareJSONSchemaSchemaLoadFailureReturnsValidationError(t *testing.T) {
+	def := promptDef(domain.FormatJSON, domain.ValidationJSONSchema, 0)
+	def.Validation.SchemaPath = "missing.schema.json"
+	validator := &fakeValidator{schemaErr: errors.New("schema unavailable")}
+	runner := NewRunner(
+		&fakePromptRepo{def: def},
+		&fakeExecutionProfileRepo{profiles: map[string]*domain.ExecutionProfile{"exec": defaultExecutionProfile()}},
+		defaultArtifactReader(),
+		defaultRenderer(),
+		&fakeLLM{forbid: true},
+		validator,
+	)
+
+	_, err := runner.Prepare(context.Background(), domain.RunRequest{
+		PromptID:  "p",
+		ProfileID: "exec",
+		Inputs:    singleInputRef(),
+	})
+	if !errors.Is(err, ErrValidation) {
+		t.Fatalf("expected ErrValidation, got %v", err)
+	}
+	if validator.schemaLoads != 1 {
+		t.Fatalf("expected one schema load attempt, got %d", validator.schemaLoads)
+	}
+	if validator.schemaLoadPath != "missing.schema.json" {
+		t.Fatalf("expected schema path missing.schema.json, got %q", validator.schemaLoadPath)
+	}
+}
+
 func TestRunnerRunJSONSchemaSchemaLoadFailureFailsBeforeLLM(t *testing.T) {
 	def := promptDef(domain.FormatJSON, domain.ValidationJSONSchema, 0)
 	def.Validation.SchemaPath = "missing.schema.json"
@@ -862,6 +894,9 @@ func TestRunnerRunAPIKeyEnvMissingEnvironmentValueFailsClearly(t *testing.T) {
 	if !errors.Is(err, ErrInvalidRequest) {
 		t.Fatalf("expected ErrInvalidRequest, got %v", err)
 	}
+	if !errors.Is(err, ErrAPIKeyEnvMissing) {
+		t.Fatalf("expected ErrAPIKeyEnvMissing, got %v", err)
+	}
 	if !strings.Contains(err.Error(), "SCRIPTORIUM_MISSING_KEY") {
 		t.Fatalf("expected missing env name in error, got %v", err)
 	}
@@ -1122,6 +1157,174 @@ func TestRunnerRunJSONSchemaRepairCarriesStructuredOutputSpec(t *testing.T) {
 	}
 }
 
+func TestExecutionProfileToTargetPopulatesAllFieldsAndCopiesExtraParams(t *testing.T) {
+	src := &domain.ExecutionProfile{
+		ID:              "exec",
+		Endpoint:        "http://profile/v1",
+		Model:           "profile-model",
+		Temperature:     0.2,
+		MaxTokens:       123,
+		TopP:            0.75,
+		TimeoutSeconds:  90,
+		ServiceTier:     "priority",
+		ReasoningEffort: "medium",
+		APIKeyEnv:       "SCRIPTORIUM_API_KEY",
+		ExtraParams: map[string]string{
+			"provider_option": "on",
+		},
+	}
+
+	target := executionProfileToTarget(src)
+	if target.Endpoint != src.Endpoint ||
+		target.Model != src.Model ||
+		target.Temperature != src.Temperature ||
+		target.MaxTokens != src.MaxTokens ||
+		target.TopP != src.TopP ||
+		target.TimeoutSeconds != src.TimeoutSeconds ||
+		target.ServiceTier != src.ServiceTier ||
+		target.ReasoningEffort != src.ReasoningEffort ||
+		target.APIKeyEnv != src.APIKeyEnv {
+		t.Fatalf("expected all profile fields to populate target, got %+v", target)
+	}
+	if !reflect.DeepEqual(target.ExtraParams, src.ExtraParams) {
+		t.Fatalf("expected extra_params to match, got %#v", target.ExtraParams)
+	}
+
+	src.ExtraParams["provider_option"] = "changed"
+	if target.ExtraParams["provider_option"] != "on" {
+		t.Fatalf("expected extra_params copy to be independent, got %#v", target.ExtraParams)
+	}
+}
+
+func TestResolveExecutionTargetProfileValuesPopulateAllSupportedFields(t *testing.T) {
+	profileValue := &domain.ExecutionProfile{
+		ID:              "exec",
+		Endpoint:        "http://profile/v1",
+		Model:           "profile-model",
+		Temperature:     0.3,
+		MaxTokens:       222,
+		TopP:            0.6,
+		TimeoutSeconds:  77,
+		ServiceTier:     "priority",
+		ReasoningEffort: "low",
+		APIKeyEnv:       "PROFILE_KEY",
+		ExtraParams: map[string]string{
+			"profile_option": "enabled",
+		},
+	}
+
+	target := resolveExecutionTarget(profileValue, nil)
+	if target.Endpoint != profileValue.Endpoint ||
+		target.Model != profileValue.Model ||
+		target.Temperature != profileValue.Temperature ||
