Complete framework characterization baseline

.gitignore

@@ -1,6 +1,5 @@
 # ---> Codex
 .codex
-AGENTS.md
 
 # ---> Go
 # If you prefer the allow list template instead of the deny list, see community template:

.woodpecker/release.yml

@@ -28,10 +28,6 @@ steps:
 
         build_binary linux amd64 ""
         build_binary linux arm64 ""
-        build_binary darwin amd64 ""
-        build_binary darwin arm64 ""
-        build_binary windows amd64 ".exe"
-        build_binary windows arm64 ".exe"
 
   - name: publish-release
     image: woodpeckerci/plugin-release

AGENTS.md

@@ -0,0 +1 @@
+Please review `docs/development.md` for initial orientation in this repository and follow its task-specific reading guide.

LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2026 eric. 
+Copyright (c) 2026 Eric Rakestraw. 
 
 Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
 

README.md

@@ -1,449 +1,45 @@
 # scriptorium
 
-Scriptorium is a generic prompt execution engine.
+Scriptorium is a narrow prompt-execution application for rendering prompt
+requests, running them against OpenAI-compatible chat-completions endpoints, and
+serving the same run workflow over HTTP.
 
-It takes:
-- a prompt definition
-- a selected or default execution profile
-- named input artifacts
-- template variables
-- optional runtime overrides
+It keeps prompt definitions, execution profiles, schemas, and input artifacts as
+separate files so prompts can be reviewed and reused without baking model
+runtime settings into application code.
 
-It returns:
-- for `run`: generated artifact, validation result, metadata
-- for `render`: prepared/rendered prompt data (no model output)
+## Quickstart
 
-## Prompt vs Profile
+From the repository root:
 
-Scriptorium separates **what** to do (Prompt) from **how** to do it (Profile).
-
-### Prompt Definition
-Defines the task logic and output contract.
-- Task description and version.
-- Message templates (system, user, etc.).
-- Required and optional input artifacts.
-- Output format and validation rules.
-- Repair settings for structured output.
-- Optional `default_profile` for convenience.
-
-### Execution Profile
-Defines the runtime environment and model settings.
-- LLM endpoint (URL).
-- Model name.
-- Generation parameters: `temperature`, `max_tokens`, `top_p`.
-- Runtime settings: `timeout`, `reasoning_effort`.
-- API key source via `api_key_env`.
-
-Callers can explicitly provide a `profile_id` to override the prompt's `default_profile`.
-
-## Precedence
-
-Scriptorium uses two precedence layers:
-
-### Application Configuration Precedence
-
-For application-level adapter settings (for example prompt/profile/schema directories, server address, and render output default), precedence is:
-
-1. **CLI Flags**
-2. **`config.yml`**
-3. **Built-in application defaults**
-
-Application config loading behavior:
-- Default config path: `/etc/scriptorium/config.yml`
-- Override path: `--config <PATH>` (supported by `run`, `render`, and `serve`)
-- If `--config` is provided, the file must exist and be valid.
-- If `--config` is omitted, missing `/etc/scriptorium/config.yml` is allowed.
-
-### Runtime Model Precedence
-
-When resolving runtime model settings, Scriptorium follows this precedence model (highest to lowest):
-
-1. **Runtime Overrides**: Provided via CLI flags or HTTP request `model` object.
-2. **Execution Profile**: Settings defined in the selected profile.
-3. **Application Defaults**: Built-in fallback values.
-
-### Profile Selection Logic
-The engine determines which profile to use in this order:
-1. Explicit `profile_id` (via `--profile` or HTTP request).
-2. The `default_profile` named in the Prompt Definition.
-3. Error: If neither is provided and no default exists.
-
-## API Key Policy
-
-To ensure security, Scriptorium does not support raw API keys in configuration files, CLI arguments, or HTTP requests.
-
-- **`api_key_env`**: Profiles and overrides specify the name of an environment variable (e.g., `SCRIPTORIUM_API_KEY`).
-- **Runtime Resolution**: The value of the environment variable is read directly from the process environment at runtime.
-- **Zero Leakage**: API key values are never included in metadata, logs, or response bodies.
-
-## CLI Usage
-
-Available CLI commands:
-- `scriptorium run`
-- `scriptorium render`
-- `scriptorium serve`
-
-All commands accept `--config <PATH>`.
-
-`prompt_dir` and `profile_dir` may be supplied by CLI flags or `config.yml`:
-- `--prompt-dir` or `config.yml` `prompt_dir`
-- `--profile-dir` or `config.yml` `profile_dir`
-
-`schema_dir` and `serve` `addr` may also be supplied by `config.yml` where applicable:
-- `--schema-dir` or `config.yml` `schema_dir`
-- `--addr` or `config.yml` `server.addr`
-
-### `scriptorium run`
-
-Runs a single prompt execution.
-
-**Required Flags:**
-- `--prompt`: The prompt ID to execute.
-- `--input`: Input mapping `name=path` (repeatable).
-
-**Required Effective Settings:**
-- Prompt directory: `--prompt-dir` or `config.yml` `prompt_dir`
-- Profile directory: `--profile-dir` or `config.yml` `profile_dir`
-
-**Optional Flags:**
-- `--config`: Application config file path. Default discovery path is `/etc/scriptorium/config.yml`.
-- `--prompt-dir`: Override prompt directory from config.
-- `--profile-dir`: Override profile directory from config.
-- `--profile`: Override the prompt's default profile.
-- `--var`: Template variable `name=value` (repeatable).
-- `--out`: Write output to a file instead of stdout.
-- `--llm-base-url`: Override endpoint.
-- `--model`: Override model name.
-- `--api-key-env`: Override API key environment variable name.
-- `--temperature`: Override temperature.
-- `--max-tokens`: Override max tokens.
-- `--top-p`: Override top_p.
-- `--timeout`: Override request timeout (e.g., `30s`, `1m`).
-- `--schema-dir`: Base directory for validation schemas.
-
-**Examples:**
-
-Using `config.yml` for prompt/profile directories:
 ```bash
-scriptorium run \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Overriding config directories explicitly:
-```bash
-scriptorium run \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Overriding the profile:
-```bash
-scriptorium run \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --profile local-quality \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Overriding model and runtime values:
-```bash
-scriptorium run \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --model gpt-4o \
-  --temperature 0.7 \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Using a local OpenAI-compatible vLLM endpoint:
-```bash
-scriptorium run \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --llm-base-url http://localhost:8000/v1 \
-  --model meta-llama-3-8b \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-### `scriptorium render`
-
-Prepares and renders a prompt without calling the LLM.
-
-`render` uses the same prompt/profile/input/variable/runtime override resolution as `run`:
-- Profile selection precedence: `--profile` -> prompt `default_profile` -> error.
-- Runtime precedence: CLI runtime overrides -> selected profile -> built-in defaults.
-
-`render` is useful for debugging:
-- prompt template rendering
-- input mappings
-- selected profile behavior
-- runtime override behavior
-
-`render` does not:
-- call the LLM
-- validate model output
-- perform repair
-- expose resolved API key values
-
-It may include `api_key_env` names where relevant.
-
-**Required Flags:**
-- `--prompt`: The prompt ID to render.
-- `--input`: Input mapping `name=path` (repeatable).
-
-**Required Effective Settings:**
-- Prompt directory: `--prompt-dir` or `config.yml` `prompt_dir`
-- Profile directory: `--profile-dir` or `config.yml` `profile_dir`
-
-**Optional Flags:**
-- `--config`: Application config file path. Default discovery path is `/etc/scriptorium/config.yml`.
-- `--prompt-dir`: Override prompt directory from config.
-- `--profile-dir`: Override profile directory from config.
-- `--profile`: Override the prompt's default profile.
-- `--var`: Template variable `name=value` (repeatable).
-- `--out`: Write output to a file instead of stdout.
-- `--format`: Render output format (`text` or `json`). Default: `text`.
-- `--llm-base-url`: Runtime override for endpoint.
-- `--model`: Runtime override for model name.
-- `--api-key-env`: Runtime override for API key environment variable name.
-- `--temperature`: Runtime override for temperature.
-- `--max-tokens`: Runtime override for max tokens.
-- `--top-p`: Runtime override for top_p.
-- `--timeout`: Runtime override for timeout (e.g., `30s`, `1m`).
-
-**Render Output Formats:**
-- `text`: Human-readable output (default).
-- `json`: Machine-readable structured output.
-
-Render formatting is modular; additional output formats can be added later without changing prepare/run core logic.
-
-**Examples:**
-
-Default text output using `config.yml` directories:
-```bash
-scriptorium render \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Explicit config path:
-```bash
-scriptorium render \
+go run ./cmd/scriptorium render \
   --config ./examples/config.yml \
   --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Explicit directory overrides:
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
-
-Explicit JSON output:
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
   --input transcript=./examples/fixtures/transcript.md \
+  --input glossary=./examples/fixtures/glossary.yml \
   --format json
 ```
 
-Using prompt `default_profile` (omit `--profile`):
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md
-```
+This command renders the prepared prompt and effective runtime settings without calling an LLM.
+For complete invocation and output behavior, see the [CLI reference](docs/cli.md).
 
-Overriding profile selection:
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --profile local-quality \
-  --input transcript=./examples/fixtures/transcript.md
-```
+## Documentation
 
-Overriding runtime settings:
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md \
-  --llm-base-url http://localhost:8000/v1 \
-  --model gpt-4o-mini \
-  --temperature 0.2 \
-  --max-tokens 800 \
-  --top-p 1.0 \
-  --timeout 45s
-```
-
-Writing rendered output to a file:
-```bash
-scriptorium render \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --prompt generic.markdown_summary \
-  --input transcript=./examples/fixtures/transcript.md \
-  --format text \
-  --out ./rendered_prompt.txt
-```
-
-### `scriptorium serve`
-
-Starts the HTTP API.
-
-**Required Effective Settings:**
-- Prompt directory: `--prompt-dir` or `config.yml` `prompt_dir`
-- Profile directory: `--profile-dir` or `config.yml` `profile_dir`
-
-**Optional Flags:**
-- `--config`: Application config file path. Default discovery path is `/etc/scriptorium/config.yml`.
-- `--addr`: Listen address (default `:8080`).
-- `--schema-dir`: Base directory for validation schemas.
-
-**Examples:**
-
-Using `config.yml`:
-```bash
-scriptorium serve
-```
-
-Overriding config for local use:
-```bash
-scriptorium serve \
-  --prompt-dir ./prompts \
-  --profile-dir ./profiles \
-  --addr :9090
-```
-
-## HTTP API
-
-### `POST /v1/runs`
-
-Executes a prompt. No built-in authentication is provided; deploy behind a trusted gateway.
-
-**Request Body:**
-```json
-{
-  "prompt_id": "generic.structured_events",
-  "profile_id": "local-quality",
-  "include_raw_output": false,
-  "inputs": {
-    "transcript": {"type": "file", "uri": "./examples/fixtures/transcript.md"}
-  },
-  "vars": {
-    "session_date": "2026-05-04"
-  },
-  "model": {
-    "endpoint": "http://localhost:8000/v1",
-    "model": "gpt-4o-mini",
-    "temperature": 0.0
-  }
-}
-```
-
-`profile_id` is optional. If omitted, Scriptorium uses the prompt's `default_profile`. If neither is available, the run fails.
-
-**Response:**
-Returns a `200 OK` with the generated artifact, validation results, and metadata including the `prompt_id` and the `selected_profile_id`.
-
-**Validation Failures:**
-If the model output fails validation (e.g., invalid JSON), the API returns `200 OK` with `validation.status = "failed"`.
-
-**Raw Output Exposure:**
-- `raw_model_output` is omitted by default.
-- Set `include_raw_output: true` in the request to include it in the response.
-- Raw output is preserved internally in run results regardless of HTTP exposure.
-
-## Prompt Definition Authoring
-
-Prompts are defined in YAML.
-
-### Canonical Shape
-```yaml
-id: generic.structured_events
-version: "1.0.0"
-description: "Extracts structured events from a transcript"
-default_profile: local-quality
-
-inputs:
-  - name: transcript
-    required: true
-    content_type: text/markdown
-    description: "The raw session transcript"
-  - name: glossary
-    required: false
-    content_type: application/yaml
-    description: "Optional glossary terms"
-
-messages:
-  - role: system
-    content: "You are a helpful assistant."
-  - role: user
-    content_file: messages/extract_events.tmpl
-
-output:
-  format: json
-  validation_mode: json_schema
-  schema_path: structured_events.schema.json
-  repair_attempts: 2
-```
-
-**Key Features:**
-- **Inline vs File**: Use `content` for short prompts or `content_file` for larger templates. Exactly one must be set per message.
-- **Path Resolution**: `content_file` paths are resolved relative to the prompt YAML file.
-- **Inputs**: Mark inputs as `required` to ensure the runner fails early if they are missing.
-- **Input Metadata**: `content_type` is currently descriptive metadata and not enforced yet.
-- **Validation**: Support `none`, `basic`, `json`, and `json_schema`.
-- **Repair**: `repair_attempts` enables bounded retries to fix structured output.
-
-## Execution Profile Authoring
-
-Profiles are defined in YAML.
-
-### Canonical Shape
-```yaml
-id: local-quality
-endpoint: http://localhost:8000/v1
-model: gpt-4o
-temperature: 0.0
-max_tokens: 4096
-top_p: 1.0
-timeout_seconds: 300
-reasoning_effort: high
-api_key_env: SCRIPTORIUM_API_KEY
-```
-
-**Constraints:**
-- **No Raw Keys**: Do not include actual API keys. Only specify the environment variable name in `api_key_env`.
-- **Local Profiles**: For local endpoints that don't require auth, `api_key_env` can be omitted.
+- [CLI reference](docs/cli.md)
+- [Configuration reference](docs/config.md)
+- [HTTP API reference](docs/api.md)
+- [Operations guide](docs/operations.md)
+- [Consumer integration overview](docs/consumers/api.md)
+- [Go library package](docs/consumers/pkg-scriptorium.md)
+- [Subprocess integration](docs/integrations/subprocess.md)
+- [OpenAI-compatible chat integration](docs/integrations/openai-compatible-chat.md)
+- [Architecture policy](docs/policy/architecture.md)
 
 ## Examples
 
-- **Prompt Definitions**: `prompts/`
-- **Execution Profiles**: `profiles/`
-- **Schemas**: `schemas/`
-- **Fixtures**: `examples/fixtures/`
-
-## Build and Test
-
-```bash
-go build -o scriptorium ./cmd/scriptorium
-go test ./...
-```
+- [Minimal configuration](examples/config.yml) and [complete configuration](examples/config.full.yml)
+- [Prompt definitions](examples/prompts/), [execution profiles](examples/profiles/), [schemas](examples/schemas/), and [synthetic input fixtures](examples/fixtures/)
+- [Render script](examples/render-markdown-summary.sh)
+- [HTTP request](examples/http-run.json)
+- [Go library example](examples/go-library/prepare/main.go)

architecture.md

@@ -1,483 +0,0 @@
-# Scriptorium Architecture
-
-## 1. Purpose and Non-Goals
-
-Scriptorium is a prompt-definition execution engine.
-
-It accepts named input artifacts, renders prompt templates, calls an LLM, validates output, optionally performs bounded structured-output repair, and returns an artifact with metadata.
-
-Scriptorium also supports rendering/preparing a prompt without calling an LLM. This allows users to inspect the fully rendered prompt messages and effective runtime settings before executing a run.
-
-Scriptorium is not an orchestrator. It must not own transcription, transcript merge/polish steps, notifications, or cross-step workflow control.
-
-For the motivating D&D workflow:
-- Narratio orchestrates.
-- WhisperX transcribes.
-- Seriatim merges transcripts.
-- Audita polishes transcripts.
-- Scriptorium generates final artifacts from prepared inputs.
-
-Core Go code remains generic.
-
-## 2. Current Architecture
-
-Scriptorium uses a ports-and-adapters architecture to decouple the core execution logic from external dependencies.
-
-### Package Responsibilities
-
-- `cmd/scriptorium`: Binary entrypoint for CLI and HTTP server.
-- `internal/domain`: Core domain contracts, including `PromptDefinition`, `ExecutionProfile`, `PreparedRun`, `RunResult`, and related metadata.
-- `internal/usecase`: `Runner` use case logic, including prompt preparation, profile selection, runtime override resolution, full run execution, validation, and bounded repair.
-- `internal/config`: Application-level config model/loader for adapter settings (for example prompt/profile/schema directories, server address, and render format default).
-- `internal/promptdef`: Repository for loading and validating Prompt Definitions from the filesystem.
-- `internal/profile`: Repository for loading Execution Profiles from the filesystem.
-- `internal/artifact`: Input artifact resolution (`inline`, `file`).
-- `internal/prompt`: Template rendering via Go templates.
-- `internal/llm`: Provider-neutral client interface and OpenAI-compatible HTTP adapter.
-- `internal/validate`: Output validation implementation (`none/basic/json/json_schema`).
-- `internal/adapter/cli`: CLI flag parsing, command dispatch, and output handling.
-- `internal/adapter/http`: HTTP request/response mapping.
-- `internal/format` or equivalent: Formatting of prepared/rendered prompt output for CLI or other adapters, if formatting grows beyond simple CLI-local helpers.
-
-Exact package names may evolve, but the architectural boundaries should remain stable.
-
-### Application Configuration
-
-`config.yml` is adapter/application setup, not domain logic.
-
-Application config is intended for application-level settings such as:
-- `prompt_dir`
-- `profile_dir`
-- `schema_dir`
-- `server.addr`
-- `defaults.render_format`
-
-Application config precedence is:
-1. CLI flags
-2. `config.yml`
-3. Built-in application defaults
-
-Runtime model settings are intentionally separate:
-- Execution profiles and runtime overrides continue to own endpoint/model/runtime behavior.
-- `config.yml` does not replace execution profiles.
-
-The core use case (`Runner.Prepare`/`Runner.Run`) does not need to know whether adapter-level settings came from CLI flags or `config.yml`; it receives resolved dependencies and requests from adapters.
-
-## 3. Core Execution Model
-
-Scriptorium has two closely related execution paths:
-
-1. Prepare/render path.
-2. Full run path.
-
-The full run path should reuse the prepare path rather than duplicating its logic.
-
-### 3.1 Prepare / Render Data Flow
-
-The prepare path should be represented in the use case layer, preferably as `Runner.Prepare(ctx, RunRequest)` or an equivalent method.
-
-It executes all pre-LLM work:
-
-1. **Validate Request**: Ensure the request includes a prompt ID and all minimum required fields.
-2. **Load Prompt Definition**: Retrieve the `PromptDefinition` by ID from the prompt repository.
-3. **Select Profile**: Determine the `profile_id` using this precedence:
-   - Explicit `profile_id` in `RunRequest`.
-   - `default_profile` specified in the `PromptDefinition`.
-   - Error if neither is available.
-4. **Load Execution Profile**: Retrieve the `ExecutionProfile` from the profile repository.
-5. **Resolve Runtime Overrides**: Merge settings based on precedence, highest to lowest:
-   - Runtime overrides from CLI flags or HTTP request `model` object.
-   - Execution Profile settings.
-   - Built-in application defaults.
-6. **Resolve Artifacts**: Load all named input artifacts defined in the request.
-7. **Render Prompt**: Apply template variables and input artifacts to the prompt templates.
-8. **Compute Metadata**: Compute hashes, selected profile ID, prompt ID/version, effective runtime settings, input hashes, rendered prompt hash, and timing information as appropriate.
-9. **Return PreparedRun**: Return a `PreparedRun` containing the rendered messages, effective runtime settings, resolved metadata, and input/prompt hashes.
-
-The prepare path must not call the LLM.
-
-The prepare path must not validate model output, because there is no model output.
-
-The prepare path must not perform structured-output repair, because repair only applies after model output exists.
-
-The prepare path should not resolve or expose raw API key values. It may include the selected `api_key_env` name in effective runtime settings or metadata, but never the environment variable value.
-
-### 3.2 Full Run Data Flow
-
-The `Runner.Run(ctx, RunRequest)` flow should reuse the prepare path:
-
-1. **Prepare**: Call the shared prepare flow to load the prompt, select the profile, resolve artifacts, render prompt messages, and compute pre-run metadata.
-2. **Call LLM**: Execute the generation request using the effective runtime settings from the prepared run.
-3. **Build Output Artifact**: Convert the model response into the configured output artifact.
-4. **Validate Output**:
-   - Validate the model output against the prompt definition's output contract.
-   - Validation content failures remain successful run results with `validation.status=failed`.
-   - Validator runtime/config errors are run errors.
-5. **Repair If Configured**:
-   - If structured validation fails and `repair_attempts > 0`, perform bounded repair attempts.
-   - Re-validate after each repair attempt.
-   - Repair loops must remain strictly bounded.
-6. **Return RunResult**: Produce a `RunResult` containing the final artifact, validation status, raw model output, usage information, and metadata.
-
-`Runner.Run` should not duplicate profile selection, artifact resolution, or prompt rendering logic that already exists in `Runner.Prepare`.
-
-## 4. Domain Model
-
-Key domain types:
-
-- `PromptDefinition`: Defines the "what" of the task: templates, inputs, output contract, validation settings, repair settings, and optional `default_profile`.
-- `ExecutionProfile`: Defines the "how" of execution: endpoint, model, generation parameters, timeout, reasoning effort, and `api_key_env`.
-- `RunRequest`: The intent to execute or prepare a prompt, including `prompt_id`, optional `profile_id`, inputs, variables, and optional runtime overrides.
-- `PreparedRun`: The result of the prepare/render phase. Contains rendered messages, effective runtime settings, selected profile ID, prompt metadata, input hashes, prompt hash, and other pre-LLM metadata.
-- `RunResult`: The result of a full run. Contains the generated `Artifact`, `ValidationResult`, raw model output, token usage, and `RunMetadata`.
-- `RunMetadata`: Detailed tracing information, including prompt ID/version, selected profile ID, effective model parameters, usage tokens, hashes, timestamps, validation status, and repair attempts where applicable.
-- `RenderedPrompt`: Provider-neutral rendered prompt structure.
-- `RenderedMessage`: Provider-neutral rendered message with role and content.
-- `ArtifactRef`: A reference to an input artifact, such as `file` or `inline`.
-- `Artifact`: Loaded artifact content with name, content type, body, source URI, size, and hash.
-- `ValidationResult`: Validation status and details for full runs.
-
-`PreparedRun` should be serializable for JSON output and should also be representable in a human-readable text format.
-
-## 5. Interfaces and Adapters
-
-### Primary Ports
-
-- `promptdef.Repository`: Lookup for prompt definitions.
-- `profile.Repository`: Lookup for execution profiles.
-- `artifact.Reader`: Loading of artifact content.
-- `prompt.Renderer`: Template rendering.
-- `llm.Client`: Model generation.
-- `validate.Validator`: Output validation.
-- `format.PreparedRunFormatter` or equivalent: Optional formatting abstraction for rendered/prepared output.
-
-### Current Adapters
-
-- **Repositories**: Filesystem YAML loaders for both prompts and profiles.
-- **Artifact Reader**: Composite reader supporting `file` and `inline`.
-- **Prompt Renderer**: Go templates with a custom `input` helper.
-- **LLM Client**: OpenAI-compatible `/chat/completions` over HTTP.
-- **Validator**: Standard validator supporting `none`, `basic`, `json`, and `json_schema`.
-- **CLI Adapter**: Supports `run`, `render`, and `serve`.
-- **HTTP Adapter**: Supports full run execution through `POST /v1/runs`.
-
-## 6. Public Contracts
-
-### CLI
-
-Scriptorium should expose at least these commands:
-
-- `scriptorium run`: Executes a prompt by preparing it, calling the LLM, validating output, optionally repairing structured output, and returning an artifact.
-- `scriptorium render`: Prepares and renders a prompt without calling the LLM.
-- `scriptorium serve`: Starts the HTTP API using infrastructure-only flags.
-
-### `scriptorium run`
-
-`run` uses flags such as:
-
-- `--prompt-dir`
-- `--profile-dir`
-- `--prompt`
-- `--profile`
-- `--input`
-- `--var`
-- `--out`
-- runtime overrides such as `--model`, `--llm-base-url`, `--temperature`, `--max-tokens`, `--top-p`, `--timeout`, and `--api-key-env` if supported.
-
-`run` should produce the generated artifact as its primary output.
-
-### `scriptorium render`
-
-`render` prepares and renders a prompt without calling an LLM.
-
-It should use the same prompt/profile/input/variable/runtime override flags as `run` where applicable:
-
-- `--prompt-dir`
-- `--profile-dir`
-- `--prompt`
-- `--profile`
-- `--input`
-- `--var`
-- runtime overrides such as `--model`, `--llm-base-url`, `--temperature`, `--max-tokens`, `--top-p`, `--timeout`, and `--api-key-env` if supported.
-- `--format`, with initial support for `text` and `json`.
-
-Default render output format should be `text`.
-
-`render` must not call the LLM.
-
-`render` should show the same rendered messages and effective runtime settings that `run` would use.
-
-`render` should include enough information to debug:
-
-- prompt ID
-- prompt version
-- selected profile ID
-- effective runtime settings
-- input hashes
-- prompt hash
-- rendered messages
-
-`render` must not include resolved API key values.
-
-`render` may include the `api_key_env` name.
-
-### Render Output Formats
-
-Initial render output formats:
-
-- `text`: Human-readable default format.
-- `json`: Machine-readable structured representation of the prepared run.
-
-Additional formats, such as `markdown`, may be added later.
-
-Render output formatting should be modular. Adding a new output format should not require changing the prepare/run core logic.
-
-The output formatting layer should consume a `PreparedRun` and produce bytes or text for the adapter. It should not reload prompts, re-resolve artifacts, re-render templates, call the LLM, or perform validation.
-
-### `scriptorium serve`
-
-`serve` starts the HTTP API.
-
-It should use infrastructure-only flags such as:
-
-- `--addr`
-- `--prompt-dir`
-- `--profile-dir`
-- `--schema-dir`
-
-`serve` should not introduce a server-level model/runtime precedence layer unless explicitly documented and intentionally implemented.
-
-### HTTP API
-
-Current HTTP API:
-
-- `POST /v1/runs`: Accepts `RunRequest` JSON and returns `RunResponse` JSON. No built-in auth.
-
-Request may include runtime overrides under `model` and an `include_raw_output` boolean.
-
-`raw_model_output` is exposed only when explicitly requested with `include_raw_output=true`.
-
-A future HTTP prepare/render endpoint may be added, such as `POST /v1/renders` or `POST /v1/runs/prepare`, but the initial render feature may be CLI-only. If added later, it should call the same usecase-level prepare path as `scriptorium render`.
-
-### YAML Shapes
-
-Prompt YAML includes:
-
-- `id`
-- `version`
-- optional `default_profile`
-- `inputs`
-- `messages`
-- `output`
-
-Inputs support:
-
-- `name`
-- `required`
-- optional `content_type`
-- `description`
-
-Messages require:
-
-- `role`
-- exactly one of `content` or `content_file`
-
-`content_file` resolves relative to the prompt YAML location.
-
-Profile YAML includes:
-
-- `id`
-- `endpoint`
-- `model`
-- generation parameters
-- timeout settings
-- reasoning settings
-- `api_key_env`
-
-Prompt content must not appear in profile YAML.
-
-Model/runtime/API-key settings must not appear in prompt YAML, except that prompt YAML may specify `default_profile`.
-
-## 7. Render Feature Design
-
-The render feature is a first-class use case, not a CLI-only shortcut.
-
-### Goals
-
-The render feature should help users:
-
-- inspect fully rendered prompt messages
-- debug missing inputs
-- verify template variable substitution
-- verify selected profile resolution
-- verify runtime override precedence
-- verify file-backed prompt loading
-- inspect input hashes and prompt hashes
-- prepare for future token budgeting and prompt-size inspection
-
-### Non-Goals
-
-The render feature should not:
-
-- call an LLM
-- validate model output
-- repair structured output
-- resolve or print API key values
-- mutate artifacts
-- save outputs to artifact storage unless a future explicit output option is added
-- become an orchestration step manager
-
-### Usecase Shape
-
-The preferred usecase shape is:
-
-- `Runner.Prepare(ctx, RunRequest) (*PreparedRun, error)`
-- `Runner.Run(ctx, RunRequest) (*RunResult, error)`
-
-`Runner.Run` should call `Runner.Prepare`.
-
-The prepare flow should be the only implementation of:
-
-- prompt loading
-- profile selection
-- runtime override resolution
-- artifact resolution
-- prompt rendering
-- pre-run metadata/hash calculation
-
-### CLI Shape
-
-The preferred command name is `render`.
-
-The command should support:
-
-- `--format text`
-- `--format json`
-
-Default format:
-
-- `text`
-
-Unknown formats should produce a clear error.
-
-Formatting should be centralized through a small formatter registry, strategy, switch, or interface so new formats can be added without modifying usecase logic.
-
-### Text Output Expectations
-
-Text output should be optimized for human inspection.
-
-It should include, at minimum:
-
-- prompt ID and version
-- selected profile ID
-- model name
-- endpoint
-- effective generation settings
-- input hashes
-- rendered prompt hash
-- rendered messages grouped by role
-
-Text output should be readable and deterministic enough for tests.
-
-It should not include raw API key values.
-
-### JSON Output Expectations
-
-JSON output should be a structured representation of `PreparedRun` or a DTO derived from it.
-
-It should include, at minimum:
-
-- prompt ID and version
-- selected profile ID
-- effective runtime settings
-- input hashes
-- rendered prompt hash
-- rendered messages
-
-JSON output should not include raw API key values.
-
-JSON output should remain stable enough to be useful for automation and integration tests.
-
-## 8. Guardrails
-
-- **Separation of Concerns**: Prompt content must not belong in execution profiles; model/API settings must not belong in prompt definitions.
-- **Security**: Raw API keys are unsupported in all configuration and transport layers. Only `api_key_env` is used.
-- **Secret Handling**: Resolved API key values must never appear in rendered output, metadata, logs, HTTP responses, or CLI output.
-- **Path Resolution**: `content_file` paths in prompt definitions resolve relative to the prompt YAML file.
-- **Integrity**: No silent prompt truncation or omission of content.
-- **Reliability**: Repair loops are strictly bounded by `repair_attempts`.
-- **No Orchestration Creep**: Scriptorium prepares and executes a single prompt request. It does not coordinate multi-stage workflows.
-- **Render Reuse**: The full run path must reuse the prepare/render path to avoid divergent behavior.
-- **Formatter Isolation**: Render output formatters must not perform usecase work. They only format a completed `PreparedRun`.
-
-## 9. Testing Strategy
-
-Tests should protect both the run path and the prepare/render path.
-
-### Prepare / Render Tests
-
-Add tests for:
-
-- preparing a prompt with explicit profile selection
-- preparing a prompt using `default_profile`
-- failing when no explicit profile and no `default_profile` exist
-- runtime overrides beating profile values
-- profile values beating application defaults
-- file-backed prompt bodies rendering correctly
-- required inputs failing when missing
-- optional inputs being absent when not referenced
-- unknown input references failing
-- input hashes being included
-- rendered prompt hash being included
-- effective runtime settings being included
-- `api_key_env` name being included where appropriate
-- resolved API key values never appearing in `PreparedRun`
-- prepare path not calling the LLM
-
-### CLI Render Tests
-
-Add tests for:
-
-- `scriptorium render` mapping flags into `RunRequest`
-- default text output
-- explicit `--format text`
-- explicit `--format json`
-- unknown format failure
-- text output includes prompt/profile/messages
-- JSON output includes prompt/profile/messages
-- rendered output never includes resolved API key values
-
-### Run Reuse Tests
-
-Add tests proving:
-
-- `Runner.Run` reuses prepare behavior
-- run and render resolve the same prompt/profile/runtime settings for equivalent inputs
-- run still validates output
-- run still performs bounded repair where configured
-
-### Existing Tests
-
-Continue testing:
-
-- prompt definition loading
-- execution profile loading
-- artifact loading
-- prompt rendering
-- LLM adapter behavior
-- validation behavior
-- HTTP request/response mapping
-
-## 10. Extension Points
-
-Future work should remain grounded in the current architecture:
-
-- **Artifacts**: Add S3 artifact references via a new `artifact.Reader`.
-- **LLM**: Implement additional provider adapters, such as Anthropic or Google.
-- **Execution**: Add token budgeting, streaming generation, and batch execution capabilities.
-- **Prepare/Render**: Add token estimates, prompt-size summaries, or additional render output formats.
-- **Repositories**: Implement database-backed repositories for prompts and profiles.
-- **Profiles**: Support more granular profile versioning and environment-specific profiles.
-- **HTTP**: Add an HTTP prepare/render endpoint if Narratio or another caller needs it.
-
-Future render formats should plug into the formatter layer and should not require changes to the usecase layer.

convert.go

@@ -0,0 +1,406 @@
+package scriptorium
+
+import (
+	"reflect"
+
+	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+)
+
+func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) {
+	execution, err := toDomainExecutionTargetOverride(req.Execution)
+	if err != nil {
+		return domain.RunRequest{}, err
+	}
+	return domain.RunRequest{
+		PromptID:      req.PromptID,
+		PromptVersion: req.PromptVersion,
+		ProfileID:     req.ProfileID,
+		APIKey:        req.APIKey,
+		Inputs:        toDomainArtifactRefMap(req.Inputs),
+		Vars:          copyStringMap(req.Vars),
+		Execution:     execution,
+		Validation:    toDomainOutputContractPtr(req.Validation),
+		Metadata:      copyStringMap(req.Metadata),
+	}, nil
+}
+
+func fromDomainPreparedRun(prepared *domain.PreparedRun) *PreparedRun {
+	if prepared == nil {
+		return nil
+	}
+	return &PreparedRun{
+		PromptID:             prepared.PromptID,
+		PromptVersion:        prepared.PromptVersion,
+		PromptHash:           prepared.PromptHash,
+		SelectedProfileID:    prepared.SelectedProfileID,
+		EffectiveModelParams: fromDomainExecutionTarget(prepared.EffectiveModelParams),
+		OutputContract:       fromDomainOutputContract(prepared.OutputContract),
+		StructuredOutput:     fromDomainStructuredOutputSpec(prepared.StructuredOutput),
+		InputHashes:          copyStringMap(prepared.InputHashes),
+		SessionID:            prepared.SessionID,
+		RenderedPromptHash:   prepared.RenderedPromptHash,
+		Messages:             fromDomainRenderedMessages(prepared.Messages),
+		StartTime:            prepared.StartTime,
+		EndTime:              prepared.EndTime,
+		DurationMS:           prepared.DurationMS,
+	}
+}
+
+func fromDomainRunResult(result *domain.RunResult) *RunResult {
+	if result == nil {
+		return nil
+	}
+	return &RunResult{
+		RunID:                result.RunID,
+		Artifact:             fromDomainArtifact(result.Artifact),
+		RawOutput:            result.RawOutput,
+		Validation:           fromDomainValidationResult(result.Validation),
+		PromptID:             result.PromptID,
+		PromptVersion:        result.PromptVersion,
+		PromptHash:           result.PromptHash,
+		RenderedPromptHash:   result.RenderedPromptHash,
+		SelectedProfileID:    result.SelectedProfileID,
+		ModelName:            result.ModelName,
+		Endpoint:             result.Endpoint,
+		EffectiveModelParams: fromDomainExecutionTarget(result.EffectiveModelParams),
+		InputHashes:          copyStringMap(result.InputHashes),
+		Usage:                fromDomainTokenUsage(result.Usage),
+		StartTime:            result.StartTime,
+		EndTime:              result.EndTime,
+		Duration:             result.Duration,
+	}
+}
+
+func fromDomainGenerateRequest(req domain.GenerateRequest) GenerateRequest {
+	return GenerateRequest{
+		Prompt:           fromDomainRenderedPrompt(req.Prompt),
+		Target:           fromDomainExecutionTarget(req.Target),
+		TargetPresence:   fromDomainExecutionTargetPresence(req.TargetPresence),
+		StructuredOutput: fromDomainStructuredOutputSpec(req.StructuredOutput),
+		APIKey:           req.Target.APIKey,
+	}
+}
+
+func toDomainGenerateResponse(resp *GenerateResponse) *domain.GenerateResponse {
+	if resp == nil {
+		return nil
+	}
+	return &domain.GenerateResponse{
+		Content: resp.Content,
+		Usage:   toDomainTokenUsage(resp.Usage),
+	}
+}
+
+func fromDomainRenderedPrompt(prompt domain.RenderedPrompt) RenderedPrompt {
+	return RenderedPrompt{
+		SessionID: prompt.SessionID,
+		Messages:  fromDomainRenderedMessages(prompt.Messages),
+	}
+}
+
+func toDomainArtifactRefMap(src map[string]ArtifactRef) map[string]domain.ArtifactRef {
+	if src == nil {
+		return nil
+	}
+	out := make(map[string]domain.ArtifactRef, len(src))
+	for k, v := range src {
+		out[k] = toDomainArtifactRef(v)
+	}
+	return out
+}
+
+func toDomainArtifactRef(ref ArtifactRef) domain.ArtifactRef {
+	return domain.ArtifactRef{
+		Type: domain.ArtifactRefType(ref.Type),
+		URI:  ref.URI,
+		Body: ref.Body,
+	}
+}
+
+func fromDomainArtifact(artifact domain.Artifact) Artifact {
+	return Artifact{
+		Name:        artifact.Name,
+		ContentType: artifact.ContentType,
+		Body:        copyBytes(artifact.Body),
+		URI:         artifact.URI,
+		Size:        artifact.Size,
+		Hash:        artifact.Hash,
+	}
+}
+
+func toDomainExecutionTargetOverride(override *ExecutionTargetOverride) (*domain.ExecutionTargetOverride, error) {
+	if override == nil {
+		return nil, nil
+	}
+	extraParams, err := copyPublicJSONMap(override.ExtraParams)
+	if err != nil {
+		return nil, err
+	}
+	return &domain.ExecutionTargetOverride{
+		Endpoint:        override.Endpoint,
+		Model:           override.Model,
+		Temperature:     copyFloat64Ptr(override.Temperature),
+		MaxTokens:       copyIntPtr(override.MaxTokens),
+		TopP:            copyFloat64Ptr(override.TopP),
+		TimeoutSeconds:  copyIntPtr(override.TimeoutSeconds),
+		ServiceTier:     override.ServiceTier,
+		ReasoningEffort: override.ReasoningEffort,
+		APIKeyEnv:       override.APIKeyEnv,
+		ExtraParams:     extraParams,
+	}, nil
+}
+
+func fromDomainExecutionTarget(target domain.ExecutionTarget) ExecutionTarget {
+	return ExecutionTarget{
+		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:     copyAnyMap(target.ExtraParams),
+	}
+}
+
+func fromDomainExecutionTargetPresence(presence domain.ExecutionTargetPresence) ExecutionTargetPresence {
+	return ExecutionTargetPresence{
+		Temperature:    presence.Temperature,
+		MaxTokens:      presence.MaxTokens,
+		TopP:           presence.TopP,
+		TimeoutSeconds: presence.TimeoutSeconds,
+	}
+}
+
+func toDomainOutputContractPtr(contract *OutputContract) *domain.OutputContract {
+	if contract == nil {
+		return nil
+	}
+	out := toDomainOutputContract(*contract)
+	return &out
+}
+
+func toDomainOutputContract(contract OutputContract) domain.OutputContract {
+	return domain.OutputContract{
+		Format:         domain.OutputFormat(contract.Format),
+		ValidationMode: domain.ValidationMode(contract.ValidationMode),
+		SchemaPath:     contract.SchemaPath,
+		RepairAttempts: contract.RepairAttempts,
+	}
+}
+
+func fromDomainOutputContract(contract domain.OutputContract) OutputContract {
+	return OutputContract{
+		Format:         OutputFormat(contract.Format),
+		ValidationMode: ValidationMode(contract.ValidationMode),
+		SchemaPath:     contract.SchemaPath,
+		RepairAttempts: contract.RepairAttempts,
+	}
+}
+
+func fromDomainValidationResult(result domain.ValidationResult) ValidationResult {
+	return ValidationResult{
+		Status:         ValidationStatus(result.Status),
+		Mode:           ValidationMode(result.Mode),
+		Errors:         copyStringSlice(result.Errors),
+		SchemaPath:     result.SchemaPath,
+		RepairAttempts: result.RepairAttempts,
+		IsValid:        result.IsValid,
+	}
+}
+
+func fromDomainTokenUsage(usage domain.TokenUsage) TokenUsage {
+	return TokenUsage{
+		PromptTokens:     usage.PromptTokens,
+		CompletionTokens: usage.CompletionTokens,
+		TotalTokens:      usage.TotalTokens,
+		CachedTokens:     usage.CachedTokens,
+		CacheWriteTokens: usage.CacheWriteTokens,
+	}
+}
+
+func toDomainTokenUsage(usage TokenUsage) domain.TokenUsage {
+	return domain.TokenUsage{
+		PromptTokens:     usage.PromptTokens,
+		CompletionTokens: usage.CompletionTokens,
+		TotalTokens:      usage.TotalTokens,
+		CachedTokens:     usage.CachedTokens,
+		CacheWriteTokens: usage.CacheWriteTokens,
+	}
+}
+
+func fromDomainRenderedMessages(messages []domain.RenderedMessage) []RenderedMessage {
+	if messages == nil {
+		return nil
+	}
+	out := make([]RenderedMessage, len(messages))
+	for i, msg := range messages {
+		out[i] = RenderedMessage{
+			Role:         msg.Role,
+			Content:      msg.Content,
+			CacheControl: fromDomainCacheControl(msg.CacheControl),
+		}
+	}
+	return out
+}
+
+func fromDomainCacheControl(cacheControl *domain.CacheControl) *CacheControl {
+	if cacheControl == nil {
+		return nil
+	}
+	return &CacheControl{
+		Type: CacheControlType(cacheControl.Type),
+		TTL:  cacheControl.TTL,
+	}
+}
+
+func fromDomainStructuredOutputSpec(spec *domain.StructuredOutputSpec) *StructuredOutputSpec {
+	if spec == nil {
+		return nil
+	}
+	out := &StructuredOutputSpec{
+		Type: StructuredOutputType(spec.Type),
+	}
+	if spec.JSONSchema != nil {
+		out.JSONSchema = &StructuredOutputJSONSpec{
+			Name:   spec.JSONSchema.Name,
+			Strict: spec.JSONSchema.Strict,
+			Schema: copyAny(spec.JSONSchema.Schema),
+		}
+	}
+	return out
+}
+
+func copyStringMap(src map[string]string) map[string]string {
+	if src == nil {
+		return nil
+	}
+	out := make(map[string]string, len(src))
+	for k, v := range src {
+		out[k] = v
+	}
+	return out
+}
+
+func copyAnyMap(src map[string]any) map[string]any {
+	if src == nil {
+		return nil
+	}
+	out := make(map[string]any, len(src))
+	for k, v := range src {
+		out[k] = copyAny(v)
+	}
+	return out
+}
+
+func copyAny(value any) any {
+	if value == nil {
+		return nil
+	}
+	switch v := value.(type) {
+	case map[string]any:
+		return copyAnyMap(v)
+	case []any:
+		out := make([]any, len(v))
+		for i, item := range v {
+			out[i] = copyAny(item)
+		}
+		return out
+	case []string:
+		return copyStringSlice(v)
+	case []byte:
+		return copyBytes(v)
+	default:
+		return copyReflectValue(reflect.ValueOf(value)).Interface()
+	}
+}
+
+func copyReflectValue(value reflect.Value) reflect.Value {
+	if !value.IsValid() {
+		return value
+	}
+
+	switch value.Kind() {
+	case reflect.Interface:
+		if value.IsNil() {
+			return reflect.Zero(value.Type())
+		}
+		copied := copyReflectValue(value.Elem())
+		if copied.IsValid() && copied.Type().AssignableTo(value.Type()) {
+			return copied
+		}
+		out := reflect.New(value.Type()).Elem()
+		out.Set(copied)
+		return out
+	case reflect.Pointer:
+		if value.IsNil() {
+			return reflect.Zero(value.Type())
+		}
+		out := reflect.New(value.Type().Elem())
+		out.Elem().Set(copyReflectValue(value.Elem()))
+		return out
+	case reflect.Map:
+		if value.IsNil() {
+			return reflect.Zero(value.Type())
+		}
+		out := reflect.MakeMapWithSize(value.Type(), value.Len())
+		iter := value.MapRange()
+		for iter.Next() {
+			out.SetMapIndex(copyReflectValue(iter.Key()), copyReflectValue(iter.Value()))
+		}
+		return out
+	case reflect.Slice:
+		if value.IsNil() {
+			return reflect.Zero(value.Type())
+		}
+		out := reflect.MakeSlice(value.Type(), value.Len(), value.Cap())
+		for i := 0; i < value.Len(); i++ {
+			out.Index(i).Set(copyReflectValue(value.Index(i)))
+		}
+		return out
+	case reflect.Array:
+		out := reflect.New(value.Type()).Elem()
+		for i := 0; i < value.Len(); i++ {
+			out.Index(i).Set(copyReflectValue(value.Index(i)))
+		}
+		return out
+	default:
+		return value
+	}
+}
+
+func copyStringSlice(src []string) []string {
+	if src == nil {
+		return nil
+	}
+	out := make([]string, len(src))
+	copy(out, src)
+	return out
+}
+
+func copyBytes(src []byte) []byte {
+	if src == nil {
+		return nil
+	}
+	out := make([]byte, len(src))
+	copy(out, src)
+	return out
+}
+
+func copyFloat64Ptr(src *float64) *float64 {
+	if src == nil {
+		return nil
+	}
+	v := *src
+	return &v
+}
+
+func copyIntPtr(src *int) *int {
+	if src == nil {
+		return nil
+	}
+	v := *src
+	return &v
+}

docs/adr/0001-adopt-canonical-documentation-ownership.md

@@ -0,0 +1,48 @@
+# ADR 0001: Adopt Canonical Documentation Ownership
+
+## Status
+
+Accepted
+
+## Date
+
+2026-07-26
+
+## Context
+
+Scriptorium's documentation grew alongside its CLI, HTTP, public Go, and
+integration interfaces. As a result, several documents repeated mutable
+contracts such as flags, configuration fields, and status behavior. Those
+parallel definitions made it unclear which document to update when behavior
+changed and increased the risk of documentation drift.
+
+## Decision
+
+Assign each documentation topic one canonical owner, as defined in
+[`docs/policy/documentation.md`](../policy/documentation.md). Non-owning
+documents may provide short orientation and links, but do not redefine volatile
+contracts. Current behavior is documented outside `docs/roadmap/`; roadmaps own
+future work, sequencing, and implementation status.
+
+## Alternatives Considered
+
+- Keep broad reference material in several audience-specific documents. This
+  would preserve local convenience but leave conflicting contract definitions
+  likely.
+- Consolidate all documentation into one reference. This would reduce duplicate
+  text but would not serve the distinct needs of users, operators, consumers,
+  and contributors.
+
+## Rationale
+
+Canonical ownership retains audience-specific guidance while making the source
+of truth for each contract discoverable. It also makes documentation changes
+reviewable alongside the implementation change that requires them.
+
+## Consequences
+
+- Changes to behavior must update the canonical owner in the same change.
+- Cross-cutting documentation links to the owner instead of copying its
+  details.
+- Documentation restructuring followed a dedicated implementation roadmap;
+  repository history, not this ADR, records its completion.