+		target.MaxTokens != profileValue.MaxTokens ||
+		target.TopP != profileValue.TopP ||
+		target.TimeoutSeconds != profileValue.TimeoutSeconds ||
+		target.ServiceTier != profileValue.ServiceTier ||
+		target.ReasoningEffort != profileValue.ReasoningEffort ||
+		target.APIKeyEnv != profileValue.APIKeyEnv {
+		t.Fatalf("expected profile values to populate target, got %+v", target)
+	}
+	if !reflect.DeepEqual(target.ExtraParams, profileValue.ExtraParams) {
+		t.Fatalf("expected profile extra_params in target, got %#v", target.ExtraParams)
+	}
+}
+
+func TestResolveExecutionTargetRuntimeOverridesBeatProfileForAllOverrideableFields(t *testing.T) {
+	profileValue := &domain.ExecutionProfile{
+		ID:              "exec",
+		Endpoint:        "http://profile/v1",
+		Model:           "profile-model",
+		Temperature:     0.2,
+		MaxTokens:       200,
+		TopP:            0.8,
+		TimeoutSeconds:  90,
+		ServiceTier:     "priority",
+		ReasoningEffort: "medium",
+		APIKeyEnv:       "PROFILE_KEY",
+		ExtraParams: map[string]string{
+			"profile_only": "yes",
+		},
+	}
+	override := &domain.ExecutionTarget{
+		Endpoint:        "http://override/v1",
+		Model:           "override-model",
+		Temperature:     0.9,
+		MaxTokens:       111,
+		TopP:            0.5,
+		TimeoutSeconds:  30,
+		ServiceTier:     "flex",
+		ReasoningEffort: "high",
+		APIKeyEnv:       "RUNTIME_KEY",
+		ExtraParams: map[string]string{
+			"runtime_only": "yes",
+		},
+	}
+
+	target := resolveExecutionTarget(profileValue, override)
+	if target.Endpoint != override.Endpoint ||
+		target.Model != override.Model ||
+		target.Temperature != override.Temperature ||
+		target.MaxTokens != override.MaxTokens ||
+		target.TopP != override.TopP ||
+		target.TimeoutSeconds != override.TimeoutSeconds ||
+		target.ServiceTier != override.ServiceTier ||
+		target.ReasoningEffort != override.ReasoningEffort ||
+		target.APIKeyEnv != override.APIKeyEnv {
+		t.Fatalf("expected runtime overrides to win for all fields, got %+v", target)
+	}
+	if !reflect.DeepEqual(target.ExtraParams, override.ExtraParams) {
+		t.Fatalf("expected runtime extra_params to replace profile extra_params, got %#v", target.ExtraParams)
+	}
+}
+
+func TestMergeExecutionTargetEmptyStringOverridesDoNotErase(t *testing.T) {
+	base := domain.ExecutionTarget{
+		Endpoint:        "http://base/v1",
+		Model:           "base-model",
+		ServiceTier:     "priority",
+		ReasoningEffort: "medium",
+		APIKeyEnv:       "BASE_KEY",
+	}
+	override := domain.ExecutionTarget{
+		Endpoint:        "http://override/v1",
+		Model:           "override-model",
+		ServiceTier:     "   ",
+		ReasoningEffort: "  ",
+		APIKeyEnv:       "",
+	}
+
+	merged := mergeExecutionTarget(base, override)
+	if merged.Endpoint != "http://override/v1" || merged.Model != "override-model" {
+		t.Fatalf("expected endpoint/model to override, got %+v", merged)
+	}
+	if merged.ServiceTier != "priority" {
+		t.Fatalf("expected empty service_tier override to be ignored, got %q", merged.ServiceTier)
+	}
+	if merged.ReasoningEffort != "medium" {
+		t.Fatalf("expected empty reasoning_effort override to be ignored, got %q", merged.ReasoningEffort)
+	}
+	if merged.APIKeyEnv != "BASE_KEY" {
+		t.Fatalf("expected empty api_key_env override to be ignored, got %q", merged.APIKeyEnv)
+	}
+}
+
+func TestMergeExecutionTargetEmptyExtraParamsDoesNotErase(t *testing.T) {
+	base := domain.ExecutionTarget{
+		ExtraParams: map[string]string{
+			"keep": "value",
+		},
+	}
+	override := domain.ExecutionTarget{
+		ExtraParams: map[string]string{},
+	}
+
+	merged := mergeExecutionTarget(base, override)
+	if !reflect.DeepEqual(merged.ExtraParams, base.ExtraParams) {
+		t.Fatalf("expected empty extra_params override not to erase base values, got %#v", merged.ExtraParams)
+	}
+}
+
 func TestBuildOutputArtifactDefaults(t *testing.T) {
 	tests := []struct {
 		name        string