docs/adr/0002-split-promptkit-from-scriptorium.md

@@ -0,0 +1,288 @@
+# ADR 0002: Split Promptkit From Scriptorium
+
+## Status
+
+Accepted
+
+## Date
+
+2026-07-26
+
+## Context
+
+Scriptorium currently combines two products in one Go module:
+
+- a reusable prompt-execution framework with a public Go facade; and
+- a runnable application with CLI and HTTP interfaces.
+
+Downstream Go projects increasingly import the framework directly and do not
+use the executable interfaces. Keeping both products in one module couples
+framework releases, dependencies, documentation, and public API evolution to
+application-specific transport concerns.
+
+Promptkit will become the framework project, and Scriptorium will become a slim
+application that consumes it. This ADR records that end-state boundary. It does
+not assert that the split has been implemented; until then, the current
+repository structure and contracts remain authoritative.
+
+## Decision
+
+### Projects And Module Paths
+
+Create a repository named `promptkit` alongside Scriptorium:
+
+| Project | Repository and Go module path | Root Go package |
+| --- | --- | --- |
+| Promptkit | `gitea.maximumdirect.net/eric/promptkit` | `promptkit` |
+| Scriptorium | `gitea.maximumdirect.net/eric/scriptorium` | No reusable root facade after migration |
+
+Promptkit will expose its supported public API from the module root. Its
+implementation packages will remain under `internal/` unless a real consumer
+extension point requires a public type or interface.
+
+Scriptorium will import only Promptkit's supported public packages. It will not
+import Promptkit implementation packages or reproduce Promptkit orchestration.
+
+### Product Responsibilities
+
+Promptkit owns application-neutral framework behavior:
+
+- the engine and its `Prepare` and `Run` workflow;
+- public request, result, profile, option, extension, and error APIs;
+- prompt-definition loading and rendering;
+- profile loading, overlays, and the embedded built-in profile registry;
+- schema loading and output validation;
+- provider-neutral model-client boundaries and the OpenAI-compatible client;
+- artifact types, artifact-reader injection, and general-purpose inline and
+  caller-selected file readers;
+- execution-setting resolution and framework defaults; and
+- framework-level secret redaction and error classification.
+
+Scriptorium owns executable and transport behavior:
+
+- the `scriptorium` process and its `run`, `render`, and `serve` commands;
+- CLI parsing, streams, output files, formatting, exit codes, and process
+  cancellation behavior;
+- application-configuration discovery and CLI-over-configuration precedence;
+- HTTP routing, strict request decoding, DTO mapping, response encoding,
+  status codes, and transport limits;
+- HTTP artifact-root containment and deployment policy;
+- server construction, server defaults, and process logging; and
+- executable release artifacts.
+
+The dependency direction is:
+
+```text
+Scriptorium CLI and HTTP adapters
+                |
+                v
+       Promptkit public API
+                |
+                v
+  injected sources, readers, and model clients
+```
+
+### Current Package Disposition
+
+Implementation may reorganize files during extraction, but each current package
+has this target owner:
+
+| Current package or file group | Target owner | Disposition |
+| --- | --- | --- |
+| Root `scriptorium` facade files and tests | Promptkit | Move and rename the public package to `promptkit`; Scriptorium retains no compatibility facade. |
+| `internal/domain`, `internal/usecase` | Promptkit | Move as internal engine implementation. |
+| `internal/promptdef`, `internal/prompt` | Promptkit | Move as internal prompt loading and rendering. |
+| `internal/profile`, `internal/profile/builtin` | Promptkit | Move with embedded built-in assets and registry tests. |
+| `internal/filecatalog` | Promptkit | Move as source-loading support. |
+| `internal/validate` | Promptkit | Move as schema and output-validation implementation. |
+| `internal/llm` | Promptkit | Move with the OpenAI-compatible integration. |
+| `internal/artifact` | Split | Move general inline/file reading to Promptkit; keep rooted, denied, and byte-limited HTTP file reading in Scriptorium behind a Promptkit reader interface. |
+| `internal/defaults` | Split | Move framework, execution, output-artifact, content-type, and model-client defaults to Promptkit; keep CLI, HTTP, and server defaults in Scriptorium. |
+| `internal/adapter/cli`, `internal/adapter/http` | Scriptorium | Keep and refactor to use Promptkit's public API. |
+| `internal/config` | Scriptorium | Keep application settings, discovery, validation, and CLI precedence. |
+| `internal/format` | Scriptorium | Keep prepared-run presentation, rewritten against Promptkit public values. |
+| `cmd/scriptorium` | Scriptorium | Keep as the process entrypoint. |
+
+Tests move with the behavior they protect. Cross-boundary tests will live with
+the consuming side: Promptkit protects framework contracts, while Scriptorium
+protects adapter mapping, HTTP containment, and executable behavior.
+
+### Public Boundary
+
+Promptkit's initial facade will preserve the useful shape of the current
+Scriptorium Go API where that reduces extraction risk. It will expose only the
+capabilities required by Promptkit consumers and by Scriptorium:
+
+- engine construction, preparation, and execution;
+- public request, result, profile, and error values;
+- prompt, profile, schema, artifact-reader, validator, and model-client source
+  or injection options that have demonstrated consumers; and
+- enough stable error identity for Scriptorium to map CLI and HTTP outcomes.
+
+Promptkit will not export its domain package, runner implementation,
+repositories, adapter DTOs, or general internal constructors merely to
+simplify the move.
+
+Scriptorium's CLI and HTTP adapters will depend on a small consumer-facing
+`Prepare`/`Run` interface where test substitution is needed. That interface
+belongs at the consuming boundary rather than forcing adapter concepts into
+Promptkit.
+
+### Artifact Reading And HTTP Containment
+
+Promptkit will define the artifact-reader extension point used during
+preparation. Its ordinary file reader may read a path deliberately supplied by
+an in-process or CLI caller and does not claim to be a deployment sandbox.
+
+Scriptorium will implement the HTTP-specific reader that:
+
+- denies file references when no artifact root is configured;
+- applies the configured artifact byte limit;
+- enforces Scriptorium's documented lexical root-containment rule; and
+- maps reader failures to Scriptorium HTTP error responses.
+
+Scriptorium will inject that reader through Promptkit's public construction
+boundary. Promptkit will not know about HTTP roots, status codes, request DTOs,
+or deployment policy.
+
+### Configuration And Default Ownership
+
+Configuration ownership follows the behavior configured, not the current file
+location:
+
+| Configuration category | Owner |
+| --- | --- |
+| Application configuration discovery, configuration-file precedence, `prompt_dir`, `profile_dir`, and `schema_dir` | Scriptorium |
+| CLI flags and their mapping to application settings or request overrides | Scriptorium |
+| `server.*`, render-output settings, HTTP byte limits, and server defaults | Scriptorium |
+| Prompt-definition, profile, and output-contract file formats | Promptkit |
+| Prompt/profile source selection, overlays, schema behavior, and built-in profiles | Promptkit |
+| Execution settings, presence-aware request overrides, and execution defaults | Promptkit |
+| Built-in OpenAI-compatible client settings, timeout behavior, and provider wire mapping | Promptkit |
+| HTTP request and response fields, including their mapping to framework values | Scriptorium |
+
+Scriptorium will translate its application settings and external request
+values into Promptkit construction options and requests. When an omitted
+Scriptorium setting means “use the framework default,” Scriptorium will omit
+the override rather than copy Promptkit's numeric default.
+
+### Compatibility And Versioning
+
+This migration is intentionally breaking:
+
+- new Go consumers will import `gitea.maximumdirect.net/eric/promptkit`;
+- Scriptorium will not provide aliases, forwarding wrappers, or deprecated
+  compatibility packages for its former Go facade;
+- existing consumers may remain pinned to the final framework-bearing
+  Scriptorium tag until migrated; and
+- intermediate migration phases need not preserve source compatibility, but
+  each merged phase must be internally buildable and tested.
+
+Promptkit's first release will be `v0.1.0`. During the migration, incompatible
+Promptkit changes may advance its minor version until a stable `v1` contract is
+declared. The first slim Scriptorium release will advance the Scriptorium minor
+version beyond the final framework-bearing release. Normal semantic-versioning
+rules apply independently to both projects after the migration.
+
+Promptkit must be tagged before Scriptorium or another consumer publishes a
+release that depends on it. Release branches must use tagged module
+dependencies, not local replacements or unpublished revisions.
+
+### Local Development And Cross-Repository Coordination
+
+For coordinated local work, place both repositories in a temporary Go
+workspace or use an uncommitted module replacement. `go.work`,
+`go.work.sum`, and local filesystem `replace` directives must not be committed
+to release branches.
+
+Cross-repository changes follow this order:
+
+1. land and tag the required Promptkit capability;
+2. update Scriptorium and other consumers to that tag;
+3. run each repository's own CI and smoke checks; and
+4. release consumers only after the Promptkit tag is available.
+
+Migration coordination must confirm out-of-band repository creation, Promptkit
+tags, and downstream migrations before dependent work proceeds.
+Cross-repository changes are coordinated, not treated as atomic commits.
+
+### Documentation And Maintained Assets
+
+Each repository will maintain its own README, contributor guide, architecture,
+documentation, testing, release, and operations material appropriate to that
+project. Cross-project documents will link to the canonical owner rather than
+copy its contract.
+
+Existing documentation and maintained assets have these target owners:
+
+| Current material | Target owner |
+| --- | --- |
+| Current README and executable quickstart | Scriptorium; Promptkit creates its own framework orientation |
+| Public Go package and Go-consumer guidance | Promptkit |
+| Prompt, profile, schema, execution-setting, and framework credential reference | Promptkit |
+| OpenAI-compatible integration contract and framework internal documents | Promptkit |
+| CLI, HTTP API, subprocess, and Scriptorium operations contracts | Scriptorium |
+| Consumer interface overview | Scriptorium, revised to route Go consumers to Promptkit |
+| Application-configuration discovery, server settings, and adapter internals | Scriptorium |
+| Current internal overview and source documentation | Split into repository-local overviews; Promptkit owns framework sources and Scriptorium owns HTTP containment |
+| This ADR and cross-project migration records | Scriptorium |
+| `examples/go-library` | Promptkit |
+| `examples/config*.yml`, `examples/render-markdown-summary.sh`, and `examples/http-run.json` | Scriptorium |
+| Example prompts, profiles, schemas, and synthetic fixtures used by the executable examples | Scriptorium |
+| Embedded built-in profile assets | Promptkit |
+| Scriptorium release workflow and executable packaging | Scriptorium |
+| Repository-level license, ignore rules, agent guidance, and development policies | Each repository maintains its own applicable copy |
+
+Promptkit will create or retain its own minimal framework examples and test
+fixtures rather than making either repository's tests depend on the other's
+working tree. Scriptorium's framework-format documentation will become a short
+version-appropriate link to Promptkit, while its maintained executable examples
+remain self-contained.
+
+## Alternatives Considered
+
+- Keep the current combined repository and improve package naming. This avoids
+  migration work but retains release and ownership coupling between the
+  framework and executable.
+- Add Promptkit as a wrapper around the Scriptorium public package. This gives
+  consumers a new import path but leaves framework ownership and dependency
+  direction inverted.
+- Extract Promptkit while retaining a Scriptorium compatibility facade. This
+  reduces immediate consumer changes but creates a second public API surface
+  and prolongs duplicate maintenance.
+- Move all artifact reading into Promptkit. This would place HTTP containment,
+  byte limits, and deployment policy in the application-neutral framework.
+- Keep Promptkit and Scriptorium as separate modules in one repository. This
+  separates imports but not repository permissions, release workflows,
+  issue ownership, or independent project evolution.
+
+## Rationale
+
+A separate Promptkit project makes the reusable framework the direct owner of
+the API that downstream Go projects already consume. Keeping Scriptorium as a
+public-API consumer exercises the same boundary as other consumers and prevents
+its adapters from relying on framework internals.
+
+The selected split keeps transport and deployment policy close to the
+Scriptorium interfaces that expose it, while allowing Promptkit to remain
+useful to in-process consumers with different IO and security requirements.
+Explicit package, configuration, documentation, and asset ownership reduces
+ambiguity during extraction and after release.
+
+## Consequences
+
+- All Go consumers of the framework must change their import path.
+- Promptkit and Scriptorium gain independent issue, release, CI, policy, and
+  documentation lifecycles.
+- Scriptorium becomes a real downstream integration test of Promptkit's public
+  facade.
+- Framework changes that affect Scriptorium require tagged, ordered
+  cross-repository coordination.
+- Some current packages, especially artifact reading and defaults, must be
+  separated by responsibility rather than moved intact.
+- Scriptorium's current configuration and documentation references must be
+  split between application and framework owners.
+- Maintainers must inventory and migrate downstream consumers explicitly; no
+  compatibility facade will hide incomplete migration.
+- Until the split is implemented, the current repository structure and
+  contracts remain authoritative.

docs/api.md

@@ -0,0 +1,145 @@
+# HTTP API Reference
+
+This is the canonical public HTTP contract for Scriptorium.
+
+## Service And Route
+
+`POST /v1/runs` runs one prompt request and returns generated output,
+validation, and metadata. The service has no built-in authentication or
+authorization; deploy it behind appropriate network and authentication controls.
+
+The service address and HTTP limits are configured as described in the
+[configuration reference](config.md). `serve` invocation is defined in the
+[CLI reference](cli.md).
+
+Requests and responses are JSON objects. Requests are decoded as JSON regardless
+of their `Content-Type`; successful JSON responses use
+`Content-Type: application/json`. There are no query parameters.
+
+## Request Limits
+
+The configured request-body limit includes inline artifact bodies. The artifact
+limit applies to HTTP `file` inputs. The response limit applies to the encoded
+response, including the artifact body and optional raw output. A limit of zero
+disables that limit.
+
+A request body over its limit returns `413 request_too_large`; an oversized
+file input returns `413 artifact_too_large`; an oversized encoded response
+returns `413 response_too_large`.
+
+## `POST /v1/runs`
+
+### Request Body
+
+The maintained [request example](../examples/http-run.json) is a complete
+copyable shape. The smallest valid shape is:
+
+```json
+{
+  "prompt_id": "generic.markdown_summary",
+  "inputs": {
+    "transcript": {"type": "inline", "body": "Source text"}
+  }
+}
+```
+
+| Field | Required | Meaning |
+| --- | --- | --- |
+| `prompt_id` | yes | Non-blank prompt ID. |
+| `prompt_version` | no | Prompt version filter. |
+| `profile_id` | no | Execution-profile ID; otherwise the prompt must set `default_profile`. |
+| `inputs` | yes | Non-empty object mapping input names to references. |
+| `vars` | no | Object mapping template-variable names to strings. |
+| `model` | no | Runtime model-override object. |
+| `include_raw_output` | no | Include `raw_model_output` when true. |
+
+An input reference has a required `type` of `file` or `inline`. A `file`
+reference requires `uri`; an `inline` reference requires `body`.
+
+HTTP file references require a configured artifact root. Relative paths resolve
+within that root. Absolute paths must be lexically within it; traversal outside
+it is rejected with `400 artifact_not_allowed`. This lexical check does not
+resolve symlinks: the operating system follows symlinks inside the root,
+including ones that target outside it. Keep the root narrow and inaccessible to
+untrusted writers.
+
+The optional `model` object accepts `endpoint`, `model`, `temperature`,
+`max_tokens`, `top_p`, `timeout_seconds`, `service_tier`,
+`reasoning_effort`, `api_key_env`, and `extra_params`. Numeric ranges and
+credential supply are defined by the [configuration reference](config.md).
+Explicit zero values for the numeric fields are overrides; zero
+`timeout_seconds` disables the per-generation deadline only, retaining the
+request context and configured transport cap. The timeout layers are defined in
+the [outbound integration contract](integrations/openai-compatible-chat.md#authentication-and-timeout).
+
+Raw API-key values are not accepted. `api_key` and any other unknown model
+field cause `400 invalid_json`.
+
+### Strict JSON
+
+Request decoding rejects malformed JSON, unknown fields at every request level,
+and trailing JSON tokens with `400 invalid_json`. A blank `prompt_id` or
+empty `inputs` object returns `400 invalid_request`.
+
+### Success Response
+
+A completed run returns `200 OK`, including when generated content fails its
+validation contract. The response contains:
+
+- `artifact`: `name`, `content_type`, `body`, `size`, `hash`, and
+  optional `uri`;
+- `validation`: `status`, `mode`, `repair_attempts`, `is_valid`, plus
+  optional `errors` and `schema_path`;
+- `metadata`: run, prompt, rendered-prompt, profile, model, input-hash, usage,
+  timing, validation, and repair-attempt metadata; and
+- optional `raw_model_output` when requested.
+
+`metadata.model_params` has `endpoint`, `model`, `temperature`,
+`max_tokens`, `top_p`, and `timeout_seconds`, plus optional
+`service_tier`, `reasoning_effort`, `api_key_env`, and `extra_params`.
+`metadata.usage` always includes `prompt_tokens`, `completion_tokens`,
+`total_tokens`, `cached_tokens`, and `cache_write_tokens`; unavailable
+cache usage is reported as zero.
+
+A validation failure has `validation.status: "failed"`, `is_valid: false`,
+and any available diagnostic errors, while still returning the artifact and
+metadata.
+
+## Error Responses
+
+Errors have this shape:
+
+```json
+{"error":{"code":"invalid_request","message":"prompt_id is required"}}
+```
+
+Messages are concise and do not expose wrapped internal causes.
+
+| Status | Code | Meaning |
+| --- | --- | --- |
+| `400` | `invalid_json` | Malformed JSON, unknown field, or trailing JSON. |
+| `400` | `invalid_request` | Missing or invalid request data or runtime override. |
+| `400` | `profile_required` | No profile ID and no prompt default profile. |
+| `400` | `prompt_load_failed` | Prompt definition failed to load. |
+| `400` | `profile_load_failed` | Profile failed to load. |
+| `400` | `artifact_not_allowed` | HTTP file input is disabled or outside the artifact root. |
+| `400` | `artifact_read_failed` | Input artifact is invalid or cannot be read. |
+| `400` | `prompt_render_failed` | Prompt template rendering failed. |
+| `400` | `api_key_env_missing` | The selected credential environment variable is unset or empty. |
+| `404` | `not_found` | Route does not exist. |
+| `404` | `prompt_not_found` | Prompt ID or version does not exist. |
+| `404` | `profile_not_found` | Profile ID does not exist. |
+| `405` | `method_not_allowed` | The route does not accept the method. |
+| `413` | `request_too_large` | Encoded request exceeds its limit. |
+| `413` | `artifact_too_large` | File input exceeds its limit. |
+| `413` | `response_too_large` | Encoded response exceeds its limit. |
+| `500` | `validation_runtime_failed` | Schema or validator runtime failure. |
+| `500` | `internal_error` | Unclassified server failure. |
+| `502` | `llm_failed` | Outbound model request failed. |
+
+## Retry And Idempotency
+
+Scriptorium provides no idempotency keys, pagination, caching headers, or rate
+limits. Clients may retry transport failures or `5xx` responses only when
+their workflow tolerates another model call: a retry can produce different
+output and incur another provider request.

docs/cli.md

@@ -0,0 +1,148 @@
+# CLI Reference
+
+This is the canonical contract for invoking Scriptorium. Configuration discovery,
+precedence, directories, profiles, and schemas are defined in the
+[configuration reference](config.md). The [HTTP API reference](api.md) owns
+service request and response behavior.
+
+## Shortest Useful 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
+```
+
+`render` prepares a request without calling an LLM.
+
+## Commands
+
+- `scriptorium run`: prepare a prompt, call the configured LLM, and write the
+  generated artifact.
+- `scriptorium render`: prepare a prompt and write prepared-run output.
+- `scriptorium serve`: start the HTTP server.
+
+All commands accept `--config <path>` and reject positional arguments. An
+effective `prompt_dir` is required for every command. Supply it through the
+configuration contract or the command's `--prompt-dir` flag.
+
+## `scriptorium run`
+
+```text
+scriptorium run [flags]
+```
+
+Required flags:
+
+| Flag | Meaning |
+| --- | --- |
+| `--prompt <id>` | Prompt ID to execute. |
+| `--input name=path` | Input file mapping; repeat or use comma-separated mappings. |
+
+Optional flags:
+
+| Flag | Meaning |
+| --- | --- |
+| `--config <path>` | Application configuration file. |
+| `--prompt-dir <dir>` | Prompt-definition directory override. |
+| `--profile-dir <dir>` | Custom profile-directory override. |
+| `--schema-dir <dir>` | Schema base-directory override. |
+| `--profile <id>` | Execution-profile override. |
+| `--var name=value` | Template-variable mapping; repeat or use comma-separated mappings. |
+| `--out <path>` | Write generated content to this file instead of stdout. |
+| `--llm-base-url <url>` | Runtime endpoint override. |
+| `--model <name>` | Runtime model override. |
+| `--api-key-env <name>` | Runtime API-key environment-variable name override. |
+| `--temperature <float>` | Runtime temperature override. |
+| `--max-tokens <int>` | Runtime maximum-token override. |
+| `--top-p <float>` | Runtime top-p override. |
+| `--timeout <duration>` | Runtime timeout override using Go duration syntax. |
+
+Deprecated aliases: `--prompt-id` for `--prompt`, and `--profile-id` for
+`--profile`.
+
+Omitted numeric runtime flags preserve the selected effective value; explicit
+zero values override it. `--timeout 0s` disables the per-generation deadline
+only; the caller context and configured transport cap remain active. CLI
+durations are converted to whole seconds by truncation toward zero, so any
+duration whose absolute value is below one second becomes an explicit
+zero-second override. The timeout layers are defined in the
+[outbound integration contract](integrations/openai-compatible-chat.md#authentication-and-timeout).
+
+There is no raw API-key flag. Use `--api-key-env`.
+
+## `scriptorium render`
+
+```text
+scriptorium render [flags]
+```
+
+`--prompt <id>` and at least one `--input name=path` are required. The
+following optional flags are supported: `--config`, `--prompt-dir`,
+`--profile-dir`, `--profile`, `--var`, `--out`, `--llm-base-url`,
+`--model`, `--api-key-env`, `--temperature`, `--max-tokens`, `--top-p`,
+`--timeout`, and `--format text|json`. Their meanings match the corresponding
+`run` flags; `--format` selects prepared-run output and otherwise uses
+`defaults.render_format`.
+
+The same deprecated aliases and numeric/timeout behavior as `run` apply.
+`render` does not accept `--schema-dir`; configure `schema_dir` through the
+configuration file. It resolves profiles and schemas as part of preparation but
+does not call an LLM.
+
+## `scriptorium serve`
+
+```text
+scriptorium serve [flags]
+```
+
+Optional flags:
+
+| Flag | Meaning |
+| --- | --- |
+| `--config <path>` | Application configuration file. |
+| `--addr <listen-address>` | HTTP listen-address override. |
+| `--prompt-dir <dir>` | Prompt-definition directory override. |
+| `--profile-dir <dir>` | Custom profile-directory override. |
+| `--schema-dir <dir>` | Schema base-directory override. |
+| `--artifact-root <dir>` | Root for HTTP `file` input references. |
+| `--max-request-bytes <n>` | Maximum encoded HTTP request-body bytes; `0` disables the limit. |
+| `--max-artifact-bytes <n>` | Maximum HTTP file-input artifact bytes; `0` disables the limit. |
+| `--max-response-bytes <n>` | Maximum encoded HTTP response bytes; `0` disables the limit. |
+
+`serve` accepts no runtime model override flags. HTTP request fields, response
+schemas, and error codes are defined in the [HTTP API reference](api.md).
+
+## Input And Variable Syntax
+
+`--input name=path` maps an input name to a local file; `--var name=value`
+maps a template variable to a string. Both flags can be repeated or contain
+comma-separated mappings. Values may contain `=` after the first separator.
+Empty names and values are rejected.
+
+CLI inputs are file references. HTTP inline inputs are defined by the
+[HTTP API reference](api.md).
+
+## Output And Exit Behavior
+
+- `run` writes generated content to stdout, or to `--out` when supplied, and
+  writes a concise summary to stderr.
+- `render` writes prepared-run output to stdout, or to `--out` when supplied,
+  without a success summary.
+- `serve` writes startup and server errors to stderr.
+
+Exit statuses:
+
+| Status | Meaning |
+| --- | --- |
+| `0` | Success. |
+| `1` | Parse, configuration, loading, rendering, generation, output-write, or other runtime error. |
+| `2` | `run` generated and wrote output, but validation failed. |
+
+## Workflows And Examples
+
+The [maintained render script](../examples/render-markdown-summary.sh) is a
+copyable render workflow. The [HTTP request example](../examples/http-run.json)
+is for a running `serve` process.

docs/config.md

@@ -0,0 +1,173 @@
+# Configuration Reference
+
+This is the canonical reference for Scriptorium application settings and the
+prompt, profile, and schema files those settings select. For command syntax,
+see the [CLI reference](cli.md); for HTTP request shapes, limits, and outcomes,
+see the [HTTP API reference](api.md).
+
+## Discovery And Precedence
+
+Application settings are resolved in this order:
+
+1. built-in defaults;
+2. a configuration file; then
+3. CLI overrides.
+
+When `--config` is omitted, Scriptorium searches
+`/usr/local/etc/scriptorium/config.yml` and then `/etc/scriptorium/config.yml`.
+If neither exists, it uses built-in defaults. An explicit `--config` path must
+exist and decode successfully.
+
+The maintained [minimal configuration](../examples/config.yml) and
+[full configuration](../examples/config.full.yml) are copyable examples.
+
+## Application Configuration File
+
+Configuration is strict YAML: unknown fields are rejected. Empty string values
+do not override a prior value. Raw API-key fields are not accepted.
+
+| Field | Default | Meaning |
+| --- | --- | --- |
+| `prompt_dir` | unset | Directory containing prompt-definition YAML. `run`, `render`, and `serve` require an effective value. |
+| `profile_dir` | unset | Directory containing custom profile YAML. Built-in profiles remain available. |
+| `schema_dir` | `.` | Base directory for relative JSON Schema paths. |
+| `server.addr` | `:8080` | Address used by `serve`. |
+| `server.artifact_root` | unset | Root that enables HTTP `file` input references. |
+| `server.max_request_bytes` | `16777216` | Maximum encoded HTTP request body bytes; `0` disables the limit. |
+| `server.max_artifact_bytes` | `16777216` | Maximum HTTP file-input artifact bytes; `0` disables the limit. |
+| `server.max_response_bytes` | `16777216` | Maximum encoded HTTP response bytes; `0` disables the limit. |
+| `defaults.render_format` | `text` | Default `render` output format: `text` or `json`. |
+
+The three size fields must be zero or greater. The HTTP contract defines how
+each limit is enforced and reported. `server.artifact_root` configures the
+deployment boundary; see the [HTTP API reference](api.md) for request-path and
+containment behavior, and [operations](operations.md) for deployment handling.
+
+## Prompt Definition Files
+
+Prompt definitions are strict YAML files anywhere below `prompt_dir`. A prompt
+is selected by its YAML `id`, not by file path; nested directories are only for
+organization. See [maintained prompt examples](../examples/prompts/).
+
+| Field | Required | Meaning |
+| --- | --- | --- |
+| `id` | yes | Prompt identifier. |
+| `version` | yes | Prompt version. |
+| `default_profile` | no | Profile used when a request omits a profile ID. |
+| `description` | no | Human-readable description. |
+| `session_id` | no | Go-template string rendered from request variables and sent to a compatible provider when non-empty. |
+| `inputs` | no | Declared input metadata. |
+| `messages` | yes | Chat-message templates. |
+| `output` | yes | Output format and validation contract. |
+
+### Inputs And Messages
+
+Each `inputs` item has a required `name` and optional `required`,
+`content_type`, and `description` fields. Input names must be unique.
+
+Each message has a required `role`, exactly one of `content` or `content_file`,
+and optional `cache_control`. A `content_file` path is relative to the prompt
+file. `cache_control.type` must be `ephemeral`; its optional `ttl` is `1h`.
+
+`session_id` uses the same template variables as messages. Empty rendered
+values are omitted. A rendered value may contain at most 256 Unicode code
+points.
+
+### Output Contract
+
+| Field | Required | Values or behavior |
+| --- | --- | --- |
+| `format` | yes | `text`, `markdown`, or `json`. |
+| `validation_mode` | yes | `none`, `basic`, `json`, or `json_schema`. |
+| `schema_path` | for `json_schema` | Schema path, relative to `schema_dir` unless absolute. |
+| `repair_attempts` | no | Integer greater than or equal to `0`; omitted means `0`. |
+
+## Profile Definition Files
+
+Profiles are strict YAML files anywhere below `profile_dir`. A profile is
+selected by YAML `id`; nested directories are organizational. See the
+[maintained profile examples](../examples/profiles/).
+
+| Field | Required | Meaning |
+| --- | --- | --- |
+| `id` | yes | Profile identifier. |
+| `endpoint` | yes | OpenAI-compatible base URL, including its API version path when needed. |
+| `model` | yes | Provider model name. |
+| `temperature` | no | Number from `0` through `2`. |
+| `max_tokens` | no | Integer zero or greater. |
+| `top_p` | no | Number from `0` through `1`. |
+| `timeout_seconds` | no | Per-generation-call deadline in whole seconds; integer zero or greater. |
+| `service_tier` | no | Non-empty provider-specific request tier. |
+| `reasoning_effort` | no | Non-empty provider-specific reasoning setting. |
+| `api_key_env` | no | Environment-variable name containing the API key. |
+| `extra_params` | no | JSON-compatible provider-specific outbound request fields. |
+
+Execution defaults before profile and request overrides are `temperature: 0`,
+`max_tokens: 0`, `top_p: 1`, and `timeout_seconds: 600`. Profile numeric values
+merge by non-zero value. Request overrides preserve presence, so an explicit
+zero can override a profile value. For `timeout_seconds`, explicit request zero
+disables the generation deadline while retaining the caller context and the
+built-in client's transport cap. See the
+[OpenAI-compatible integration contract](integrations/openai-compatible-chat.md#authentication-and-timeout)
+for the complete timeout interaction.
+
+Custom profiles take precedence over built-ins with the same ID. Invalid custom
+profiles are errors; they do not fall back to a built-in profile. Raw `api_key`
+is rejected. Use `api_key_env`, or the public Go package's request-scoped key
+mechanism described in the [package contract](consumers/pkg-scriptorium.md).
+
+`extra_params` keys must be non-empty and cannot be `model`, `session_id`,
+`messages`, `temperature`, `max_tokens`, `top_p`, `service_tier`,
+`reasoning_effort`, or `response_format`.
+
+### Built-In Profile Catalog
+
+Each embedded profile uses `OPENROUTER_API_KEY`.
+
+| Provider | ID | Model |
+| --- | --- | --- |
+| aion-labs | `aion-2` | `aion-labs/aion-2.0` |
+| anthropic | `claude-fable-latest` | `~anthropic/claude-fable-latest` |
+| anthropic | `claude-haiku-latest` | `~anthropic/claude-haiku-latest` |
+| anthropic | `claude-opus-latest` | `~anthropic/claude-opus-latest` |
+| anthropic | `claude-sonnet-latest` | `~anthropic/claude-sonnet-latest` |
+| deepseek | `deepseek-3-2` | `deepseek/deepseek-v3.2` |
+| deepseek | `deepseek-4-flash` | `deepseek/deepseek-v4-flash` |
+| deepseek | `deepseek-4-pro` | `deepseek/deepseek-v4-pro` |
+| google | `gemini-2-flash` | `google/gemini-2.5-flash` |
+| google | `gemini-2-flash-lite` | `google/gemini-2.5-flash-lite` |
+| google | `gemini-2-pro` | `google/gemini-2.5-pro` |
+| google | `gemini-3-flash-lite` | `google/gemini-3.1-flash-lite` |
+| google | `gemini-flash-latest` | `~google/gemini-flash-latest` |
+| google | `gemini-pro-latest` | `~google/gemini-pro-latest` |
+| google | `gemma-4-31b` | `google/gemma-4-31b-it:exacto` |
+| minimax | `minimax-m2` | `minimax/minimax-m2.5` |
+| minimax | `minimax-m3` | `minimax/minimax-m3` |
+| mistral | `mistral-large-2512` | `mistralai/mistral-large-2512` |
+| mistral | `mistral-medium-3-5` | `mistralai/mistral-medium-3-5` |
+| mistral | `mistral-small-3` | `mistralai/mistral-small-3.2-24b-instruct` |
+| mistral | `mistral-small-4` | `mistralai/mistral-small-2603` |
+| nvidia | `nemotron-3-ultra` | `nvidia/nemotron-3-ultra-550b-a55b` |
+| openai | `gpt-5-mini` | `openai/gpt-5.4-mini` |
+| openai | `gpt-5-nano` | `openai/gpt-5.4-nano` |
+
+## Schemas
+
+Schemas are JSON files, normally below `schema_dir`. `json_schema` output
+requires a `schema_path`. Relative paths resolve from `schema_dir`; absolute
+paths are used directly. Referenced nested schemas use relative paths and are
+not discovered by basename. An unreadable or invalid schema is a runtime
+validation error; generated content that fails JSON or schema validation is a
+validation result.
+
+## Credentials
+
+Keep secrets in environment variables. Store only an environment-variable name
+in `api_key_env`; do not place raw keys in configuration, prompt or profile
+files, CLI arguments, examples, or HTTP payloads.
+
+## Related References
+
+- [CLI reference](cli.md)
+- [HTTP API reference](api.md)
+- [OpenAI-compatible outbound contract](integrations/openai-compatible-chat.md)

docs/config/config-yml.md

@@ -1,48 +0,0 @@
-# Main `config.yml`
-
-`config.yml` defines application-level defaults used by CLI commands.
-
-By default, Scriptorium looks for `/usr/local/etc/scriptorium/config.yml` and, if not present, then for `/etc/scriptorium/config.yml`. You can also pass `--config PATH`.
-
-## Complete Example
-
-```yaml
-prompt_dir: ./prompts
-profile_dir: ./profiles
-schema_dir: ./schemas
-
-server:
-  addr: :8080
-
-defaults:
-  render_format: text
-```
-
-## Available Options
-
-- `prompt_dir` (optional): Default directory for prompt definition YAML files.
-- `profile_dir` (optional): Default directory for execution profile YAML files.
-- `schema_dir` (optional): Base directory for JSON schema files used by `json_schema` validation.
-
-### `server`
-
-- `addr` (optional): HTTP server listen address for `scriptorium serve`.
-
-### `defaults`
-
-- `render_format` (optional): Default output format for `scriptorium render`.
-- Allowed values: `text`, `json`.
-
-## Precedence
-
-For run/render/serve settings, precedence is:
-
-1. Explicit CLI flags
-2. `config.yml`
-3. Built-in defaults
-
-## Notes and Rules
-
-- Unknown YAML fields fail to load (strict decoding).
-- This file does not accept API keys.
-- `config.yml` sets directory/server defaults only; prompt/profile content remains in their own files.

docs/config/profile-definitions.md

@@ -1,41 +0,0 @@
-# Execution Profile Definitions
-
-Execution Profiles define **how** Scriptorium calls an LLM endpoint.
-
-A profile file is YAML, typically stored under `profiles/`, for example `profiles/local-quality.yaml`.
-
-## Complete Example
-
-```yaml
-id: local-quality
-endpoint: http://localhost:8000/v1
-model: gpt-4.1
-temperature: 0.0
-max_tokens: 1200
-top_p: 1.0
-timeout_seconds: 180
-reasoning_effort: medium
-api_key_env: SCRIPTORIUM_API_KEY
-extra_params:
-  provider: openrouter
-  route: fallback
-```
-
-## Available Options
-
-- `id` (required): Unique profile identifier used by `--profile` or prompt `default_profile`.
-- `endpoint` (required): OpenAI-compatible base URL, usually ending in `/v1`.
-- `model` (required): Model name to request at that endpoint.
-- `temperature` (optional): Sampling temperature. Valid range is `0` to `2`.
-- `max_tokens` (optional): Max completion tokens. Must be `>= 0`.
-- `top_p` (optional): Nucleus sampling parameter. Valid range is `0` to `1`.
-- `timeout_seconds` (optional): Request timeout in seconds. Must be `>= 0`.
-- `reasoning_effort` (optional): Provider/model-specific reasoning level string.
-- `api_key_env` (optional): Environment variable name that holds the API key.
-- `extra_params` (optional): String key/value map for provider-specific parameters.
-
-## Notes and Rules
-
-- Raw API keys are not supported. Do **not** add `api_key` fields.
-- Unknown YAML fields fail to load (strict decoding).
-- If `api_key_env` is set, the environment variable must be present when the run executes.

docs/config/prompt-definitions.md

@@ -1,73 +0,0 @@
-# Prompt Definition Files
-
-Prompt Definitions define **what** Scriptorium should do.
-
-A prompt file is YAML, typically stored under `prompts/`, for example `prompts/generic.structured_events.yaml`.
-
-## Complete Example
-
-```yaml
-id: generic.structured_events
-version: "1.0.0"
-default_profile: local-quality
-description: Extract events from a transcript into structured JSON.
-
-inputs:
-  - name: transcript
-    required: true
-    content_type: text/markdown
-    description: Source transcript
-  - name: glossary
-    required: false
-    content_type: text/yaml
-    description: Optional glossary context
-
-messages:
-  - role: system
-    content: |
-      You are a structured extraction assistant.
-      Return only JSON.
-  - role: user
-    content_file: ./generic.structured_events.user.md
-
-output:
-  format: json
-  validation_mode: json_schema
-  schema_path: structured_events.schema.json
-  repair_attempts: 1
-```
-
-## Available Options
-
-- `id` (required): Prompt identifier used by `--prompt` / `prompt_id`.
-- `version` (required): Prompt version string.
-- `default_profile` (optional): Execution profile ID used when no explicit profile is provided.
-- `description` (optional): Human-readable description.
-
-### `inputs[]`
-
-- `name` (required): Logical input name referenced in templates via `{{input "name"}}`.
-- `required` (optional): If `true`, run fails when input is missing.
-- `content_type` (optional): Metadata only (not enforced yet).
-- `description` (optional): Human-readable input description.
-
-### `messages[]`
-
-- `role` (required): Message role such as `system` or `user`.
-- `content` (optional): Inline Go-template message body.
-- `content_file` (optional): Path to a template file.
-
-Each message must set **exactly one** of `content` or `content_file`.
-
-### `output`
-
-- `format` (required): One of `text`, `markdown`, `json`.
-- `validation_mode` (required): One of `none`, `basic`, `json`, `json_schema`.
-- `schema_path` (required when `validation_mode: json_schema`): Path to JSON Schema file.
-- `repair_attempts` (required): Number of bounded repair retries (`>= 0`).
-
-## Notes and Rules
-
-- Unknown YAML fields fail to load (strict decoding).
-- `content_file` paths are resolved relative to the prompt YAML file.
-- For `json_schema` validation mode, Scriptorium also sends provider-level structured output requests automatically.

docs/config/schema-definitions.md

@@ -1,82 +0,0 @@
-# JSON Schema Definition Files
-
-Schema definition files describe the expected JSON output contract for prompts that use:
-
-- `output.format: json`
-- `output.validation_mode: json_schema`
-
-Schema files are JSON, typically stored under `schemas/`, for example `schemas/structured_events.schema.json`.
-
-## Complete Example
-
-```json
-{
-  "$schema": "https://json-schema.org/draft/2020-12/schema",
-  "$id": "https://example.com/schemas/structured-events.schema.json",
-  "title": "Structured Events",
-  "description": "Expected shape for extracted event output",
-  "type": "object",
-  "properties": {
-    "summary": {
-      "type": "string",
-      "minLength": 1,
-      "description": "High-level session summary"
-    },
-    "events": {
-      "type": "array",
-      "items": {
-        "type": "object",
-        "properties": {
-          "title": { "type": "string" },
-          "type": {
-            "type": "string",
-            "enum": ["discovery", "combat", "social", "travel", "downtime", "other"]
-          },
-          "notes": { "type": "string" }
-        },
-        "required": ["title", "type"],
-        "additionalProperties": false
-      },
-      "minItems": 0
-    }
-  },
-  "required": ["summary", "events"],
-  "additionalProperties": false,
-  "$defs": {
-    "nonEmptyString": {
-      "type": "string",
-      "minLength": 1
-    }
-  }
-}
-```
-
-## Available Options
-
-Scriptorium does not define custom schema keywords. It expects a valid JSON Schema document and passes it to the validator/provider.
-
-Commonly used JSON Schema options include:
-
-- `$schema`: Draft identifier URI.
-- `$id`: Schema identifier URI.
-- `title`: Human-readable schema title.
-- `description`: Human-readable schema description.
-- `type`: Expected JSON type (`object`, `array`, `string`, etc.).
-- `properties`: Object field definitions.
-- `required`: Required object fields.
-- `additionalProperties`: Whether undeclared fields are allowed.
-- `items`: Array item schema.
-- `enum`: Allowed literal values.
-- `const`: Single allowed literal value.
-- `oneOf`, `anyOf`, `allOf`: Composition rules.
-- `minimum`, `maximum`: Numeric bounds.
-- `minLength`, `maxLength`, `pattern`: String constraints.
-- `minItems`, `maxItems`: Array constraints.
-- `$defs`: Reusable local definitions.
-- `$ref`: Reference to another schema/definition.
-
-## Notes and Rules
-
-- Schema path comes from prompt `output.schema_path` and is resolved relative to `schema_dir`.
-- If schema loading fails for `json_schema` mode, the run fails before the LLM request.
-- Keep schemas strict (`additionalProperties: false`) when you want predictable output shape.

docs/consumers/api.md

@@ -0,0 +1,59 @@
+# Consumer Integration Overview
+
+This guide helps applications choose a Scriptorium interface and understand
+their responsibilities. The linked contracts own interface syntax and wire
+semantics.
+
+| Interface | Use when |
+| --- | --- |
+| Go package | The consumer is Go and needs typed requests, results, or an injected LLM client. |
+| CLI subprocess | The consumer needs process isolation or is not written in Go. |
+| HTTP API | The consumer needs a service boundary or remote access. |
+
+- Go package: [package contract](pkg-scriptorium.md)
+- CLI subprocess: [subprocess integration](../integrations/subprocess.md)
+- HTTP service: [HTTP API reference](../api.md)
+- Prompt, profile, schema, and credential configuration: [configuration reference](../config.md)
+
+## Minimal Go Use
+
+```go
+engine, err := scriptorium.NewEngine(scriptorium.Config{
+	PromptDir:  "./examples/prompts",
+	ProfileDir: "./examples/profiles",
+})
+if err != nil {
+	return err
+}
+
+prepared, err := engine.Prepare(ctx, scriptorium.RunRequest{
+	PromptID: "generic.markdown_summary",
+	Inputs: map[string]scriptorium.ArtifactRef{
+		"transcript": scriptorium.File("./examples/fixtures/transcript.md"),
+	},
+})
+if err != nil {
+	return err
+}
+_ = prepared
+```
+
+For a maintained program, see
+[`examples/go-library/prepare`](../../examples/go-library/prepare).
+
+## Consumer Responsibilities
+
+Consumers are responsible for:
+
+- selecting and deploying prompt, profile, and schema assets;
+- supplying required inputs and template variables;
+- supplying credentials through the applicable interface;
+- protecting rendered prompts and generated artifacts as potentially sensitive;
+- deciding whether validation-failed output is usable; and
+- retrying only when another model call is acceptable.
+
+Scriptorium does not persist run state. A retry can produce different output and
+can incur another provider request. CLI exit behavior belongs to the
+[CLI reference](../cli.md); HTTP status behavior belongs to the
+[HTTP API reference](../api.md); package errors and results belong to the
+[package contract](pkg-scriptorium.md).

docs/consumers/pkg-scriptorium.md

@@ -0,0 +1,183 @@
+# Package `scriptorium`
+
+Import path:
+
+```go
+import "gitea.maximumdirect.net/eric/scriptorium"
+```
+
+This is the canonical public Go contract for in-process prompt preparation and
+execution. Prompt, profile, and schema file formats are defined in the
+[configuration reference](../config.md).
+
+## Engine Construction
+
+`NewEngine(Config, ...Option)` constructs an engine. `Config` has these
+fields:
+
+| Field | Meaning |
+| --- | --- |
+| `PromptDir` | Prompt-definition directory, required unless a prompt source option is supplied. |
+| `ProfileDir` | Optional custom profile directory over built-ins. |
+| `SchemaDir` | Schema directory; empty uses `.`. |
+| `Timeout` | Transport-wide safety cap for the built-in OpenAI-compatible client when `HTTPClient` is absent or has a non-positive timeout. A non-positive value uses the internal ten-minute default. |
+| `HTTPClient` | Optional HTTP client for that built-in client. It is cloned; a positive `Timeout` on it is the transport cap and takes precedence over `Config.Timeout`. A non-positive client timeout is treated as unset. |
+
+Nil options are ignored. Invalid construction, including
+`WithLLMClient(nil)`, returns an error matching `ErrInvalidConfig`.
+
+Profile and request `timeout_seconds` values select a per-generation-call
+deadline independently of the transport cap. An explicit request override of
+zero disables that generation deadline only. The complete interaction with the
+caller context is defined in the
+[OpenAI-compatible integration contract](../integrations/openai-compatible-chat.md#authentication-and-timeout).
+
+Source options replace their matching directory source:
+
+- prompts: `WithPromptFS(fsys, root)`, `WithPromptFile(path)`;
+- profiles: `WithProfileFS(fsys, root)`, `WithProfileFile(path)`, and
+  `WithProfiles(profiles...)`;
+- schemas: `WithSchemaFS(fsys, root)`, `WithSchemaFile(path)`; and
+- LLM client: `WithLLMClient(client)`.
+
+`fs.FS` prompt-content and schema paths stay inside their configured roots.
+Single-file prompt and profile sources are selected by their YAML `id`, not
+their file names. `WithPromptFile` resolves relative `content_file` paths from
+the prompt file's directory. `WithSchemaFile` exposes its schema by the schema
+file's base name. In-memory profiles take precedence over an explicit or
+directory-backed profile source, which in turn takes precedence over built-ins.
+File and filesystem sources use the format and credential rules in the
+[configuration reference](../config.md).
+
+## Prepare And Run
+
+`Prepare(ctx, request)` resolves the prompt, profile, input artifacts,
+validation contract, and rendered messages without calling an LLM.
+`Run(ctx, request)` performs that preparation, calls the configured client,
+and validates generated content.
+
+```go
+engine, err := scriptorium.NewEngine(scriptorium.Config{
+	PromptDir:  "./examples/prompts",
+	ProfileDir: "./examples/profiles",
+})
+if err != nil {
+	return err
+}
+
+prepared, err := engine.Prepare(ctx, scriptorium.RunRequest{
+	PromptID: "generic.markdown_summary",
+	Inputs: map[string]scriptorium.ArtifactRef{
+		"transcript": scriptorium.File("./examples/fixtures/transcript.md"),
+	},
+})
+if err != nil {
+	return err
+}
+_ = prepared.Messages
+```
+
+The maintained package example is
+[`examples/go-library/prepare`](../../examples/go-library/prepare).
+
+`PreparedRun` exposes prompt, selected-profile, effective-model, output
+contract, structured-output, input-hash, rendered-message, and timing
+information. It does not include a resolved API key, model output, validation
+result, or target-presence metadata.
+
+`RunResult` adds run ID, artifact, raw output, validation, model metadata,
+usage, and duration. Generated-content validation failures return a result with
+`Validation.Status == ValidationFailed`; schema or validator runtime failures
+return an error matching `ErrValidation`.
+
+## Public Values
+
+`ArtifactRef` has `Type`, `URI`, and `Body`; `Artifact` has `Name`,
+`ContentType`, `Body`, `URI`, `Size`, and `Hash`. `ExecutionTarget` exposes the
+effective endpoint, model, numeric settings, credential-environment name,
+service tier, reasoning effort, and extra parameters. `ValidationResult`
+contains status, mode, errors, schema path, repair attempts, and validity.
+
+The exported constants define these serialized values:
+
+- artifact types: `inline` and `file`;
+- output formats: `text`, `markdown`, and `json`;
+- validation modes: `none`, `basic`, `json`, and `json_schema`; and
+- validation statuses: `passed`, `failed`, and `skipped`.
+
+`TokenUsage` reports prompt, completion, total, cached, and cache-write token
+counts. `RenderedPrompt`, `RenderedMessage`, `CacheControl`, and
+`StructuredOutputSpec` are the public shapes used by injected LLM clients.
+
+## Requests, Inputs, And Overrides
+
+`RunRequest` fields are `PromptID`, `PromptVersion`, `ProfileID`,
+`APIKey`, `Inputs`, `Vars`, `Execution`, `Validation`, and
+`Metadata`.
+
+Input helpers are:
+
+- `File(path)` for a file-backed artifact;
+- `Inline(body)` for inline content; and
+- `InlineWithURI(uri, body)` for inline content with URI metadata.
+
+Required declared inputs must be supplied. Template rendering must also resolve
+every input name the prompt actually references. Extra entries in `Inputs`
+are not rejected solely because they are undeclared.
+
+`ExecutionTargetOverride` supplies endpoint, model, credential-environment,
+service-tier, reasoning-effort, and extra-parameter overrides. Its numeric
+fields (`Temperature`, `MaxTokens`, `TopP`, and `TimeoutSeconds`) are
+pointers so explicit zero values are preserved. `OutputContract` supplies
+`Format`, `ValidationMode`, `SchemaPath`, and `RepairAttempts`.
+
+`ExtraParams` accepts JSON-compatible values: strings, booleans, finite
+numbers, objects with string keys, arrays or slices, and nil. Unsupported
+values, non-string map keys, non-finite floats, and cycles return
+`ErrInvalidConfig` for profiles or `ErrInvalidRequest` for request
+overrides.
+
+## Profiles And Credentials
+
+`OpenAICompatibleProfile(OpenAICompatibleProfileConfig)` creates an
+in-memory `Profile`. Its public fields are `ID`, `Endpoint`, `Model`,
+`Temperature`, `MaxTokens`, `TopP`, `TimeoutSeconds`, `ServiceTier`,
+`ReasoningEffort`, `APIKeyRequired`, and `ExtraParams`.
+`WithProfiles` rejects duplicate IDs in one call.
+
+A direct `RunRequest.APIKey` is request-scoped and takes precedence over
+`api_key_env` for the built-in client. It is excluded from JSON output and
+from `PreparedRun` and `RunResult`. The package's `String` and
+`GoString` methods report only whether a direct key is set. Do not use
+reflection-based dumps of request structs, which can bypass that redaction.
+
+## Injected LLM Clients
+
+`LLMClient` implements:
+
+```go
+Generate(context.Context, GenerateRequest) (*GenerateResponse, error)
+```
+
+Injected clients receive the rendered prompt, effective execution target, numeric
+target-presence metadata, optional structured-output specification, and direct
+request API key. `GenerateResponse` returns content and `TokenUsage`.
+Custom clients should avoid logging raw prompts or credentials.
+
+## Errors
+
+Public methods preserve these sentinel checks through `errors.Is`:
+
+- `ErrInvalidConfig`
+- `ErrInvalidRequest`
+- `ErrPromptNotFound`
+- `ErrProfileNotFound`
+- `ErrPromptLoad`
+- `ErrProfileLoad`
+- `ErrArtifactLoad`
+- `ErrPromptRender`
+- `ErrLLMGenerate`
+- `ErrValidation`
+
+For interface selection and operational responsibilities, see the
+[consumer integration overview](api.md).

docs/development.md

@@ -0,0 +1,63 @@
+# Development
+
+This is the contributor entry point for Scriptorium. Use the task-specific
+reading guide below before making changes. Canonical architecture, contracts,
+component behavior, and policies remain in their owning documents.
+
+## Initial Orientation
+
+Before starting work:
+
+1. inspect the working tree and preserve unrelated changes;
+2. read the architecture policy for code or design work;
+3. read the policy, contract, and internal documents listed for the task;
+4. inspect the relevant implementation and tests before deciding how to change
+   them.
+
+Start with:
+
+- [Architecture policy](policy/architecture.md) for system boundaries,
+  invariants, and non-goals;
+- [Internal component overview](internal/overview.md) for the current package
+  and component map;
+- [Documentation policy](policy/documentation.md) before changing
+  documentation;
+- [Testing policy](policy/testing.md) before adding, rewriting, or deleting
+  tests.
+
+## Task-Specific Reading Guide
+
+| Task | Read before changing |
+| --- | --- |
+| Repository orientation or component responsibility | [Internal component overview](internal/overview.md) and [architecture policy](policy/architecture.md) |
+| Public Go package or engine behavior | [Go package consumer contract](consumers/pkg-scriptorium.md), [internal component overview](internal/overview.md), [runner internals](internal/runner.md), [adapter internals](internal/adapters.md), and [source internals](internal/sources.md) |
+| CLI commands, flags, output, or exit behavior | [CLI contract](cli.md), [internal component overview](internal/overview.md), and [adapter internals](internal/adapters.md) |
+| HTTP routes, DTOs, limits, or status mapping | [HTTP API contract](api.md), [internal component overview](internal/overview.md), [adapter internals](internal/adapters.md), and [source internals](internal/sources.md) |
+| Application configuration | [Configuration contract](config.md), [internal component overview](internal/overview.md), [adapter internals](internal/adapters.md), and [source internals](internal/sources.md) |
+| Prompt, profile, schema, or artifact loading | [Configuration contract](config.md), [internal component overview](internal/overview.md), and [source internals](internal/sources.md) |
+| Runner orchestration, rendering, validation, or repair | [Runner internals](internal/runner.md) and [source internals](internal/sources.md) |
+| OpenAI-compatible request or response behavior | [OpenAI-compatible integration](integrations/openai-compatible-chat.md), [LLM internals](internal/llm.md), [runner internals](internal/runner.md), and [adapter internals](internal/adapters.md) |
+| Subprocess behavior | [Subprocess integration](integrations/subprocess.md) and [CLI contract](cli.md) |
+| Runtime operation or recovery | [Operations](operations.md) |
+| Examples or copyable assets | The owning contract for the demonstrated behavior and the related files under `examples/` |
+| Architecture decisions or future work | The [documentation policy](policy/documentation.md), relevant accepted ADRs such as [ADR 0001](adr/0001-adopt-canonical-documentation-ownership.md), and relevant roadmap documents under `roadmap/` |
+
+For cross-cutting changes, follow every applicable row. Internal component
+documents own detailed subsystem change recipes.
+
+## Baseline Validation
+
+Use focused checks while iterating, then run validation proportionate to the
+change and the risks described by the testing policy.
+
+The repository-level baseline for code changes is:
+
+```bash
+go test ./...
+go vet ./...
+go build ./cmd/scriptorium
+```
+
+Documentation-only work does not require the full Go suite unless it changes
+commands, examples, generated output, or another behavior that the suite
+validates. Always check changed links, paths, examples, and canonical ownership.

docs/integrations/narratio.md

@@ -1,322 +0,0 @@
-# Narratio -> Scriptorium CLI Integration
-
-## 1. Purpose
-
-This document defines how Narratio should invoke Scriptorium through the **public CLI**.
-
-This is a **subprocess integration contract**, not an internal Go API contract.
-
-## 2. Assumptions
-
-- `scriptorium` is installed and available on `PATH`.
-- Scriptorium is configured with `config.yml`.
-- `config.yml` provides `prompt_dir`, `profile_dir`, and `schema_dir` as needed.
-- Prompt and profile libraries are already deployed for the environment.
-- Narratio provides prepared artifact files (for example polished transcript, glossary, previous recap, campaign notes).
-- Initial integration is synchronous subprocess execution.
-- Narratio remains the orchestrator.
-
-In normal operation, Narratio does not need to pass `--prompt-dir` and `--profile-dir` if they are supplied by Scriptorium config.
-
-Narratio may pass `--config <PATH>` when it must use a non-default Scriptorium config file.
-
-## 3. Core Commands Narratio May Call
-
-Primary commands for subprocess integration:
-
-- `scriptorium run`
-- `scriptorium render`
-
-For production generation, use `scriptorium run`.
-
-`scriptorium render` is for debugging, dry-runs, test assertions, and validating command construction without LLM execution.
-
-Note: `scriptorium serve` and HTTP API exist, but they are not the initial integration path.
-
-## 4. Command Selection Guidance
-
-- Use `run` to generate an output artifact.
-- Use `render` to inspect the prepared prompt and effective settings without calling the LLM.
-- Use `render --format json` when Narratio/tests need structured prepare output.
-
-## 5. Recommended `run` Invocation Shape
-
-Production shape:
-
-```bash
-scriptorium run \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --out <output-artifact-path>
-```
-
-Common optional additions:
-
-- `--config <path>`: use a specific Scriptorium config file.
-- `--profile <profile_id>`: override prompt default profile.
-- `--var name=value` (repeatable): small metadata values.
-- `--input name=path` (repeatable): additional named artifacts.
-- `--timeout <duration>`: per-run timeout override.
-- Runtime model override flags (`--llm-base-url`, `--model`, etc.) only for exceptional/operator-directed cases.
-
-## 6. Recommended `render` Invocation Shape
-
-Human-readable debug shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format text
-```
-
-Structured debug/test shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format json \
-  --out <render-debug-path>
-```
-
-`render` does **not** call the LLM, does **not** validate model output, and does **not** perform repair.
-
-## 7. Inputs
-
-- Pass inputs as repeated `--input name=path` flags.
-- `name` must match the Prompt Definition input name.
-- Prefer absolute paths, or paths relative to a working directory controlled by Narratio.
-- Pass Audita output as the primary transcript input.
-- Additional inputs may include glossary, previous recap, campaign notes, event logs, final state maps, or other prompt-specific artifacts.
-- Scriptorium reads input files directly; Narratio does not need to inline file content for CLI use.
-
-## 8. Variables
-
-Use repeated `--var name=value` for small metadata values.
-
-Typical examples:
-
-- `session_date`
-- `session_id`
-- `campaign_name`
-- `previous_session_id`
-- `output_kind`
-
-Large content belongs in input files, not `--var` values.
-
-## 9. Prompt IDs and Output Artifact Types
-
-Narratio should treat prompt IDs as configuration, not hardcoded business logic.
-
-Narratio config may map stage/output names to prompt IDs, for example:
-
-- session recap prompt
-- structured event extraction prompt
-- glossary suggestion prompt
-- player-facing summary prompt
-
-Prompt IDs used by Narratio should come from the deployed Scriptorium prompt library.
-
-## 10. Profiles
-
-- Prompts may declare `default_profile`.
-- Narratio may omit `--profile` to use prompt default profile.
-- Narratio may pass `--profile` to force profile selection.
-- This enables environment/profile selection like `local-fast`, `local-quality`, `frontier`, `batch`, or test profiles.
-- Profile names should generally be Narratio configuration values.
-
-## 11. Runtime Overrides
-
-Supported runtime override flags:
-
-- `--llm-base-url`
-- `--model`
-- `--api-key-env`
-- `--temperature`
-- `--max-tokens`
-- `--top-p`
-- `--timeout`
-
-Guidance:
-
-- Keep normal model/runtime settings in Execution Profiles.
-- Use runtime overrides only for explicit per-run exceptions, tests, or operator overrides.
-- Never pass raw API keys on the command line.
-- `--api-key-env` names an environment variable; Narratio must ensure that variable is set in subprocess environment.
-
-## 12. Config Behavior
-
-- Default config path: `/etc/scriptorium/config.yml`.
-- `--config <PATH>` overrides default path.
-- Missing default config is allowed by Scriptorium.
-- If `--config` is provided explicitly, the file must exist and be valid.
-- CLI flags override `config.yml`.
-- `config.yml` overrides built-in application defaults.
-
-Narratio can either:
-
-- rely on system default config path, or
-- carry an explicit config path and pass `--config`.
-
-## 13. Environment Handling
-
-Subprocess environment recommendations:
-
-- Pass through required API-key environment variables referenced by `api_key_env`.
-- Do not pass raw API keys as CLI arguments.
-- Avoid logging full environment dumps.
-- Capture stdout and stderr separately.
-- Use a controlled working directory.
-- Prefer absolute artifact paths.
-
-## 14. Output Handling
-
-For `scriptorium run`:
-
-- Use `--out` when Narratio needs durable artifact files.
-- Without `--out`, artifact content is written to stdout.
-- Preferred orchestration pattern: always use `--out`, then treat the file as stage output artifact.
-- Capture stderr for diagnostics.
-
-For `scriptorium render`:
-
-- Use `--out` to store render diagnostics.
-- Use `--format json` when tests need to inspect selected profile, effective runtime settings, input hashes, prompt hash, and rendered messages.
-
-## 15. Exit Status and Errors
-
-Current CLI behavior (verified from implementation/tests):
-
-- `0`: success.
-- `1`: runtime/parse/config/load/render/generation/IO error.
-- `2`: run completed but output validation failed (`ValidationFailed`).
-
-Additional details:
-
-- On `run`, output artifact write happens before exit code selection. If validation fails, artifact may still be written and exit code is `2`.
-- `stderr` carries both errors and normal run summary output; non-empty stderr alone does not imply failure.
-- `render` returns `0` on success and `1` on failures.
-
-Narratio should treat non-zero exit codes as failed stage execution, but may record generated artifact paths if a run exited `2` and output file exists.
-
-## 16. Recommended Narratio Integration Pattern
-
-1. Build CLI args from Narratio stage configuration.
-2. Use subprocess context cancellation/timeout.
-3. Pass absolute input paths.
-4. Pass `--out` to a session-scoped artifact path.
-5. Add `--var` metadata values.
-6. Optionally add `--config`.
-7. Optionally add `--profile`.
-8. Ensure required API-key env vars are present.
-9. Run subprocess synchronously.
-10. Capture stdout/stderr separately.
-11. On success, store output artifact path and invocation metadata in stage artifacts.
-12. On failure, store exit code and stderr diagnostics in stage status.
-
-## 17. Suggested Narratio Configuration Shape
-
-Illustrative (not required schema):
-
-```yaml
-scriptorium:
-  config_path: /etc/scriptorium/config.yml
-  stages:
-    session_recap:
-      prompt_id: dnd.session_recap
-      profile_id: local-quality   # optional
-      inputs: [transcript, glossary, previous_recap]
-      vars: [session_id, session_date, campaign_name]
-      output_path_template: artifacts/{session_id}/session_recap.md
-      timeout: 2m
-      render_debug: false
-```
-
-The key idea: map Narratio stage/artifact names to prompt ID, optional profile, expected inputs, and output destination.
-
-## 18. Testing Strategy for Narratio Integration
-
-- Use `scriptorium render --format json` to verify command construction without LLM calls.
-- Use dedicated test prompt/profile libraries for integration tests.
-- Use small fixture transcripts.
-- Verify missing-input failure behavior.
-- Verify prompt `default_profile` behavior.
-- Verify explicit `--profile` override behavior.
-- Verify `--config` behavior (default and explicit).
-- Verify output file creation when `--out` is used.
-- Verify stderr capture on failures.
-- Avoid real API keys in tests.
-
-## 19. Security and Privacy Notes
-
-- Never pass raw API keys on command line.
-- Do not log full rendered prompts by default; transcripts may contain sensitive content.
-- Avoid logging prompt content unless explicit debug mode is enabled.
-- Treat generated artifacts as potentially sensitive.
-- Use session-scoped, access-controlled output paths.
-- `api_key_env` names should come from environment management, not embedded secrets.
-
-## 20. Initial D&D Artifact Generation Examples
-
-These are examples only. Use prompt IDs from the deployed prompt library.
-
-Session recap:
-
-```bash
-scriptorium run \
-  --prompt dnd.session_recap \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input glossary=/work/session-42/glossary.yml \
-  --out /work/session-42/artifacts/session_recap.md
-```
-
-Structured events:
-
-```bash
-scriptorium run \
-  --prompt dnd.structured_events \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --out /work/session-42/artifacts/structured_events.json
-```
-
-Glossary suggestions:
-
-```bash
-scriptorium run \
-  --prompt dnd.glossary_suggestions \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input previous_recap=/work/session-41/artifacts/session_recap.md \
-  --out /work/session-42/artifacts/glossary_suggestions.md
-```
-
-Player-facing summary:
-
-```bash
-scriptorium run \
-  --prompt dnd.player_summary \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input structured_events=/work/session-42/artifacts/structured_events.json \
-  --out /work/session-42/artifacts/player_summary.md
-```
-
-## 21. Non-Goals
-
-Initial Narratio integration should not:
-
-- call Scriptorium internal Go packages
-- use HTTP API as the primary path
-- expect Scriptorium to read S3 refs directly
-- make Scriptorium responsible for Narratio stage state
-- make Scriptorium responsible for notification
-- require Scriptorium to understand D&D workflow semantics beyond prompt definitions
-
-## 22. Future Extension Notes
-
-Possible later extensions:
-
-- HTTP API integration
-- S3 artifact references if Scriptorium adds S3 reader support
-- storing render diagnostics alongside generated artifacts
-- token budgeting/prompt-size checks
-- batch execution if Scriptorium later adds batch support

docs/integrations/openai-compatible-chat.md

@@ -0,0 +1,121 @@
+# OpenAI-Compatible Chat Integration
+
+This is the outbound wire contract for Scriptorium's OpenAI-compatible
+chat-completions client.
+
+## Endpoint And Method
+
+Scriptorium uses the request endpoint override when present; otherwise it uses
+the configured client base URL. It removes a trailing slash and sends
+`POST /chat/completions`.
+
+For example, `http://localhost:8000/v1` becomes
+`http://localhost:8000/v1/chat/completions`.
+
+## Request Payload
+
+The payload always contains `model` and rendered `messages`. It additionally
+contains these fields when applicable:
+
+| Field | Inclusion |
+| --- | --- |
+| `session_id` | Non-empty rendered prompt session ID. |
+| `temperature` | Non-zero effective value or an explicit zero override. |
+| `max_tokens` | Non-zero effective value or an explicit zero override. |
+| `top_p` | Non-zero effective value or an explicit zero override. |
+| `service_tier` | Any non-empty configured value. |
+| `reasoning_effort` | Any non-empty configured value. |
+| `response_format` | Structured output is requested. |
+| provider-specific fields | Flattened from `extra_params`. |
+
+`service_tier` and `reasoning_effort` are forwarded without a provider value
+catalog; the selected backend decides which values it supports.
+
+`extra_params` are top-level JSON fields, not a nested object. Keys cannot be
+empty or collide with `model`, `session_id`, `messages`, `temperature`,
+`max_tokens`, `top_p`, `service_tier`, `reasoning_effort`, or
+`response_format`. Values must be JSON-serializable.
+
+A rendered `session_id` is sent as a top-level JSON field, not as a header.
+Empty values are omitted. The maximum length is 256 Unicode code points.
+
+Messages without cache control use string `content`. A message with cache
+control uses one text block:
+
+```json
+{
+  "role": "system",
+  "content": [{
+    "type": "text",
+    "text": "rendered text",
+    "cache_control": {"type": "ephemeral", "ttl": "1h"}
+  }]
+}
+```
+
+When the prompt omits cache-control `ttl`, the payload omits `ttl`.
+Structured JSON Schema output is sent as:
+
+```json
+{
+  "response_format": {
+    "type": "json_schema",
+    "json_schema": {
+      "name": "schema name",
+      "strict": true,
+      "schema": {"type": "object"}
+    }
+  }
+}
+```
+
+## Authentication And Timeout
+
+When a direct request API key is present, Scriptorium sends
+`Authorization: Bearer <key>` and does not read `api_key_env`. Otherwise, it
+resolves the configured non-empty `api_key_env` at request time and sends the
+same header. If neither mechanism supplies a key, it sends no
+`Authorization` header.
+
+The transport-wide safety cap is chosen at client construction. A positive
+timeout on a supplied `http.Client` takes precedence over a positive
+`Config.Timeout`; if neither is positive, the internal ten-minute default is
+used. The supplied client is cloned, and zero or negative timeout values are
+treated as unset.
+
+Separately, a positive effective `timeout_seconds` creates a deadline for each
+outbound generation call. Its value follows the execution-setting hierarchy:
+an explicit request override, then a non-zero profile value, then the
+600-second framework default. An explicit request override of zero disables
+only this generation deadline. Negative values are rejected before a request
+is sent.
+
+The complete observable rule is that the earliest caller-context deadline,
+transport cap, or positive generation deadline terminates the call. Transport
+and cancellation failures retain the generation-error classification.
+
+## Response Subset And Failures
+
+A successful provider response must supply non-empty
+`choices[0].message.content`. Scriptorium reads these optional or required
+usage fields when present:
+
+- `usage.prompt_tokens`
+- `usage.completion_tokens`
+- `usage.total_tokens`
+- `usage.prompt_tokens_details.cached_tokens`
+- `usage.cache_write_tokens`
+
+Missing cache usage is reported as zero. Invalid JSON, an empty choices array,
+or empty first-choice content is a malformed provider response. Network and
+request-construction failures, non-2xx responses, and malformed responses fail
+the outbound call. Provider response bodies are not exposed by this client.
+
+The client does not implement built-in retries, tool calls, top-level
+`cache_control`, or multi-request payload modes.
+
+## Related References
+
+Prompt schema preparation and runner orchestration are described in
+[runner internals](../internal/runner.md). Prompt and profile configuration is
+defined by the [configuration reference](../config.md).

docs/integrations/subprocess.md

@@ -0,0 +1,41 @@
+# Subprocess Integration
+
+This document covers process-boundary behavior for callers that invoke
+Scriptorium as a child process. Command syntax, flags, output, and exit codes
+are defined by the [CLI reference](../cli.md). Interface selection belongs in
+the [consumer integration overview](../consumers/api.md).
+
+## Process Contract
+
+Use `scriptorium render` when the caller needs prepared output without a model
+call, and `scriptorium run` for generation. Pass an explicit `--config` or
+make the configuration search paths available to the child process; configuration
+discovery, fields, profile selection, and credential mechanisms are defined in
+the [configuration reference](../config.md).
+
+Pass required API-key environment variables through the child environment. Do
+not place raw API keys in arguments. Keep the environment limited to the values
+needed for the selected profile.
+
+## Streams And Output Ownership
+
+Capture stdout and stderr separately. Stdout contains the requested artifact or
+prepared output unless the caller selects an output file; stderr contains
+summaries, diagnostics, and server messages. The exact destinations and status
+meanings are part of the [CLI reference](../cli.md), not a stable stderr data
+protocol.
+
+When using `--out`, the caller owns the output path, its permissions, and
+cleanup. Treat rendered prompts, generated artifacts, stdout, and stderr as
+potentially sensitive.
+
+## Cancellation And Recovery
+
+A CLI invocation performs one synchronous request and creates no durable run
+state. A supervising process that needs cancellation must terminate the child
+process according to its own process-management policy. A later invocation is a
+new request and can make another model call; there is no resume or checkpoint
+protocol.
+
+For deployment, filesystem permissions, and sensitive-artifact handling, see
+the [operations guide](../operations.md).

docs/internal/adapters.md

@@ -0,0 +1,113 @@
+# Adapter Internals
+
+## Purpose
+
+Adapters translate external inputs into domain requests, compose dependencies,
+and translate domain results or errors back to their interface. They own IO and
+presentation mechanics; use-case decisions remain in `internal/usecase`.
+
+External contracts are canonical in the [CLI reference](../cli.md), [HTTP API
+reference](../api.md), and [Go package contract](../consumers/pkg-scriptorium.md).
+
+## Components And Collaborators
+
+- `cmd/scriptorium` passes process arguments and streams to
+  `internal/adapter/cli`.
+- `internal/adapter/cli` parses commands, resolves application settings through
+  `internal/config`, constructs a runner, and owns process output handling.
+- `internal/adapter/http` decodes DTOs, maps them to `domain.RunRequest`, calls
+  a runner interface, and maps errors and results to HTTP DTOs.
+- The root `scriptorium` package maps its public types and options to internal
+  collaborators and maps selected internal errors to public sentinels.
+- `internal/format` formats prepared runs for the CLI; `internal/llm`,
+  `internal/prompt`, and source packages supply runner dependencies.
+
+## Wiring Flows
+
+### CLI
+
+The CLI resolves configuration before constructing dependencies. `run` builds a
+runner with the ordinary composite artifact reader and invokes `Runner.Run`;
+`render` uses the same wiring and invokes `Runner.Prepare`; `serve` replaces the
+file reader with the restricted artifact reader, builds an HTTP handler, and
+starts the server.
+
+Parser state records whether numeric runtime values were explicitly supplied.
+That presence is carried into `domain.ExecutionTargetOverride`, allowing the
+runner to distinguish omitted values from explicit zero overrides.
+
+### HTTP
+
+The handler first enforces transport limits, strict JSON decoding, and the
+minimal request shape. It maps DTO values to domain types without deciding
+prompt selection, source behavior, or validation semantics. On success it maps
+the domain result to the response DTO; on failure it uses `errors.Is` over
+runner, source, artifact, and profile errors to choose the public error mapping.
+
+The [HTTP API reference](../api.md) owns the route, DTO schema, status codes,
+and externally observable limit behavior.
+
+### Public Go Facade
+
+`NewEngine` applies public options, selects filesystem, `fs.FS`, single-file,
+or in-memory dependencies, and constructs a runner. The conversion functions
+copy maps and slices across the boundary so callers do not receive internal
+domain values. The facade maps selected internal errors to the public sentinel
+set and keeps direct request API keys out of public results.
+
+## Package-Local Guarantees
+
+- Adapters do not embed runner orchestration or source-loading decisions.
+- Configuration is resolved before adapter dependency composition.
+- CLI and HTTP create runners without a repairer; a repairer is available only
+  through explicit internal runner construction.
+- DTO conversion preserves explicit numeric-override presence.
+- Error mapping matches error identities, not error text.
+- No adapter creates durable run state; caller-selected output files are not
+  application state.
+
+## Failure And Verification Boundaries
+
+Keep external error payloads concise, preserve strict external decoding, and do
+not serialize resolved secret values. Validation content failures remain result
+state; runtime failures remain errors for the relevant adapter to map.
+
+Inspect focused tests when changing this area:
+
+- `internal/adapter/cli/run_test.go`
+- `internal/adapter/http/handler_test.go`
+- `engine_test.go`
+- `internal/format/prepared_run_test.go`
+
+Run the affected adapter package tests and recheck the relevant canonical
+contract. The [testing policy](../policy/testing.md) owns global test
+sufficiency guidance.
+
+## Change Recipes
+
+### Application Configuration Fields
+
+1. Add the field to the relevant `internal/config` shape and default handling.
+2. Parse and validate it, then preserve configuration and CLI-override
+   precedence while wiring it through its consuming adapter.
+3. Add focused configuration and adapter tests for parsing, mapping, and
+   effective behavior.
+4. Update the [configuration contract](../config.md) and any affected external
+   contract.
+
+### CLI Flags
+
+1. Add the flag to the relevant parser in `internal/adapter/cli/run.go`.
+2. Keep command scope and application-configuration precedence intentional.
+3. Add or update parser and command tests in
+   `internal/adapter/cli/run_test.go`.
+4. Update the [CLI contract](../cli.md) and affected maintained examples.
+
+### Adapter Capabilities
+
+1. Define or reuse the appropriate domain or use-case interface boundary.
+2. Implement translation and IO behavior without moving use-case decisions out
+   of `internal/usecase`.
+3. Add focused mapping, parsing, and error-behavior tests.
+4. Update this document and the affected public or integration contract. Update
+   [source internals](sources.md) when source-loading behavior changes.

docs/internal/llm.md

@@ -0,0 +1,86 @@
+# LLM Internals
+
+## Purpose
+
+`internal/llm` defines the provider-neutral `Client` interface and the
+OpenAI-compatible client implementation. The [OpenAI-compatible integration
+contract](../integrations/openai-compatible-chat.md) owns the outbound HTTP wire
+format and protocol behavior.
+
+## Construction
+
+`NewOpenAICompatibleClient` validates a non-empty configured base URL, records
+an optional default model, and resolves one transport cap. A supplied client
+with a positive timeout supplies that cap; otherwise a positive configured
+timeout is used, then the internal default.
+
+When callers supply an `http.Client`, construction clones it rather than
+mutating the caller's instance. A supplied client with a zero or negative
+timeout receives the resolved transport cap in the clone. The client stores the
+trimmed base URL, default model, and cloned client.
+
+## Generate Flow
+
+`Generate` receives a `domain.GenerateRequest` from the runner:
+
+1. validate the effective timeout and choose the request endpoint;
+2. map the domain request to the internal wire-request representation;
+3. validate and flatten extra parameters and encode JSON;
+4. derive a child context when the effective generation timeout is positive,
+   then create the HTTP request with that context;
+5. prefer a direct API key, otherwise resolve the configured key environment
+   variable;
+6. execute with the construction-time HTTP client, reject non-success status
+   responses without returning
+   provider response bodies; and
+7. decode the response subset into `domain.GenerateResponse`.
+
+`openAIChatRequestFromGenerateRequest` is the conversion boundary for effective
+model defaults, explicit numeric-presence state, rendered messages, structured
+output, and session-ID validation. `openAIChatRequestPayload` protects reserved
+fields and JSON encoding before an HTTP call. The external payload shape is
+defined only in the [integration contract](../integrations/openai-compatible-chat.md).
+
+## Error Categories
+
+The package uses these internal sentinels:
+
+- `ErrInvalidConfig` for invalid client construction;
+- `ErrInvalidRequest` for invalid effective generation input;
+- `ErrRequestFailed` for request construction or transport failures;
+- `ErrUnexpectedStatus` for non-success HTTP responses; and
+- `ErrMalformedResponse` for invalid or incomplete successful-response data.
+
+The runner maps an invalid LLM request to its invalid-request category and
+other LLM failures to its generation category. Adapters then apply their public
+error contracts.
+
+## Package-Local Guarantees
+
+- The default-model fallback happens before wire encoding.
+- Per-generation timeout handling derives a request context; it never replaces
+  or mutates the configured HTTP client's transport cap.
+- Direct API keys take precedence over environment lookup within this client.
+- Provider response bodies are discarded for non-success status responses.
+- The client does not implement retries, tool calls, or a stateful session
+  store.
+
+## Verification And Change Recipe
+
+Inspect:
+
+- `internal/llm/openai_compatible_client_test.go`
+- `internal/usecase/runner_test.go`
+- `internal/adapter/http/handler_test.go`
+
+When changing the client:
+
+1. keep domain-to-wire mapping inside `internal/llm` and preserve the `Client`
+   interface;
+2. test construction, timeout selection, mapping, and error categorization;
+3. update the [OpenAI-compatible integration contract](../integrations/openai-compatible-chat.md)
+   for any observable wire or protocol change; and
+4. update [runner internals](runner.md) if the client boundary or structured
+   output handoff changes.
+
+The [testing policy](../policy/testing.md) owns global test sufficiency.

docs/internal/overview.md

@@ -0,0 +1,48 @@
+# Internal Component Overview
+
+## Purpose
+
+This is the inventory of Scriptorium's implemented components for contributors.
+The [architecture policy](../policy/architecture.md) owns normative boundaries
+and invariants; public behavior belongs in the linked contracts.
+
+## Public And Command Entrypoints
+
+| Component | Implemented responsibility | References |
+| --- | --- | --- |
+| Root package `scriptorium` | Public Go facade that constructs the engine, exposes request/result types and options, and maps internal errors. | [Go package contract](../consumers/pkg-scriptorium.md), [adapter internals](adapters.md) |
+| `cmd/scriptorium` | Process entrypoint that delegates command execution to the CLI adapter. | [CLI contract](../cli.md), [adapter internals](adapters.md) |
+
+## Adapters, Domain, And Use Case
+
+| Component | Implemented responsibility | References |
+| --- | --- | --- |
+| `internal/adapter/cli` | Parses CLI commands, applies application wiring, and handles process input and output. | [CLI contract](../cli.md), [adapter internals](adapters.md) |
+| `internal/adapter/http` | Maps HTTP requests and responses to domain operations and maps public errors. | [HTTP API contract](../api.md), [adapter internals](adapters.md) |
+| `internal/domain` | Defines core request, result, output-contract, and LLM-boundary types. | [runner internals](runner.md) |
+| `internal/usecase` | Implements `Runner` preparation, execution, validation coordination, and the repairer boundary. | [runner internals](runner.md) |
+
+## Configuration And Sources
+
+| Component | Implemented responsibility | References |
+| --- | --- | --- |
+| `internal/config` | Loads application settings, applies defaults, and applies CLI overrides. | [configuration contract](../config.md), [adapter internals](adapters.md) |
+| `internal/defaults` | Holds compile-time default values used when application settings are resolved. | [configuration contract](../config.md) |
+| `internal/promptdef` | Loads prompt definitions from filesystem and `fs.FS` sources. | [configuration contract](../config.md), [source internals](sources.md) |
+| `internal/profile` | Loads filesystem and `fs.FS` execution profiles and combines profile repositories. | [configuration contract](../config.md), [source internals](sources.md) |
+| `internal/profile/builtin` | Provides embedded built-in execution profiles as a repository. | [configuration contract](../config.md), [source internals](sources.md) |
+| `internal/filecatalog` | Provides shared YAML discovery and source-root helpers. | [source internals](sources.md) |
+| `internal/artifact` | Reads inline and file-backed input artifacts. | [configuration contract](../config.md), [HTTP API contract](../api.md), [source internals](sources.md) |
+| `internal/prompt` | Renders prompt templates into messages. | [runner internals](runner.md) |
+
+## Formatting, Validation, And Model Access
+
+| Component | Implemented responsibility | References |
+| --- | --- | --- |
+| `internal/format` | Formats prepared-run information for CLI output. | [CLI contract](../cli.md), [adapter internals](adapters.md) |
+| `internal/validate` | Defines validation interfaces and provides standard filesystem and `fs.FS` schema validation. | [configuration contract](../config.md), [source internals](sources.md), [runner internals](runner.md) |
+| `internal/llm` | Defines the provider-neutral LLM client boundary and its OpenAI-compatible implementation. | [OpenAI-compatible integration](../integrations/openai-compatible-chat.md), [LLM internals](llm.md), [runner internals](runner.md) |
+
+Focused internal documents describe the components that have detailed
+orchestration, adapter, or source behavior. Package tests live alongside the
+implementation and are identified in those focused documents where relevant.

docs/internal/runner.md

@@ -0,0 +1,119 @@
+# Runner Internals
+
+## Purpose
+
+`internal/usecase.Runner` is the prompt-execution orchestrator. It prepares
+domain requests, invokes an injected LLM client, validates output, and returns
+domain results. Transport parsing, response mapping, and public type conversion
+remain outside this package.
+
+The [configuration reference](../config.md) owns prompt, profile, schema, and
+runtime-setting definitions. Public error behavior is defined by the
+[HTTP API](../api.md) and [Go package](../consumers/pkg-scriptorium.md)
+contracts.
+
+## Dependencies And Construction
+
+`Runner` receives these collaborators:
+
+- `promptdef.Repository`;
+- `profile.Repository`;
+- `artifact.Reader`;
+- `prompt.Renderer`;
+- `llm.Client`;
+- `validate.Validator`; and
+- an optional `OutputRepairer`.
+
+`NewRunner` constructs a runner without a repairer. `NewRunnerWithRepairer`
+accepts one explicitly. Adapters and the public engine choose concrete
+repositories and readers; the runner does not load application configuration.
+
+## Prepare Flow
+
+`Prepare` performs one deterministic preparation pass for a request:
+
+1. validate the prompt ID and load the prompt definition;
+2. hash the definition and select the explicit or default profile;
+3. load the profile and resolve effective execution settings;
+4. validate endpoint, model, and credential availability;
+5. resolve the output contract and, for JSON Schema output, load a structured
+   schema document before model execution;
+6. read and hash input artifacts;
+7. render messages and the session ID; and
+8. return a `PreparedRun` containing the effective state and rendered-prompt
+   hash.
+
+Execution settings merge defaults, profile values, and a request override.
+Numeric override presence is retained so explicit zero values are not confused
+with omissions.
+
+## Run And Validation Flow
+
+`Run` creates a run ID and timestamps, then calls `Prepare` rather than
+duplicating preparation. It sends the prepared prompt, effective target,
+target-presence state, and optional structured-output specification to the LLM
+client. It converts the returned content to an output artifact, validates it,
+and returns the artifact, validation, hashes, usage, and timing metadata.
+
+A validator can return a content result or an operational error. Content
+failures stay in the result; schema loading, compilation, and validator
+operational failures are returned as `ErrValidation`. The canonical distinction
+for callers is documented by the public contracts.
+
+## Repair Boundary
+
+Repair is an internal optional loop. It starts only when a repairer is present,
+the output contract permits one or more attempts, validation failed, and the
+validation mode is JSON or JSON Schema. Each repair receives the previous
+output, validation errors, effective target, structured-output specification,
+and attempt metadata; every repaired result is validated again.
+
+`NewDefaultOutputRepairer` delegates to the injected LLM client. CLI, HTTP, and
+the public engine use `NewRunner` and therefore do not inject this repairer.
+
+## Error Translation
+
+Runner sentinels identify failure categories for adapters:
+
+- `ErrInvalidRequest`
+- `ErrProfileRequired`
+- `ErrAPIKeyEnvMissing` and `ErrAPIKeyRequired`
+- `ErrPromptLoad`, `ErrProfileLoad`, and `ErrArtifactLoad`
+- `ErrPromptRender`
+- `ErrLLMGenerate`
+- `ErrValidation`
+
+Wrap errors with those sentinels and preserve their identities through
+`errors.Is`; adapters must not classify errors by message text. The runner
+passes direct keys only to the LLM boundary and never includes resolved key
+values in prepared or run results.
+
+## Package-Local Guarantees
+
+- `Run` always reuses `Prepare`.
+- Schema documents are loaded before the initial LLM call when structured output
+  is required.
+- Output validation records attempts used, including repair attempts.
+- Runner state is per request; the package does not create a durable run store
+  or manifest.
+- Source, renderer, validator, and LLM implementations remain injected
+  boundaries.
+
+## Verification And Change Recipe
+
+Inspect:
+
+- `internal/usecase/runner_test.go`
+- `internal/usecase/integration_test.go`
+- `engine_test.go`
+
+When changing orchestration:
+
+1. identify the collaborator boundary and the affected `Prepare` or `Run` state;
+2. preserve the `Run`-through-`Prepare` path and error identity;
+3. add focused runner or integration tests for changed state transitions,
+   validation, or repair behavior; and
+4. update the owning external contract and any affected source or LLM internal
+   document.
+
+The [testing policy](../policy/testing.md) owns global test sufficiency.

docs/internal/sources.md

@@ -0,0 +1,97 @@
+# Source Internals
+
+## Purpose
+
+This document describes how source packages load prompt definitions, profiles,
+schemas, and artifacts. The [configuration reference](../config.md) owns their
+user-facing formats and settings. The [HTTP API reference](../api.md) owns
+HTTP-visible artifact outcomes; [operations](../operations.md) owns deployment
+handling.
+
+## Prompt Definitions
+
+`internal/promptdef` provides filesystem and `fs.FS` repositories. Both use
+`internal/filecatalog` for recursive YAML discovery, deterministic ordering,
+display paths, and root cleaning.
+
+Repositories select a prompt by YAML ID and optional version rather than by
+path. They decode through strict YAML handling, reject duplicate matching
+definitions, and resolve `content_file` relative to the definition. The `fs.FS`
+implementation resolves content paths inside its source root; absolute paths and
+traversal outside that root are rejected before file access.
+
+## Profiles And Built-Ins
+
+`internal/profile` provides filesystem, `fs.FS`, and overlay repositories.
+`internal/profile/builtin` exposes embedded assets through the same repository
+interface.
+
+An overlay asks its primary source first. It falls back only when the primary
+reports `ErrProfileNotFound`; invalid YAML, duplicate IDs, validation failures,
+and raw-key failures are returned rather than hidden by fallback. This makes a
+custom ID override a built-in ID while retaining errors in the custom source.
+
+The public engine can overlay in-memory profiles ahead of both file-backed and
+built-in repositories. Profile field definitions, validation ranges, and the
+built-in catalog remain in the [configuration reference](../config.md).
+
+## Schemas
+
+`internal/validate` supplies `StandardValidator` for filesystem sources and
+`FSValidator` for `fs.FS` sources. Directory-backed validation loads the named
+schema path; it does not search directories by basename. `fs.FS` schema paths
+are cleaned and checked against their configured root, while a single-file
+source matches its file base name.
+
+The runner requests a schema document before generation when it needs
+structured output. JSON and schema mismatches in generated content are
+validation results; source access, decoding, registration, and compilation
+failures are operational errors.
+
+## Artifacts
+
+`internal/artifact` composes inline and file readers. The ordinary composite
+reader used by CLI and the public engine reads file references from the process
+filesystem. The restricted composite reader used by the HTTP adapter combines
+inline reading with a rooted file reader and optional byte limit.
+
+The rooted reader cleans paths and applies lexical containment without resolving
+symlinks. It checks relative references against the configured root and accepts
+absolute references only when they remain inside that lexical root. The OS still
+follows symlinks after that check. The public containment outcome is documented
+by the [HTTP API reference](../api.md); deployment permissions belong in
+[operations](../operations.md).
+
+## Failure Boundaries
+
+Source packages report repository, decoding, duplicate, validation, and read
+failures to their callers. They do not select public status codes or response
+schemas. The runner wraps source failures with use-case categories; adapters map
+them to their own external contract.
+
+Source reads use current filesystem or `fs.FS` content for each request. These
+packages create no manifests, checkpoints, or durable run state.
+
+## Verification And Change Recipe
+
+Inspect:
+
+- `internal/promptdef/repository_test.go`
+- `internal/profile/repository_test.go`
+- `internal/profile/builtin/repository_test.go`
+- `internal/artifact/reader_test.go`
+- `internal/validate/standard_validator_test.go`
+- `internal/usecase/integration_test.go`
+- `engine_test.go`
+
+When updating prompt, profile, schema, or built-in assets:
+
+1. keep assets valid for the strict loader and the relevant source boundary;
+2. update the [configuration reference](../config.md) when a file-format,
+   catalog, or default changes;
+3. run focused source and integration tests, including the built-in repository
+   test when embedded assets change; and
+4. update this document when discovery, precedence, containment, or failure
+   mechanics change.
+
+The [testing policy](../policy/testing.md) owns global test sufficiency.

docs/operations.md

@@ -0,0 +1,157 @@
+# Operations Guide
+
+## Scope And References
+
+This runbook covers deployment, normal operation, capacity planning, and safe
+recovery for Scriptorium. It does not redefine invocation syntax, configuration
+fields, or HTTP wire behavior.
+
+- [CLI reference](cli.md): commands, output destinations, and exit codes.
+- [Configuration reference](config.md): configuration, prompt/profile/schema
+  formats, defaults, and credentials.
+- [HTTP API reference](api.md): route, request/response schema, status codes,
+  limits, and HTTP artifact access.
+- [Consumer integration overview](consumers/api.md): caller responsibilities.
+
+## Operational Model And State
+
+Scriptorium handles one prompt request for each CLI invocation or HTTP request.
+It has no durable run store, archive, checkpoint, cache, or resume mechanism.
+A failed or interrupted request is recovered by correcting its inputs,
+configuration, or environment and submitting a new request.
+
+Generated artifacts, rendered prompts, model output, and run metadata are
+caller-owned data. Retention, encryption, backup, and deletion are deployment
+responsibilities.
+
+## Deploy The Filesystem And Process
+
+Provide the process with readable prompt, profile, and schema sources. Keep
+prompt templates adjacent to the prompt definitions that reference them. For an
+HTTP deployment that accepts file artifacts, use a dedicated, narrow artifact
+directory rather than a general-purpose or sensitive filesystem tree.
+
+Run Scriptorium under an identity that can:
+
+- read only the prompt, profile, schema, and allowed input-artifact paths it
+  needs;
+- read the required credential environment variables without writing them to
+  files or logs; and
+- write only caller-selected output locations when CLI output files are used.
+
+Do not make the HTTP artifact directory writable by untrusted users. The HTTP
+artifact containment behavior is lexical and the operating system follows
+symlinks; account for that when choosing ownership and mount boundaries. See
+the [HTTP API reference](api.md) for the externally observable behavior.
+
+## Supply Credentials And Protect Runtime Data
+
+Set secret values in the process environment and configure only their
+environment-variable names. Do not put raw keys in configuration, prompt or
+profile files, process arguments, HTTP payloads, captured command lines, or
+debug dumps.
+
+Treat stdout, stderr, prepared-run output, generated artifacts, and HTTP
+responses as potentially sensitive. Send service logs to a controlled collector
+and apply the same retention and access rules as for model input and output.
+
+## Run A Normal Workflow
+
+Before changing production inputs, profiles, or schemas:
+
+1. confirm the deployed configuration selects the intended sources and model
+   credentials;
+2. use [`render`](cli.md) with the same request inputs and variables to confirm
+   preparation without a model call;
+3. use [`run`](cli.md) for generation; and
+4. retain or discard validation-failed output according to the caller's
+   policy.
+
+The [maintained render script](../examples/render-markdown-summary.sh) is a
+copyable preflight example. The CLI reference owns its complete invocation and
+exit semantics.
+
+## Expose The HTTP Service
+
+The HTTP service has no built-in authentication or authorization. Place it on a
+trusted network or behind an authenticated reverse proxy, API gateway, or
+equivalent access control. Restrict who can reach it and who can read the
+artifact root.
+
+Use a service manager or supervisor appropriate to the deployment to manage
+process lifetime, restart policy, log capture, and environment injection. The
+[HTTP API reference](api.md) owns client request shapes, status behavior, and
+artifact-access outcomes.
+
+## Plan Capacity And Limits
+
+Capacity is primarily determined by concurrent model calls, input and output
+sizes, schema complexity, provider latency, and network behavior. Size limits
+protect request bodies, HTTP file artifacts, and encoded responses; configure
+them through the [configuration reference](config.md) and rely on the
+[HTTP API reference](api.md) for their response effects.
+
+Before increasing a limit:
+
+1. measure representative input, generated-output, and optional raw-output
+   sizes;
+2. confirm memory, network, and upstream-provider capacity;
+3. retain an upstream request-size and authentication boundary; and
+4. test the intended workload in a non-production environment.
+
+For large local inputs, prefer a controlled file-artifact directory over
+placing arbitrary paths on the service host. Avoid disabling a limit unless an
+equivalent trusted control exists elsewhere.
+
+## Diagnose And Recover
+
+### Preparation Or Configuration Failure
+
+Capture the CLI diagnostic or HTTP error response, then verify the selected
+configuration, prompt ID, profile selection, source readability, and input
+mapping. Use `render` with the same request when it is unclear whether failure
+occurs before model execution. Consult the [CLI reference](cli.md), the
+[configuration reference](config.md), and the [HTTP API reference](api.md) for
+the exact interface contract.
+
+### Credential Or Provider Failure
+
+Confirm that the process environment contains the configured credential name
+without printing the secret. Check endpoint reachability and provider health
+from the process network. If preparation succeeds but generation fails, inspect
+the selected model settings in prepared output and the service's controlled
+logs. Correct the deployment or provider issue, then submit a new request.
+
+### Artifact Or Permission Failure
+
+Verify that the process can read the intended local input. For HTTP file
+artifacts, verify the deployment's artifact root, ownership, path layout, and
+file size. Do not widen filesystem permissions or the allowed root merely to
+make an arbitrary path work; move or copy the required artifact into the
+controlled location instead.
+
+### Validation Failure
+
+A generated-content validation failure is distinct from a runtime failure.
+CLI `run` reports the validation result and error count in its success summary;
+it does not print the individual validation messages. For HTTP, inspect the
+validation object in the response according to the [HTTP API reference](api.md).
+
+Use rendered input and generated output to determine whether prompt instructions,
+the selected model, or the schema needs correction. If schema loading or
+compilation itself fails, correct the source deployment or schema document
+before rerunning.
+
+### HTTP Limit Or Request Failure
+
+Compare the request, artifact, or expected response size with the deployed
+configuration, and validate the request against the [HTTP API reference](api.md).
+Reduce the payload, use an appropriate controlled artifact source, omit
+unneeded raw output, or adjust the deployment limit after capacity review.
+
+## Cleanup And Reruns
+
+Because no run state is retained, cleanup concerns caller-owned output files,
+logs, and artifacts only. Remove or rotate them using the deployment's normal
+retention policy. After a correction, rerun the request from the beginning;
+there is no safe resume point.

docs/policy/architecture.md

@@ -0,0 +1,76 @@
+# Architecture
+
+This document is the development architecture policy for Scriptorium.
+
+It is for developers and LLM coding agents. User-facing behavior belongs in `README.md` and the docs under `docs/` that target operators/users.
+
+## System Shape
+
+Scriptorium is a narrow prompt-execution application with three executable
+entry paths: CLI `run`, CLI `render`, and the HTTP service started by `serve`.
+It also provides a public Go package for in-process use. Its current component
+inventory is maintained in the [internal overview](../internal/overview.md).
+
+Domain behavior is centralized in `internal/usecase` and `internal/domain`.
+
+## Core Principles
+
+- Keep orchestration narrow: Scriptorium executes one prompt request; it is not a multi-step workflow engine.
+- Keep adapter logic thin: adapters map external shapes to domain requests/results and should not hold domain decisions.
+- Keep boundaries explicit: repositories/loaders/renderers/validators/LLM client stay behind package interfaces.
+- Keep external decoding strict: configuration, prompt, and profile YAML and
+  HTTP JSON should reject unknown fields.
+- Keep secrets out of payloads: raw API key values must not be accepted or emitted.
+
+## Dependency Direction
+
+- Adapters translate external shapes and IO concerns; they do not make
+  use-case decisions.
+- Use-case and domain code depend on explicit repository, renderer, validator,
+  and LLM interfaces rather than adapter implementations.
+- Source, rendering, validation, and LLM implementations remain behind their
+  package boundaries.
+- Dependency-specific types must not leak across unrelated package boundaries.
+- Prefer the standard library; add an external dependency only when it
+  materially reduces risk or complexity.
+
+## State And Persistence Policy
+
+Scriptorium has no durable run-state store.
+
+- No built-in resume/checkpoint/archive behavior.
+- Recovery model is rerun after correcting inputs/config/environment.
+
+## Contract Ownership
+
+The [CLI](../cli.md), [configuration](../config.md), [HTTP API](../api.md),
+[public Go package](../consumers/pkg-scriptorium.md), and
+[integration](../integrations/) documents own their respective external
+contracts. This policy keeps only the architectural boundaries that govern
+their implementation.
+
+## Error Handling And Logging
+
+- Wrap errors with domain/operation context.
+- Map domain errors to adapter-appropriate statuses/codes without leaking sensitive internals.
+- Never emit raw secret values.
+
+## Testing And Documentation
+
+Testing philosophy and change-validation expectations are defined by the
+[testing policy](testing.md). Documentation ownership and maintenance rules are
+defined by the [documentation policy](documentation.md).
+
+## Architectural Invariants
+
+- `Runner.Run` reuses `Runner.Prepare` flow.
+- Raw API key values must not be accepted through external configuration or
+  request payloads, and resolved secret values must not be emitted.
+
+## Non-Goals
+
+- Do not move orchestration responsibilities from external callers into Scriptorium.
+- Do not add adapter-specific business logic in `internal/adapter/*` packages.
+- Do not bypass repository/renderer/validator/LLM boundaries by introducing cross-package coupling.
+
+Work that is not implemented belongs in `docs/roadmap/`.

docs/policy/documentation.md

@@ -0,0 +1,163 @@
+# Documentation Policy
+
+## Purpose
+
+This policy assigns each documentation topic to one canonical owner. Its goal is
+to keep this repository's documentation accurate, concise, discoverable, and
+resistant to drift for users, operators, developers, integrators, and LLM
+coding agents.
+
+## Core Rules
+
+### One Canonical Owner
+
+Each authoritative fact belongs in one document. A non-owning document may give
+a short, stable summary for orientation, but it must link to the canonical owner
+instead of repeating volatile details.
+
+Volatile details include commands, flags, configuration fields and defaults,
+module keys, schemas, file names, paths, status codes, retry behavior, and
+runtime guarantees. If readers could reasonably treat a statement as a
+contract, maintain it only in the owning document.
+
+Minimal tested usage examples are allowed outside the owning contract when this
+policy assigns them an orientation or instructional purpose. They must link to
+the canonical contract and must not redefine complete syntax, defaults, or
+semantics.
+
+### Current And Future Behavior
+
+Outside `docs/roadmap/`, documentation describes implemented behavior only.
+Partial features may be described only to their implemented boundary.
+
+ADRs are the narrow exception: an ADR may record an accepted architectural
+decision before implementation, but acceptance must not be presented as proof
+that the behavior exists. The roadmap owns implementation status and sequencing
+until the decision is implemented. Current architecture, user, operator,
+integration, and internal documentation are updated when the behavior lands.
+
+### Audience And Detail
+
+Write for the document's stated audience and include only the detail needed for
+its owned topic. User and operator docs should not expose implementation detail.
+Developer docs should link to user-facing and external contracts rather than
+restate them.
+
+### Examples
+
+Complete copyable files belong in `examples/`. Documentation may use the
+smallest illustrative snippet needed to explain its owned topic, but should link
+to maintained examples instead of embedding a second complete copy.
+
+Examples must be valid, secret-free, and tested where practical. Commands and
+configuration used in documentation should match the application.
+
+### Security And Privacy
+
+Documentation and examples must not contain real credentials, private keys,
+private environment dumps, sensitive source material, or private infrastructure
+details unless intentionally public. Document secret-handling mechanisms, not
+secret values.
+
+## Canonical Ownership
+
+| Topic | Canonical owner | Owned content | Content owned elsewhere |
+| --- | --- | --- | --- |
+| Product orientation and minimal end-to-end quickstart | `README.md` | What this project is, why it is useful, one shortest successful invocation, and links onward. | Complete command reference, configuration reference, operational procedures, implementation detail. |
+| Contributor entry point | `docs/development.md` | Task-oriented reading guide, minimal contributor orientation, baseline validation commands, and links to canonical docs. | Package inventory, architecture rules, subsystem behavior, and detailed change recipes, which belong in the relevant internal component document. |
+| Current application architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
+| Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression-test policy, and criteria for adding, rewriting, or deleting tests. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
+| CLI contract | `docs/cli.md` | Commands, arguments, flags, invocation semantics, and exit codes. | End-to-end operating procedures, configuration field definitions, runtime filesystem layout, module implementation details. |
+| Configuration contract | `docs/config.md` | Discovery and precedence, file schema, fields, defaults, environment overrides, validation rules, and user-selectable module or validator keys. | Complete example files, CLI syntax, runtime state lifecycle, module implementation details. |
+| Operations | `docs/operations.md` | Runtime workflows, physical filesystem and state layout, output, cache, and debug handling, resume, cleanup, permissions, recovery, and operational limits. | CLI flag syntax, configuration field definitions, logical output schemas, implementation mechanics. |
+| Public HTTP contract | `docs/api.md` | Routes, authentication, media types, request and response schemas, status codes, pagination, caching, idempotency, rate limits, and HTTP retry semantics. | Client walkthroughs, upstream or downstream integration internals, implementation detail. |
+| Consumer guidance | `docs/consumers/` | Task-oriented use of the public interface, minimal client examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| External and durable integration contracts | `docs/integrations/` | External file formats and protocols, upstream and downstream contracts, logical output bundle paths and schemas, media types, and compatibility behavior. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
+| Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
+| Internal component behavior | Other files under `docs/internal/` | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
+| Architectural decision history | `docs/adr/` | Significant decisions, context, alternatives, rationale, consequences, and supersession history. | Current behavior reference, implementation status, task sequencing. |
+| Future work and implementation status | `docs/roadmap/` | Proposed, accepted, deferred, or rejected work; implementation status; sequencing; and task breakdowns. | Implemented behavior reference and architectural decision rationale. |
+| Complete copyable artifacts | `examples/` | Maintained configuration, inputs, and other files intended to be copied or run. | Field-by-field reference, command reference, prose explanation. |
+
+Documents that do not exist are required only when the corresponding interface
+or responsibility exists. Do not create placeholder API, consumer, integration,
+or operations documents for behavior the application does not have.
+
+## Boundary Rules
+
+### Orientation
+
+The README owns product orientation. The developer guide routes contributors.
+Architecture owns normative structure. Internal overview owns the current
+concrete component map. These documents may link to one another but should not
+maintain parallel package or behavior descriptions.
+
+### Commands, Configuration, And Operations
+
+CLI documentation answers how to invoke the application. Configuration
+documentation answers what settings mean. Operations answers what happens to
+runtime state and how to operate or recover the application. When a workflow
+crosses these topics, choose the document that owns the task and link to the
+other contracts.
+
+### Contracts And Implementation
+
+Integration and API documents define externally observable shapes and
+semantics. Internal documents explain how this project implements or consumes
+those contracts. Internal docs may name a field, file, or protocol to identify
+a dependency, but must link to its canonical contract for the definition.
+
+### Security Topics
+
+This policy owns what documentation and examples may contain. Architecture owns
+application security invariants. Configuration owns credential-supply
+mechanisms. Operations owns permissions and handling of sensitive runtime
+artifacts. Internal docs own implementation mechanisms only.
+
+## Architecture Decision Records
+
+Use sequentially numbered ADR filenames such as
+`0001-record-architecture-decisions.md`. Follow the lightweight Nygard format:
+
+1. title;
+2. status;
+3. date;
+4. context;
+5. decision;
+6. alternatives considered;
+7. consequences.
+
+Use one of these statuses:
+
+- **Proposed:** the decision is under consideration and may change;
+- **Accepted:** the decision is approved, whether or not implementation is
+  complete;
+- **Rejected:** the proposed decision was considered and not adopted;
+- **Superseded:** a later ADR replaces the accepted decision.
+
+A proposed ADR transitions to accepted or rejected. An accepted ADR transitions
+to superseded only when a later accepted ADR replaces it. An ADR may be created
+as accepted when the decision has already been made.
+
+Treat the decision content of an accepted ADR as immutable. Its status and
+supersession metadata may be updated, but a changed decision requires a new ADR.
+A superseded ADR must link to its replacement, and the replacement must link
+back to the superseded ADR. Rejected architectural alternatives belong in the
+ADR; rejected product ideas belong in the roadmap.
+
+## Maintenance
+
+When behavior changes, update its canonical owner in the same change. If
+ownership moves, remove the old definition and replace it with a link where
+navigation remains useful.
+
+Before completing documentation work:
+
+- verify affected behavior and examples;
+- check commands, flags, fields, defaults, schemas, and paths against their
+  implementation;
+- keep unimplemented behavior in the roadmap, subject to the ADR exception;
+- remove stale references and validate links;
+- confirm that non-owning documents summarize and link rather than redefine;
+- confirm that no secrets or sensitive private data were added.

docs/policy/testing.md

@@ -0,0 +1,296 @@
+# Testing Policy
+
+## Purpose
+
+Our tests exist to make **incorrect changes expensive and correct changes cheap**.
+
+We do not optimize for test count, line coverage, exhaustive isolation, or the fewest possible tests. We optimize for sufficient confidence in important behavior while imposing as little unnecessary friction as possible on future development.
+
+## Every test has a cost
+
+Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
+
+A test must be:
+
+- written and reviewed;
+- understood by future maintainers and coding agents;
+- executed in local and CI workflows;
+- diagnosed when it fails;
+- updated when legitimate behavior changes;
+- maintained as fixtures, APIs, and dependencies evolve; and
+- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
+
+Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+
+A test is warranted only when the confidence it provides justifies these costs.
+
+Apply this cost-benefit analysis at two levels:
+
+1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
+2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
+
+The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
+
+Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
+
+The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
+
+## Default testing style
+
+Use a **classical/Detroit-style** approach:
+
+- Test observable behavior, resulting state, contracts, and invariants.
+- Use real internal collaborators when they are fast and deterministic.
+- Use fakes, stubs, or mocks primarily at expensive, nondeterministic, destructive, or external boundaries.
+- Prefer package-level behavioral tests over tests coupled to private helpers or internal call sequences.
+- Treat exact collaborator interactions as testable behavior only when the interaction itself is a requirement.
+
+Examples of appropriate seams include clocks, randomness, subprocesses, remote APIs, object storage, email, and paid LLM calls.
+
+## Test execution requirements
+
+Tests in the default suite must be deterministic, offline, and independent of real credentials. They must not invoke paid APIs or depend on mutable external services. Tests that require live infrastructure must be explicitly opt-in and clearly separated from the default suite.
+
+Control clocks, randomness, environment variables, and other process-global or machine-specific state when they affect behavior. Tests should be safe to run repeatedly and alongside other tests without depending on execution order or state left by an earlier test.
+
+## What deserves tests
+
+Prioritize tests for:
+
+1. Public and package-level contracts.
+2. Domain rules and important invariants.
+3. Boundary conditions and malformed input.
+4. Failure handling, cancellation, retries, recovery, and partial success.
+5. Serialization, schemas, compatibility, and round trips.
+6. Previously observed or plausible regressions.
+7. Representative integration and end-to-end workflows.
+
+A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
+
+For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+
+Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
+
+## Choose the right test boundary
+
+Test through the narrowest stable boundary that expresses the behavior clearly.
+
+This is often the package API, but it may instead be:
+
+- a smaller pure function when dense domain logic is most clearly isolated there;
+- a package-level operation when several internal collaborators jointly produce the behavior; or
+- a larger integration boundary when correctness emerges from interaction with a real dependency.
+
+Do not force all behavior through oversized end-to-end tests. Do not test every private helper merely because it exists. Choose the boundary that gives durable confidence with the least incidental coupling.
+
+## Test behavior, not implementation
+
+A test should protect a decision, contract, or invariant—not memorialize the current implementation.
+
+Before adding or retaining a test, ask:
+
+> What realistic defect would this test catch?
+
+A test is suspect when its main purpose is to detect that someone:
+
+- changed an internal constant;
+- renamed or split a private helper;
+- reordered equivalent internal operations;
+- changed incidental formatting;
+- replaced one correct algorithm with another; or
+- refactored internal object structure without changing behavior.
+
+Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
+
+A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
+
+## Expected effects of different changes
+
+Use the following expectations when evaluating test failures and test maintenance:
+
+| Change | Expected effect on tests |
+|---|---|
+| Internal refactor that preserves behavior | Existing tests should normally remain unchanged and continue to pass. |
+| Change to an internal default with no contractual significance | Behavioral tests should normally remain unchanged; tests should derive expectations from configuration or relationships rather than duplicate the old value. |
+| Intentional change to public behavior, policy, schema, or compatibility guarantees | The relevant tests should be reviewed and changed deliberately. |
+| Accidental violation of a contract or invariant | Tests should fail; fix the production code rather than rewriting the tests to accept the defect. |
+
+A test failing is not the same as a test needing to be edited. Many tests may correctly fail because of one production defect. The maintenance smell is a correct internal change that requires unrelated expectation updates throughout the suite.
+
+## Separate mechanism from policy
+
+Configurable thresholds and defaults must not be duplicated throughout the test suite.
+
+For example, do not encode an internal concurrency limit indirectly:
+
+```go
+// Production policy:
+const maxConcurrency = 4
+
+// Brittle test:
+err := startProcesses(5)
+require.Error(t, err)
+```
+
+Instead, test the mechanism relationally:
+
+```go
+const limit = 2
+runner := NewRunner(limit)
+
+require.NoError(t, runner.Start(limit))
+require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
+```
+
+The test should prove:
+
+- the configured limit is accepted; and
+- one beyond the configured limit is rejected.
+
+The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
+
+Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
+
+For concurrency limits, test both kinds of behavior when relevant:
+
+1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
+
+Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
+
+## Avoid semantic duplication across layers
+
+Each behavior should have a clear test owner.
+
+- Parser tests own parsing cases.
+- Validator tests own validation rules.
+- Domain tests own transformations and invariants.
+- Adapter tests own external integration behavior.
+- Orchestrator tests own coordination and failure propagation.
+- CLI tests own argument and configuration mapping.
+- End-to-end tests prove that representative assembled workflows work.
+
+Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
+
+Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
+
+## Use test doubles deliberately
+
+Choose the least elaborate test double that provides the required control or observation.
+
+As a default:
+
+1. Prefer real collaborators when they are fast and deterministic.
+2. Use small in-memory fakes when realistic stateful behavior is helpful.
+3. Use stubs when a dependency only needs to provide controlled responses.
+4. Use mocks when the interaction itself is contractual.
+
+Mocks are appropriate when the contract includes facts such as:
+
+- a notification is sent exactly once;
+- a transaction is committed only after successful writes;
+- cancellation reaches a subprocess;
+- an expensive API is called no more than once; or
+- a security audit event is emitted.
+
+Do not use mocks merely to isolate every object or reproduce the implementation's call graph.
+
+## Go-specific guidance
+
+Use:
+
+- table-driven tests for meaningful behavioral categories and boundaries;
+- `t.TempDir()` for real filesystem behavior;
+- `httptest.Server` for realistic HTTP interactions;
+- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- golden files only when the complete output is intentionally stable;
+- integration tests where correctness depends on component interaction; and
+- a small number of representative end-to-end tests.
+
+Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
+
+At CLI boundaries, prefer exit classifications, structured output, and the smallest stable semantic fragment needed to identify the error. Do not snapshot complete diagnostic wording unless it is contractual.
+
+Golden-file updates must require an explicit local flag. CI must not update golden files automatically, and reviewers must inspect the semantic diff before accepting an update.
+
+Keep tests readable and direct. Test helpers and fixture frameworks must earn their own maintenance cost; do not build elaborate test infrastructure for small or isolated needs.
+
+## Coverage
+
+Coverage is a diagnostic, not a target.
+
+Use it to find untested critical branches and unexpectedly weak packages. Do not write low-value tests solely to increase a percentage, and do not infer test quality from coverage alone.
+
+Pure domain logic will often warrant higher coverage than CLI wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
+
+Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
+
+## Regression tests
+
+A bug fix should normally include a regression test that fails before the fix and passes afterward.
+
+Retain the test when the defect could realistically recur and its consequences justify the ongoing cost. Prefer the narrowest durable test of the violated contract or invariant; do not preserve accidental implementation details from the original bug.
+
+Not every historical bug requires a permanent test. If the underlying design has made recurrence impossible, the test has become redundant, or a stronger invariant test now subsumes it, remove or consolidate it.
+
+## Deleting or rewriting tests
+
+Tests are maintained code, not permanent historical artifacts.
+
+Delete or rewrite a test when its maintenance cost exceeds the confidence it provides.
+
+Strong candidates include tests that:
+
+- require updates after harmless internal changes;
+- directly assert private constants without protecting a real contract;
+- duplicate the same policy across several layers;
+- verify mock choreography rather than outcomes;
+- snapshot large amounts of incidental output;
+- test trivial private helpers already exercised through stable package behavior;
+- protect risks already covered more effectively elsewhere;
+- are flaky, misleading, obsolete, or disproportionately expensive to diagnose; or
+- no longer correspond to a plausible failure mode.
+
+Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
+
+Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
+
+## Reviewing a proposed test
+
+Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
+
+1. What realistic defect would it catch?
+2. How likely is that defect?
+3. How consequential would it be?
+4. Is the behavior already protected elsewhere?
+5. At which layer should this behavior be owned?
+6. Does the test assert a durable contract or an incidental implementation detail?
+7. Could the implementation be refactored without changing the behavior and without editing this test?
+8. What should cause this test to fail?
+9. What legitimate changes should not cause this test to fail?
+10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
+11. Is there a smaller or more direct test that protects the same risk?
+
+Do not add the test when its expected lifetime cost exceeds its expected protective value.
+
+When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+
+## Definition of sufficient
+
+A test suite is sufficient when:
+
+- important contracts and invariants are protected;
+- meaningful boundaries and failure modes are exercised;
+- realistic and consequential regressions are credibly protected against silent recurrence;
+- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
+- important external boundaries have realistic integration coverage;
+- representative complete workflows are tested;
+- failures provide useful signal rather than redundant noise;
+- legitimate internal changes usually do not require test edits; and
+- additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
+
+Sufficiency is a risk judgment, not a coverage percentage or test count. Reassess it as the application, its users, and the consequences of failure evolve.
+
+The governing rule is:
+
+> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.

docs/roadmap/implementation.md

@@ -0,0 +1,340 @@
+# Migration Step 3 Implementation Plan
+
+## Status
+
+Completed on 2026-07-27. The framework characterization baseline is complete,
+and Migration Step 4 is the next planned work.
+
+## Objective
+
+Implement the target state in the
+[Step 3 framework-characterization roadmap](step3.md): make
+Promptkit-destined tests independent of Scriptorium-owned executable examples,
+close the identified public contract gaps, and leave every migration-critical
+behavior with a clear test owner.
+
+Follow the accepted ownership boundary in
+[ADR 0002](../adr/0002-split-promptkit-from-scriptorium.md) and the test-value
+and non-duplication rules in the [testing policy](../policy/testing.md).
+
+## Constraints
+
+- Execute the stages in order and satisfy each gate before proceeding.
+- Do not change production behavior, public types, method signatures, package
+  boundaries, or application interfaces.
+- Limit implementation changes to testdata, tests, and roadmap status.
+- Keep all default-suite tests deterministic, offline, and independent of real
+  credentials.
+- Preserve unrelated working-tree changes.
+- Do not copy the complete `examples/` tree into testdata.
+- Do not add tests merely to raise statement coverage.
+- Prefer extending or consolidating an existing test over adding a parallel
+  test for the same behavior.
+
+## Stage 1: Create The Framework Contract Corpus
+
+Create this Promptkit-destined fixture tree:
+
+```text
+testdata/framework/
+├── fixtures/
+│   ├── glossary.yml
+│   └── transcript.md
+├── profiles/
+│   ├── contract-fast.yaml
+│   └── contract-quality.yaml
+├── prompts/
+│   ├── contract.markdown_summary.system.md
+│   ├── contract.markdown_summary.user.md
+│   ├── contract.markdown_summary.yaml
+│   ├── contract.structured_events.system.md
+│   ├── contract.structured_events.user.md
+│   └── contract.structured_events.yaml
+└── schemas/
+    └── structured_events.schema.json
+```
+
+Define the corpus as follows:
+
+- `contract.markdown_summary`
+  - version `1.0.0`;
+  - default profile `contract-fast`;
+  - required `transcript` input and optional `glossary` input;
+  - system and user messages loaded through relative `content_file` paths; and
+  - Markdown output with basic validation and no repair attempts.
+- `contract.structured_events`
+  - version `1.0.0`;
+  - default profile `contract-quality`;
+  - the same two inputs;
+  - system and user messages loaded through relative `content_file` paths; and
+  - JSON output with JSON Schema validation through
+    `structured_events.schema.json` and no repair attempts.
+- `contract-fast`
+  - endpoint `http://localhost:8000/v1`;
+  - model `contract-fast-model`;
+  - temperature `0.2`, max tokens `500`, top-p `1`, and timeout `90`; and
+  - no credential requirement.
+- `contract-quality`
+  - endpoint `http://localhost:8000/v1`;
+  - model `contract-quality-model`;
+  - temperature `0.1`, max tokens `1000`, top-p `0.9`, and timeout `120`; and
+  - no credential requirement.
+- The schema requires an object containing an `events` array. Keep it small but
+  valid for the same JSON Schema draft currently used by the maintained
+  structured-output example.
+- The transcript and glossary contain short synthetic values suitable for
+  rendering and hash assertions. They must contain no private or real-world
+  data.
+
+Do not add application configuration, HTTP requests, executable scripts, or
+provider credentials to this corpus.
+
+In `engine_test.go`, add shared constants for the contract root, prompt IDs,
+profile IDs, and fixture paths. Replace
+`TestPrepareWorksWithExampleDirectoriesAndFileInputs` with
+`TestPrepareWorksWithFrameworkContractCorpus`, using table cases for the
+ordinary and structured prompts. Construct the public engine from the new
+directories, use the new file artifacts, and assert that each prompt renders
+with its intended profile. For the structured case, also assert that
+`PreparedRun.StructuredOutput` contains the loaded JSON Schema specification.
+This test is the real-parser acceptance check for the corpus.
+
+Run:
+
+```bash
+go test .
+go test ./internal/promptdef ./internal/profile ./internal/validate
+```
+
+### Stage 1 Gate
+
+- Every corpus file loads through its real owning parser.
+- Relative prompt content resolves from the prompt file location.
+- The structured schema decodes through the real validator's schema-document
+  loader and appears in the prepared structured-output specification.
+- No production or executable-example file changed.
+
+## Stage 2: Move Framework Tests Onto Framework-Owned Fixtures
+
+Update `engine_test.go` to use the new corpus.
+
+1. Reuse the contract constants introduced in Stage 1.
+2. Rename:
+   - `newExampleEngine` to `newContractEngine`;
+   - `newExampleEngineWithOptions` to `newContractEngineWithOptions`; and
+   - `exampleConfig` to `contractConfig`.
+3. Make `contractConfig` point at the corpus prompt, profile, and schema
+   directories.
+4. Replace each `./examples/...` dependency in `engine_test.go` with the
+   matching contract fixture or a purpose-built `t.TempDir`,
+   `fstest.MapFS`, or in-memory profile.
+5. Update assertions that intentionally identify fixture prompt IDs, profile
+   IDs, models, rendered text, or hashes to the contract values. Do not change
+   assertions that express independent public behavior.
+6. Rename tests whose names say “example” when they now exercise contract
+   testdata.
+
+Add `TestEngineRunWithDirectorySourcesAndFileInputs` to `engine_test.go`. It
+must assemble the public engine from the contract prompt, profile, schema, and
+file-artifact directories; inject a deterministic `LLMClient`; run
+`contract.structured_events`; and assert:
+
+- the prompt-selected `contract-quality` profile;
+- a non-empty run ID, prompt hash, rendered-prompt hash, and both input hashes;
+- provider-level JSON Schema structured output on the captured generation
+  request;
+- passed JSON Schema validation;
+- `application/json` artifact content;
+- preserved raw output and injected token usage; and
+- non-zero ordered timestamps with non-negative duration.
+
+Move the unique protection from
+`internal/usecase/integration_test.go` into this public test, then delete that
+internal integration test. Do not retain both assembled workflows.
+
+The only test references to `examples/` after this stage should be
+Scriptorium-owned adapter or maintained-example checks. In particular, this
+command must return no matches:
+
+```bash
+rg -n 'examples/' engine_test.go internal/usecase
+```
+
+Run:
+
+```bash
+go test .
+go test ./internal/usecase
+```
+
+### Stage 2 Gate
+
+- Public and framework-internal tests pass without reading Scriptorium-owned
+  executable examples.
+- The new public assembled workflow subsumes the deleted internal integration
+  test.
+- Scriptorium's maintained examples are unchanged.
+- No production file changed.
+
+## Stage 3: Consolidate And Complete Public Characterization
+
+### Execution-Setting Precedence
+
+Add a table-driven `TestEngineExecutionSettingPrecedence` in `engine_test.go`.
+Run through the public engine with an injected recording `LLMClient`. Cover
+these cases:
+
+1. a profile with zero-valued optional settings receives the documented
+   framework numeric defaults;
+2. non-zero profile settings replace those defaults;
+3. request settings replace profile settings; and
+4. explicit request numeric zero replaces non-zero profile settings.
+
+Across the table, verify the effective endpoint, model, temperature,
+max-tokens, top-p, timeout, service tier, reasoning effort, API-key environment
+name, and `extra_params` where the relevant layer supplies them. Verify
+`ExecutionTargetPresence` is false for omitted numeric request settings and
+true for every explicitly supplied numeric setting, including zero.
+
+Use relationally distinct values for each layer. Assert literal framework
+defaults only in the framework-default case because those values are part of
+the documented public contract. Use `t.Setenv` for every non-empty profile or
+request API-key environment name, assert only the environment-variable names,
+and never expose the test secret values.
+
+Consolidate overlapping assertions:
+
+- remove `TestPreparePreservesExplicitZeroExecutionOverrides` once the new
+  table protects that behavior; and
+- retain `TestRunPassesPreparedRequestToInjectedLLMClient` for rendered prompt,
+  direct-key, and structured-output handoff, but remove execution-precedence
+  assertions now owned by the table.
+
+### Caller Cancellation
+
+Add `TestEngineRunPropagatesCallerCancellation` using the built-in
+OpenAI-compatible client and a custom `RoundTripper`.
+
+- The transport must signal through a channel when `RoundTrip` begins.
+- It must block on `req.Context().Done()` and return the context error.
+- Start `Engine.Run` in a goroutine, wait for the transport signal, cancel the
+  caller context, and collect the result through a buffered channel.
+- Assert that the call returns and the error matches `ErrLLMGenerate`.
+- Do not use sleeps or elapsed-time assertions.
+
+### Injected Nil Response
+
+Add a valid-request case to `TestPublicErrorsSupportErrorsIs` whose injected
+`LLMClient` returns `(nil, nil)`. Assert `ErrLLMGenerate`. Extend the existing
+fake only as needed to express this case; do not create a mock framework.
+
+### Reserved Provider Parameters
+
+Add `TestRunRejectsReservedExtraParamsBeforeProviderCall`.
+
+- Use the built-in client with a custom immediate `RoundTripper` that records
+  whether it was invoked.
+- Supply a valid contract prompt and profile plus request
+  `ExtraParams: map[string]any{"model": "collision"}`.
+- Assert `ErrInvalidRequest`.
+- Assert that the transport was not invoked.
+
+Run:
+
+```bash
+go test .
+go test -count=20 .
+```
+
+### Stage 3 Gate
+
+- The four-layer precedence table and presence assertions pass.
+- Cancellation is deterministic and contains no wall-clock sleeps.
+- Nil injected responses and reserved parameters preserve their public error
+  categories.
+- Superseded assertions or tests have been removed rather than duplicated.
+- No production file changed.
+
+## Stage 4: Audit Ownership And Validate The Baseline
+
+### Ownership Audit
+
+Review the behavior list in [step3.md](step3.md) against the final suite.
+Confirm:
+
+- root facade and public contract tests are Promptkit-destined;
+- `internal/domain`, `internal/usecase`, `internal/promptdef`,
+  `internal/prompt`, `internal/profile`, `internal/profile/builtin`,
+  `internal/filecatalog`, general `internal/artifact`, `internal/validate`, and
+  `internal/llm` tests move with Promptkit-owned behavior;
+- CLI, application configuration, prepared formatting, HTTP DTO, strict JSON,
+  HTTP limit, and rooted artifact-containment tests remain Scriptorium-owned;
+- HTTP and CLI tests that currently construct internal runners or inspect
+  internal sentinels retain their observable assertions and are explicitly
+  deferred for boundary rewrites in Migration Step 4; and
+- no consequential behavior in the feature roadmap lacks a test owner.
+
+Do not create a permanent test-inventory document. Record any unexpected
+ownership exception in `step3.md`; otherwise the ownership table there is the
+complete disposition.
+
+### Full Validation
+
+Run:
+
+```bash
+go test ./...
+go vet ./...
+build_dir="$(mktemp -d)"
+go build -o "$build_dir/scriptorium" ./cmd/scriptorium
+go test -count=20 .
+go test ./internal/adapter/http -run TestMaintainedHTTPRunExampleMatchesRequestContract
+bash ./examples/render-markdown-summary.sh
+go run ./cmd/scriptorium render \
+  --config ./examples/config.full.yml \
+  --prompt generic.markdown_summary \
+  --input transcript=./examples/fixtures/transcript.md \
+  --input glossary=./examples/fixtures/glossary.yml \
+  --format text
+go run ./examples/go-library/prepare
+git diff --check
+```
+
+Validate every local Markdown link and path in the changed roadmap files.
+Confirm both configuration examples were accepted through the real
+configuration loader by the two render commands.
+
+Inspect the final diff and confirm the implementation changed only:
+
+- `testdata/framework/**`;
+- `engine_test.go`;
+- `internal/usecase/integration_test.go` by deletion;
+- `docs/roadmap/step3.md`;
+- `docs/roadmap/implementation.md`; and
+- the Step 3 status in `docs/roadmap/migration.md`.
+
+If a necessary change falls outside that list, stop and revise the plan or
+request direction rather than expanding scope implicitly.
+
+### Completion Bookkeeping
+
+After every check passes:
+
+1. update `step3.md` to state that the target state is complete and summarize
+   the characterized baseline without reintroducing an implementation log;
+2. add a Step 3 gate-status entry to `migration.md` with the completion date
+   and a short validation summary; and
+3. mark this implementation plan complete.
+
+Do not begin Migration Step 4 in the same change.
+
+### Stage 4 Gate
+
+- Every completion criterion in `step3.md` is satisfied.
+- The full suite and maintained examples pass offline.
+- The diff contains no production behavior or API change.
+- The main migration roadmap identifies Step 3 as complete and Step 4 as next.
+
+## Open Questions
+
+None.

docs/roadmap/migration.md

@@ -0,0 +1,308 @@
+# Promptkit Migration Roadmap
+
+## Status
+
+Accepted plan. Steps 1 through 3 are complete. Steps 4 through 9 remain
+proposed and are not yet implemented.
+
+## Objective
+
+Split the current repository into two projects:
+
+- **Promptkit**: the reusable Go framework, public Go facade, execution engine,
+  source and validation support, OpenAI-compatible client, extension
+  interfaces, and built-in execution-profile registry.
+- **Scriptorium**: a slim runnable application that imports Promptkit and
+  provides the CLI and HTTP interfaces.
+
+Scriptorium will become another downstream Promptkit consumer rather than the
+owner of the framework.
+
+## Compatibility And Migration Policy
+
+This is an intentionally breaking change.
+
+- New and migrated Go consumers must import Promptkit instead of Scriptorium.
+- Scriptorium will not retain type aliases, forwarding packages, deprecated
+  facade APIs, or other source-compatibility shims.
+- Existing consumers may continue using a previously tagged Scriptorium module
+  version until they are migrated.
+- The migration does not need to preserve compatibility between intermediate
+  development states. Each completed phase must instead leave the affected
+  repository internally consistent and tested.
+- Promptkit should initially preserve the useful shape and behavior of the
+  current public Go facade where doing so reduces extraction risk. Broader API
+  redesign should follow the split unless required to establish the new
+  boundary.
+
+## Target Ownership
+
+Promptkit should own application-neutral framework behavior:
+
+- public engine, request, result, option, extension, and error APIs;
+- prompt-definition loading and rendering;
+- execution profiles, overlays, and the built-in profile registry;
+- artifact-loading interfaces and general-purpose `file` and `inline` support;
+- schema loading and output validation;
+- LLM client boundaries and the OpenAI-compatible implementation;
+- preparation and execution orchestration;
+- framework and execution defaults.
+
+Scriptorium should own executable and transport concerns:
+
+- the `scriptorium` command and its `run`, `render`, and `serve` interfaces;
+- CLI parsing, output formatting, exit codes, and process behavior;
+- application-config discovery and CLI precedence;
+- HTTP routing, request and response DTOs, limits, and error/status mapping;
+- HTTP artifact-root and deployment security policy;
+- server and adapter defaults;
+- executable examples, operations guidance, and transport documentation.
+
+The intended dependency direction is:
+
+```text
+Scriptorium CLI and HTTP adapters
+                |
+                v
+             Promptkit
+                |
+                v
+     consumer-supplied sources and clients
+```
+
+Scriptorium must use Promptkit's public API. It must not depend on Promptkit
+implementation packages or reproduce framework orchestration.
+
+## Migration Steps
+
+### Step 1: Refresh And Synchronize Documentation
+
+Perform a repository-wide documentation refresh before migration development.
+
+At minimum:
+
+- reconcile all current-behavior documentation with the code, tests, examples,
+  defaults, and current public contracts;
+- introduce the planned documentation-policy updates;
+- establish an architecture decision record policy and canonical ADR location;
+- resolve stale, duplicated, or misplaced material;
+- validate documentation links and maintained examples;
+- leave future migration behavior in `docs/roadmap/` until implemented.
+
+**Gate:** Do not begin architectural migration work until the documentation
+refresh and policy updates are merged and the repository has an agreed,
+accurate baseline.
+
+**Gate status:** Complete as of 2026-07-26. The completed documentation
+refresh, follow-up verification, and layered-timeout correction remain recorded
+in repository history. Step 1 remains complete after that validation.
+
+### Step 2: Record The Architectural Decision And Detailed Boundary
+
+Create an ADR, under the policy established in Step 1, that records:
+
+- the decision to split Promptkit from Scriptorium;
+- the target ownership and dependency direction;
+- the selected Promptkit repository and Go module paths;
+- the breaking-change and versioning policy;
+- ownership of configuration fields and defaults;
+- artifact-reader and HTTP containment responsibilities;
+- local multi-repository development and release coordination;
+- documentation ownership after the split.
+
+Use the ADR to resolve any remaining public-boundary decisions before code is
+moved.
+
+**Gate:** The ADR is accepted, and every existing package, public contract,
+configuration category, and maintained asset has a target owner.
+
+**Gate status:** Complete as of 2026-07-26.
+[ADR 0002: Split Promptkit From Scriptorium](../adr/0002-split-promptkit-from-scriptorium.md)
+is accepted and records the required ownership and coordination decisions.
+
+### Step 3: Characterize Existing Framework Behavior
+
+Strengthen or add contract-focused tests where needed so extraction can be
+verified without relying on package placement.
+
+The accepted implementation scope and intended completion state are in the
+[Step 3 framework-characterization roadmap](step3.md).
+
+Preserve coverage of:
+
+- `Prepare` and `Run` behavior;
+- prompt, profile, execution-default, and request-override precedence;
+- presence-aware numeric overrides;
+- built-in profile fallback and custom-profile overlays;
+- strict YAML and JSON decoding;
+- prompt, profile, schema, and artifact source behavior;
+- structured-output requests and output validation;
+- validation failures versus validation runtime errors;
+- secret handling and redaction;
+- public error classification;
+- HTTP artifact restrictions and transport mappings.
+
+**Gate:** Current framework and adapter contracts are represented by passing
+tests sufficient to detect behavioral regressions during the split.
+
+**Gate status:** Complete as of 2026-07-27. The framework contract corpus,
+public `Engine` characterization, ownership audit, full test and vet suites,
+temporary executable build, and maintained offline examples passed. Step 4 is
+next.
+
+### Step 4: Make Scriptorium Adapters Consume The Public Facade
+
+Within the current repository, refactor the CLI and HTTP adapters to use the
+public framework facade rather than constructing or importing internal runner
+components directly.
+
+Add only the minimum public capabilities needed to support this boundary. These
+may include:
+
+- a small `Run`/`Prepare` consumer interface;
+- injectable artifact-reading behavior for Scriptorium's restricted HTTP
+  policy;
+- source options currently available only through internal constructors;
+- prepared-run formatting based on public types;
+- stable public error classification required by CLI and HTTP mappings.
+
+Do not broadly export internal repositories, domain types, or use-case
+implementations.
+
+**Gate:** The CLI and HTTP adapters use only the public framework API for
+framework behavior, and all tests and documented smoke commands pass.
+
+### Step 5: Create The Promptkit Repository
+
+Create the Promptkit repository and Go module as an explicit out-of-band
+operation.
+
+Establish:
+
+- repository access, ownership, and branch protections;
+- the module path selected by the ADR;
+- baseline development, architecture, documentation, and release policies;
+- CI for build, test, vet, and other agreed checks;
+- an initial package layout centered on a small public facade with internal
+  implementation packages;
+- a local development workflow for coordinated Promptkit and Scriptorium
+  changes, using a workspace or temporary uncommitted module replacement where
+  appropriate.
+
+Do not commit local filesystem `replace` directives to release branches.
+
+**Gate:** The Promptkit repository exists, is accessible to maintainers, has
+working CI and policy scaffolding, and can receive the extracted framework.
+Do not begin cross-repository extraction until this out-of-band work is
+confirmed complete.
+
+### Step 6: Extract And Stabilize Promptkit
+
+Move the application-neutral framework and built-in profile assets into
+Promptkit. Preserve implementation packages as internal where practical.
+
+The initial public API should remain focused on the established engine workflow
+and the source and client extension points required by real consumers. Avoid
+combining the extraction with unrelated API redesign.
+
+Move or recreate the relevant:
+
+- framework implementation;
+- public package tests and framework contract tests;
+- built-in profile assets and registry tests;
+- Go consumer examples;
+- framework, consumer, configuration-format, and integration documentation.
+
+Verify that Promptkit can be built, tested, and consumed independently of the
+Scriptorium repository.
+
+**Gate:** Promptkit independently provides the agreed framework contract,
+passes its CI checks, and has a tagged version that Scriptorium and other
+consumers can import.
+
+### Step 7: Slim Scriptorium And Adopt Promptkit
+
+Update Scriptorium to import the tagged Promptkit module and remove the
+framework implementation and public Go facade that Promptkit replaces.
+
+Retain only Scriptorium-owned executable and transport behavior. In particular:
+
+- wire CLI and HTTP requests through Promptkit's public API;
+- keep application config and transport defaults in Scriptorium;
+- keep restricted HTTP artifact policy in Scriptorium while injecting it
+  through Promptkit's supported boundary;
+- remove obsolete framework packages, tests, and documentation;
+- update Scriptorium examples and docs to describe the CLI and HTTP application;
+- direct Go framework consumers to Promptkit without providing compatibility
+  aliases or forwarding APIs.
+
+**Gate:** Scriptorium builds and passes all tests using a tagged Promptkit
+dependency, contains no duplicate framework implementation, and its current
+documentation describes only the slimmed application.
+
+### Step 8: Migrate Downstream Consumers To Promptkit
+
+Inventory downstream Go consumers and migrate each from the Scriptorium package
+to Promptkit. This work may occur in external repositories and must be tracked
+explicitly.
+
+For each consumer:
+
+- update module imports and dependencies;
+- adapt to any intentionally changed public API;
+- run its tests and relevant integration or smoke checks;
+- confirm configuration, source, validation, and error behavior;
+- release or deploy the migrated consumer through its normal process.
+
+Consumers that cannot migrate immediately may remain pinned to the last
+framework-bearing Scriptorium tag. No compatibility work is required in the new
+Scriptorium project for those consumers.
+
+**Gate:** All in-scope downstream consumers are either migrated and verified or
+explicitly recorded as remaining on the previous Scriptorium version with an
+owner and follow-up plan. Do not declare the ecosystem migration complete until
+the required out-of-band consumer changes are confirmed.
+
+### Step 9: Complete Release And Documentation Cutover
+
+Complete the coordinated project transition:
+
+- publish Promptkit before dependent Scriptorium releases;
+- release the breaking Scriptorium version against the tagged Promptkit
+  dependency;
+- publish migration guidance that maps the former Scriptorium Go API to
+  Promptkit;
+- update cross-project links, examples, package documentation, and release
+  notes;
+- verify that no release artifact depends on local workspaces or replacements;
+- archive completed roadmap material according to the documentation policy in
+  effect at that time.
+
+**Gate:** Promptkit and Scriptorium are independently releasable, their
+documentation has distinct and accurate ownership, and the migration status of
+all identified downstream consumers is recorded.
+
+## Cross-Cutting Constraints
+
+- Preserve the invariant that execution orchestration remains narrow and
+  application-neutral.
+- Keep adapter-specific decisions out of Promptkit.
+- Keep Scriptorium dependent only on Promptkit's supported public API.
+- Preserve strict external decoding, error classification, validation
+  semantics, and secret redaction throughout the migration.
+- Keep each repository buildable and testable at merged phase boundaries.
+- Coordinate cross-repository changes through tagged dependencies and explicit
+  gates rather than assuming atomic commits across repositories.
+- Document only implemented behavior outside roadmap files.
+
+## Completion Criteria
+
+The migration is complete when:
+
+- Promptkit is the independent owner of the reusable framework and built-in
+  profiles;
+- Scriptorium is a slim CLI and HTTP consumer of Promptkit;
+- Scriptorium no longer exposes or maintains the former public Go framework;
+- all required downstream migrations and external repository work have been
+  completed or explicitly dispositioned;
+- both repositories build, test, document, version, and release independently.

docs/roadmap/step3.md

@@ -0,0 +1,178 @@
+# Migration Step 3: Framework Characterization
+
+## Status
+
+Completed on 2026-07-27. The target state below is characterized by the
+passing baseline; its ordered implementation record is in
+[implementation.md](implementation.md).
+
+## Purpose
+
+Characterize the framework and adapter contracts that must survive the
+Promptkit extraction. The work should make those contracts portable across the
+future repository boundary without duplicating behavior at every test layer or
+changing production APIs ahead of Migration Step 4.
+
+The [testing policy](../policy/testing.md) governs test value, ownership, and
+sufficiency. The [accepted split decision](../adr/0002-split-promptkit-from-scriptorium.md)
+governs which project will own each behavior.
+
+## Baseline Findings
+
+The current suite already provides broad coverage at the public facade,
+framework package, CLI, HTTP, source, validation, and model-client boundaries.
+Step 3 is therefore a portability and risk-closing exercise rather than a
+general coverage expansion.
+
+The principal extraction risk is fixture ownership:
+
+- public `Engine` tests rely extensively on executable assets under
+  `examples/`;
+- the assembled runner integration test also reads those assets; and
+- the accepted split leaves executable examples in Scriptorium while moving
+  the public facade and framework tests to Promptkit.
+
+Those dependencies would either prevent the tests from moving or create
+unwanted cross-repository fixture coupling. A smaller set of public-boundary
+gaps also remains around complete setting precedence, cancellation, malformed
+injected-client behavior, and reserved provider parameters.
+
+## Target State
+
+The completed baseline provides:
+
+- tests destined for Promptkit use only Promptkit-destined testdata or
+  fixtures generated within the test;
+- Scriptorium's maintained executable examples remain independently validated
+  by Scriptorium-owned checks;
+- each behavior named in the main migration roadmap has one clear test owner;
+- representative public `Engine` tests protect assembled framework behavior;
+- focused package tests continue to own strict parsing, source mechanics,
+  validation rules, and provider mapping without higher-level duplication;
+- Scriptorium adapter tests continue to protect CLI, HTTP, containment, limits,
+  and transport mappings; and
+- no production API or package boundary has changed as part of Step 3.
+
+## Policy Choices
+
+### Framework-Owned Testdata
+
+Promptkit-destined tests will use a compact framework contract corpus under root
+`testdata/` or a purpose-built fixture created inside the test. The shared
+corpus will contain only the assets needed to express durable framework
+behavior:
+
+- one ordinary prompt definition;
+- one JSON Schema structured-output prompt definition;
+- representative execution profiles;
+- one JSON Schema document; and
+- small file-artifact inputs.
+
+The corpus should exercise directory-backed loading and relative prompt content
+where those behaviors matter. Use small inline `fstest.MapFS` or temporary
+fixtures for cases that do not benefit from shared files.
+
+The complete executable example tree will not be copied. Testdata will remain
+minimal, synthetic, secret-free, and distinct from user-facing examples.
+
+### Test Boundaries And Consolidation
+
+Representative assembled behavior belongs at the public `Engine` boundary.
+Focused parsing, source, validation, provider, and adapter mechanics remain
+with their package-level owners. Existing tests should be consolidated when a
+new public contract test would otherwise duplicate the same risk.
+
+The assembled runner integration behavior will be protected through the public
+facade rather than through a second test tied to internal domain and runner
+types. Promptkit-destined tests will not read Scriptorium-owned `examples/`
+assets. Scriptorium adapter and maintained-example checks may continue to do so
+where the example itself is the contract under test.
+
+### Required Public Characterization
+
+The public suite will characterize:
+
+- an assembled directory-backed `Engine.Run` workflow with file inputs,
+  structured output, schema validation, hashes, usage, and timing;
+- framework-default, profile, and request execution-setting precedence;
+- explicit numeric-zero propagation and target-presence metadata;
+- caller-context cancellation at the outbound generation boundary and its
+  public error classification;
+- nil responses from injected model clients; and
+- reserved provider parameters failing before a provider call.
+
+These tests will use deterministic synchronization, real local collaborators
+where inexpensive, and fakes only at the model-provider boundary.
+
+## Target Test Ownership
+
+| Test category | Future disposition |
+| --- | --- |
+| Public `Engine`, facade, model-client extension, and public error contracts | Move to Promptkit. |
+| Framework domain, runner, prompt, profile, built-in registry, general artifact, validation, and LLM package tests | Move with their Promptkit-owned implementation. |
+| CLI parsing, application configuration, prepared-run formatting, and process behavior | Remain in Scriptorium. |
+| HTTP DTOs, strict JSON, limits, response mapping, and rooted artifact containment | Remain in Scriptorium. |
+| Tests that construct internal runners or classify internal framework sentinels from Scriptorium adapters | Preserve their observable assertions, then rewrite against the public Promptkit boundary during Step 4. |
+| Maintained Go consumer example | Move to Promptkit. |
+| Maintained executable configuration, render, HTTP, and fixture examples | Remain in Scriptorium. |
+
+Existing focused tests remain the owners of:
+
+- strict prompt, profile, and application YAML decoding;
+- strict HTTP JSON decoding;
+- prompt, profile, schema, and artifact source mechanics;
+- built-in profile validation and overlay fallback;
+- structured-output encoding and schema validation;
+- validation content failures versus operational failures;
+- credential handling and redaction;
+- OpenAI-compatible wire behavior; and
+- HTTP artifact restrictions and transport mappings.
+
+Coverage will be added only if the ownership audit identifies a consequential
+behavior with no credible existing owner.
+
+## Required Validation Outcome
+
+The characterized baseline must pass the complete Go test and vet suites, a
+temporary-output executable build, repeated public contract tests, both
+maintained application configurations, maintained render and Go consumer
+examples, the maintained HTTP request-example check, documentation-link
+validation, and whitespace validation. All checks must remain offline and
+independent of real credentials.
+
+## Out Of Scope
+
+Step 3 does not:
+
+- add artifact-reader, repository, validator, or other production extension
+  APIs;
+- refactor CLI or HTTP adapters to consume the public facade;
+- move restricted HTTP artifact behavior out of its current package;
+- create the Promptkit repository or change the Go module path;
+- move implementation packages between repositories;
+- create compatibility aliases or forwarding APIs;
+- redesign the public facade; or
+- add tests solely to increase a coverage percentage.
+
+Those changes belong to later migration steps.
+
+## Completion Criteria
+
+Step 3 is complete when:
+
+- Promptkit-destined tests have no dependency on Scriptorium-owned executable
+  examples;
+- the targeted public contract gaps are covered with deterministic tests;
+- the full roadmap behavior list has a clear, non-duplicative test owner;
+- tests that require Step 4 rewrites are explicitly identified;
+- no production behavior or public API changed;
+- all required validation passes; and
+- the main migration roadmap records the Step 3 gate as complete.
+
+Migration Step 4 must not begin until these criteria are satisfied.
+
+## Lifecycle
+
+This completed implementation roadmap remains the concise characterization
+record for the migration. Repository history retains the detailed implementation
+record.

engine.go

@@ -0,0 +1,322 @@
+package scriptorium
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io/fs"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+
+	artifactadapter "gitea.maximumdirect.net/eric/scriptorium/internal/artifact"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/defaults"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/llm"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile/builtin"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/prompt"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/promptdef"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/usecase"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/validate"
+)
+
+// ErrInvalidConfig indicates invalid public engine configuration.
+var ErrInvalidConfig = errors.New("invalid engine configuration")
+
+var (
+	ErrInvalidRequest  = errors.New("invalid run request")
+	ErrPromptNotFound  = errors.New("prompt not found")
+	ErrProfileNotFound = errors.New("profile not found")
+	ErrPromptLoad      = errors.New("failed to load prompt definition")
+	ErrProfileLoad     = errors.New("failed to load execution profile")
+	ErrArtifactLoad    = errors.New("failed to load artifact")
+	ErrPromptRender    = errors.New("failed to render prompt")
+	ErrLLMGenerate     = errors.New("failed to generate output")
+	ErrValidation      = errors.New("failed to validate output")
+)
+
+// Engine prepares and runs Scriptorium prompt requests.
+type Engine struct {
+	runner *usecase.Runner
+}
+
+// Config configures a public Scriptorium engine.
+type Config struct {
+	PromptDir  string
+	ProfileDir string
+	SchemaDir  string
+	// Timeout is the transport-wide safety cap for the built-in LLM client
+	// when HTTPClient is absent or has a non-positive timeout.
+	Timeout time.Duration
+	// HTTPClient is cloned for the built-in LLM client. Its positive Timeout
+	// takes precedence over Config.Timeout as the transport-wide safety cap.
+	HTTPClient *http.Client
+}
+
+// Option customizes engine construction.
+type Option interface {
+	apply(*engineOptions) error
+}
+
+type optionFunc func(*engineOptions) error
+
+func (f optionFunc) apply(options *engineOptions) error {
+	return f(options)
+}
+
+type engineOptions struct {
+	llmClient       llm.Client
+	promptDefs      promptdef.Repository
+	profiles        profile.Repository
+	memoryProfiles  profile.Repository
+	validator       validate.Validator
+	promptSource    bool
+	profileSource   bool
+	memorySource    bool
+	validatorSource bool
+}
+
+// WithLLMClient injects a custom LLM client for execution.
+func WithLLMClient(client LLMClient) Option {
+	return optionFunc(func(options *engineOptions) error {
+		if client == nil {
+			return ErrInvalidConfig
+		}
+		options.llmClient = publicLLMClientAdapter{client: client}
+		return nil
+	})
+}
+
+// WithPromptFS loads prompt definitions from fsys under root.
+//
+// The source uses the same strict prompt YAML rules as configured prompt
+// directories, and prompt content_file paths resolve within this source.
+func WithPromptFS(fsys fs.FS, root string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		if fsys == nil {
+			return ErrInvalidConfig
+		}
+		if strings.TrimSpace(root) == "" {
+			return ErrInvalidConfig
+		}
+		options.promptDefs = promptdef.NewFSRepository(fsys, root)
+		options.promptSource = true
+		return nil
+	})
+}
+
+// WithPromptFile loads prompt definitions from the single prompt file at path.
+//
+// Relative prompt content_file paths resolve from the file's directory.
+func WithPromptFile(path string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		fsys, root, err := fileSource(path)
+		if err != nil {
+			return err
+		}
+		options.promptDefs = promptdef.NewFSRepository(fsys, root)
+		options.promptSource = true
+		return nil
+	})
+}
+
+// WithProfileFS loads execution profiles from fsys under root.
+//
+// Profiles from this source overlay built-in profiles. Profile YAML must use
+// api_key_env for environment-based credentials; raw API keys are rejected.
+func WithProfileFS(fsys fs.FS, root string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		if fsys == nil {
+			return ErrInvalidConfig
+		}
+		if strings.TrimSpace(root) == "" {
+			return ErrInvalidConfig
+		}
+		options.profiles = profile.NewFSRepository(fsys, root)
+		options.profileSource = true
+		return nil
+	})
+}
+
+// WithProfileFile loads execution profiles from the single profile file at path.
+//
+// The profile overlays built-in profiles. Profile YAML must use api_key_env for
+// environment-based credentials; raw API keys are rejected.
+func WithProfileFile(path string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		fsys, root, err := fileSource(path)
+		if err != nil {
+			return err
+		}
+		options.profiles = profile.NewFSRepository(fsys, root)
+		options.profileSource = true
+		return nil
+	})
+}
+
+// WithProfiles configures in-memory profiles that take precedence over
+// configured profile files and built-in profiles.
+func WithProfiles(profiles ...Profile) Option {
+	return optionFunc(func(options *engineOptions) error {
+		repo, err := newMemoryProfileRepository(profiles)
+		if err != nil {
+			return err
+		}
+		options.memoryProfiles = repo
+		options.memorySource = true
+		return nil
+	})
+}
+
+// WithSchemaFS loads JSON Schema documents from fsys under root.
+//
+// Prompt schema_path values resolve within this source when schema validation
+// or structured output is requested.
+func WithSchemaFS(fsys fs.FS, root string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		if fsys == nil {
+			return ErrInvalidConfig
+		}
+		if strings.TrimSpace(root) == "" {
+			return ErrInvalidConfig
+		}
+		options.validator = validate.NewFSValidator(fsys, root)
+		options.validatorSource = true
+		return nil
+	})
+}
+
+// WithSchemaFile loads JSON Schema documents from the single schema file at path.
+//
+// Prompt schema_path values refer to the file's base name.
+func WithSchemaFile(path string) Option {
+	return optionFunc(func(options *engineOptions) error {
+		fsys, root, err := fileSource(path)
+		if err != nil {
+			return err
+		}
+		options.validator = validate.NewFSValidator(fsys, root)
+		options.validatorSource = true
+		return nil
+	})
+}
+
+// NewEngine constructs an Engine using the same default internal components as
+// the CLI and HTTP adapters.
+func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
+	var options engineOptions
+	for _, opt := range opts {
+		if opt == nil {
+			continue
+		}
+		if err := opt.apply(&options); err != nil {
+			return nil, fmt.Errorf("%w: %v", ErrInvalidConfig, err)
+		}
+	}
+
+	promptDefs := options.promptDefs
+	if !options.promptSource {
+		if strings.TrimSpace(cfg.PromptDir) == "" {
+			return nil, fmt.Errorf("%w: prompt directory is required", ErrInvalidConfig)
+		}
+		promptDefs = promptdef.NewFilesystemRepository(cfg.PromptDir)
+	}
+
+	profiles := builtin.NewRepositoryWithDirectory(cfg.ProfileDir)
+	if options.profileSource {
+		profiles = builtin.NewRepositoryWithPrimary(options.profiles)
+	}
+	if options.memorySource {
+		profiles = profile.NewOverlayRepository(options.memoryProfiles, profiles)
+	}
+
+	validator := options.validator
+	if !options.validatorSource {
+		schemaDir := cfg.SchemaDir
+		if strings.TrimSpace(schemaDir) == "" {
+			schemaDir = defaults.SchemaDirDefault
+		}
+		validator = validate.NewStandardValidator(schemaDir)
+	}
+
+	llmClient := options.llmClient
+	if llmClient == nil {
+		var err error
+		llmClient, err = llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
+			Timeout:    cfg.Timeout,
+			HTTPClient: cfg.HTTPClient,
+		})
+		if err != nil {
+			return nil, fmt.Errorf("%w: %v", ErrInvalidConfig, err)
+		}
+	}
+
+	return &Engine{
+		runner: usecase.NewRunner(
+			promptDefs,
+			profiles,
+			artifactadapter.NewCompositeReader(),
+			prompt.NewGoRenderer(),
+			llmClient,
+			validator,
+		),
+	}, nil
+}
+
+func fileSource(name string) (fs.FS, string, error) {
+	cleanName := strings.TrimSpace(name)
+	if cleanName == "" {
+		return nil, "", ErrInvalidConfig
+	}
+	dir := filepath.Dir(cleanName)
+	base := filepath.Base(cleanName)
+	if base == "." || base == string(filepath.Separator) || strings.TrimSpace(base) == "" {
+		return nil, "", ErrInvalidConfig
+	}
+	info, err := os.Stat(cleanName)
+	if err != nil {
+		return nil, "", fmt.Errorf("%w: failed to access source file %q: %v", ErrInvalidConfig, cleanName, err)
+	}
+	if info.IsDir() {
+		return nil, "", fmt.Errorf("%w: source path %q must be a file", ErrInvalidConfig, cleanName)
+	}
+	return os.DirFS(dir), filepath.ToSlash(base), nil
+}
+
+// Prepare resolves a prompt request without calling an LLM.
+func (e *Engine) Prepare(ctx context.Context, req RunRequest) (*PreparedRun, error) {
+	if e == nil || e.runner == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+
+	domainReq, err := toDomainRunRequest(req)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
+	}
+
+	prepared, err := e.runner.Prepare(ctx, domainReq)
+	if err != nil {
+		return nil, mapPublicError(err)
+	}
+	return fromDomainPreparedRun(prepared), nil
+}
+
+// Run executes a prompt request and returns the generated artifact and metadata.
+func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
+	if e == nil || e.runner == nil {
+		return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
+	}
+
+	domainReq, err := toDomainRunRequest(req)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
+	}
+
+	result, err := e.runner.Run(ctx, domainReq)
+	if err != nil {
+		return nil, mapPublicError(err)
+	}
+	return fromDomainRunResult(result), nil
+}

errors.go

@@ -0,0 +1,79 @@
+package scriptorium
+
+import (
+	"errors"
+	"fmt"
+
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/promptdef"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/usecase"
+)
+
+func mapPublicError(err error) error {
+	if err == nil {
+		return nil
+	}
+	if hasPublicError(err) {
+		return err
+	}
+	publicErr := publicErrorFor(err)
+	if publicErr == nil {
+		return err
+	}
+	return fmt.Errorf("%w: %w", publicErr, err)
+}
+
+func hasPublicError(err error) bool {
+	for _, publicErr := range []error{
+		ErrInvalidConfig,
+		ErrInvalidRequest,
+		ErrPromptNotFound,
+		ErrProfileNotFound,
+		ErrPromptLoad,
+		ErrProfileLoad,
+		ErrArtifactLoad,
+		ErrPromptRender,
+		ErrLLMGenerate,
+		ErrValidation,
+	} {
+		if errors.Is(err, publicErr) {
+			return true
+		}
+	}
+	return false
+}
+
+func publicErrorFor(err error) error {
+	switch {
+	case errors.Is(err, promptdef.ErrPromptDefinitionNotFound):
+		return ErrPromptNotFound
+	case errors.Is(err, profile.ErrProfileNotFound):
+		return ErrProfileNotFound
+	case errors.Is(err, usecase.ErrPromptLoad):
+		return ErrPromptLoad
+	case errors.Is(err, usecase.ErrProfileLoad):
+		return ErrProfileLoad
+	case errors.Is(err, promptdef.ErrInvalidYAML), errors.Is(err, promptdef.ErrInvalidPromptDefinition):
+		return ErrPromptLoad
+	case isProfileLoadCause(err):
+		return ErrProfileLoad
+	case errors.Is(err, usecase.ErrArtifactLoad):
+		return ErrArtifactLoad
+	case errors.Is(err, usecase.ErrPromptRender):
+		return ErrPromptRender
+	case errors.Is(err, usecase.ErrLLMGenerate):
+		return ErrLLMGenerate
+	case errors.Is(err, usecase.ErrValidation):
+		return ErrValidation
+	case errors.Is(err, usecase.ErrInvalidRequest):
+		return ErrInvalidRequest
+	default:
+		return nil
+	}
+}
+
+func isProfileLoadCause(err error) bool {
+	return errors.Is(err, profile.ErrInvalidYAML) ||
+		errors.Is(err, profile.ErrInvalidProfile) ||
+		errors.Is(err, profile.ErrRawAPIKeyNotAllowed)
+}

examples/config.full.yml

@@ -0,0 +1,13 @@
+prompt_dir: ./examples/prompts
+profile_dir: ./examples/profiles
+schema_dir: ./examples/schemas
+
+server:
+  addr: 127.0.0.1:8080
+  artifact_root: .
+  max_request_bytes: 16777216
+  max_artifact_bytes: 16777216
+  max_response_bytes: 16777216
+
+defaults:
+  render_format: text

examples/config.yml

@@ -1,9 +1,10 @@
-prompt_dir: ./prompts
-profile_dir: ./profiles
-schema_dir: ./schemas
+prompt_dir: ./examples/prompts
+profile_dir: ./examples/profiles
+schema_dir: ./examples/schemas
 
 server:
   addr: :8080
+  artifact_root: .
 
 defaults:
   render_format: text

examples/go-library/prepare/main.go

@@ -0,0 +1,50 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"log"
+	"os"
+
+	"gitea.maximumdirect.net/eric/scriptorium"
+)
+
+func main() {
+	engine, err := scriptorium.NewEngine(scriptorium.Config{
+		PromptDir:  "./examples/prompts",
+		ProfileDir: "./examples/profiles",
+		SchemaDir:  "./examples/schemas",
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	prepared, err := engine.Prepare(context.Background(), scriptorium.RunRequest{
+		PromptID: "generic.markdown_summary",
+		Inputs: map[string]scriptorium.ArtifactRef{
+			"transcript": scriptorium.File("./examples/fixtures/transcript.md"),
+			"glossary":   scriptorium.File("./examples/fixtures/glossary.yml"),
+		},
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	summary := struct {
+		PromptID          string            `json:"prompt_id"`
+		SelectedProfileID string            `json:"selected_profile_id"`
+		Model             string            `json:"model"`
+		MessageCount      int               `json:"message_count"`
+		InputHashes       map[string]string `json:"input_hashes"`
+	}{
+		PromptID:          prepared.PromptID,
+		SelectedProfileID: prepared.SelectedProfileID,
+		Model:             prepared.EffectiveModelParams.Model,
+		MessageCount:      len(prepared.Messages),
+		InputHashes:       prepared.InputHashes,
+	}
+
+	if err := json.NewEncoder(os.Stdout).Encode(summary); err != nil {
+		log.Fatal(err)
+	}
+}

examples/http-run.json

@@ -0,0 +1,18 @@
+{
+  "prompt_id": "generic.markdown_summary",
+  "profile_id": "local-fast",
+  "inputs": {
+    "transcript": {
+      "type": "file",
+      "uri": "./examples/fixtures/transcript.md"
+    },
+    "glossary": {
+      "type": "file",
+      "uri": "./examples/fixtures/glossary.yml"
+    }
+  },
+  "vars": {
+    "session_date": "2026-05-04"
+  },
+  "include_raw_output": false
+}

examples/render-markdown-summary.sh

@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+
+cd "$repo_root"
+
+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 text

formatting.go

@@ -0,0 +1,51 @@
+package scriptorium
+
+import "fmt"
+
+// String returns a concise request summary without exposing direct API keys.
+func (r RunRequest) String() string {
+	return r.redactedString()
+}
+
+// GoString returns a concise request summary without exposing direct API keys.
+func (r RunRequest) GoString() string {
+	return r.redactedString()
+}
+
+func (r RunRequest) redactedString() string {
+	return fmt.Sprintf(
+		"scriptorium.RunRequest{PromptID:%q PromptVersion:%q ProfileID:%q APIKeySet:%t Inputs:%d Vars:%d ExecutionSet:%t ValidationSet:%t Metadata:%d}",
+		r.PromptID,
+		r.PromptVersion,
+		r.ProfileID,
+		r.APIKey != "",
+		len(r.Inputs),
+		len(r.Vars),
+		r.Execution != nil,
+		r.Validation != nil,
+		len(r.Metadata),
+	)
+}
+
+// String returns a concise request summary without exposing direct API keys or
+// rendered prompt content.
+func (r GenerateRequest) String() string {
+	return r.redactedString()
+}
+
+// GoString returns a concise request summary without exposing direct API keys or
+// rendered prompt content.
+func (r GenerateRequest) GoString() string {
+	return r.redactedString()
+}
+
+func (r GenerateRequest) redactedString() string {
+	return fmt.Sprintf(
+		"scriptorium.GenerateRequest{Messages:%d Model:%q APIKeySet:%t StructuredOutputSet:%t ExtraParams:%d}",
+		len(r.Prompt.Messages),
+		r.Target.Model,
+		r.APIKey != "",
+		r.StructuredOutput != nil,
+		len(r.Target.ExtraParams),
+	)
+}

internal/adapter/cli/run.go

@@ -19,7 +19,7 @@ import (
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
 	renderformat "gitea.maximumdirect.net/eric/scriptorium/internal/format"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/llm"
-	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile/builtin"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/prompt"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/promptdef"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/usecase"
@@ -34,7 +34,6 @@ const (
 
 const (
 	errPromptDirRequired = "prompt directory is required; provide --prompt-dir or config.yml prompt_dir"
-	errProfileDirRequired = "profile directory is required; provide --profile-dir or config.yml profile_dir"
 )
 
 type runConfig struct {
@@ -79,6 +78,22 @@ type serveConfig struct {
 	promptDir        string
 	profileDir       string
 	schemaDir        string
+	artifactRoot     string
+	maxRequestBytes  int64
+	maxArtifactBytes int64
+	maxResponseBytes int64
+}
+
+type commonCommandSettings struct {
+	promptDir           string
+	profileDir          string
+	schemaDir           string
+	serverAddr          string
+	artifactRoot        string
+	maxRequestBytes     int64
+	maxArtifactBytes    int64
+	maxResponseBytes    int64
+	defaultRenderFormat renderformat.PreparedRunOutputFormat
 }
 
 type listFlag []string
@@ -125,22 +140,13 @@ func runCommand(args []string, stdout, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	llmClient, err := llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
-		Timeout: defaults.LLMRequestTimeoutDefault,
-	})
+	llmClient, err := newOpenAIClient()
 	if err != nil {
 		fmt.Fprintf(stderr, "llm client error: %v\n", err)
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		llmClient,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
+	runner := newRunner(cfg.promptDir, cfg.profileDir, cfg.schemaDir, llmClient)
 
 	res, runErr := runner.Run(context.Background(), req)
 	if runErr != nil {
@@ -170,14 +176,7 @@ func renderCommand(args []string, stdout, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		nil,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
+	runner := newRunner(cfg.promptDir, cfg.profileDir, cfg.schemaDir, nil)
 
 	prepared, prepErr := runner.Prepare(context.Background(), req)
 	if prepErr != nil {
@@ -205,24 +204,24 @@ func serveCommand(args []string, stderr io.Writer) int {
 		return ExitRuntimeError
 	}
 
-	llmClient, err := llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{
-		Timeout: defaults.LLMRequestTimeoutDefault,
-	})
+	llmClient, err := newOpenAIClient()
 	if err != nil {
 		fmt.Fprintf(stderr, "llm client error: %v\n", err)
 		return ExitRuntimeError
 	}
 
-	runner := usecase.NewRunner(
-		promptdef.NewFilesystemRepository(cfg.promptDir),
-		profile.NewFilesystemRepository(cfg.profileDir),
-		artifactadapter.NewCompositeReader(),
-		prompt.NewGoRenderer(),
-		llmClient,
-		validate.NewStandardValidator(cfg.schemaDir),
-	)
+	artifactReader, err := artifactadapter.NewRestrictedCompositeReaderWithLimit(cfg.artifactRoot, cfg.maxArtifactBytes)
+	if err != nil {
+		fmt.Fprintf(stderr, "artifact root error: %v\n", err)
+		return ExitRuntimeError
+	}
 
-	h := httpadapter.NewHandler(runner)
+	runner := newRunnerWithArtifactReader(cfg.promptDir, cfg.profileDir, cfg.schemaDir, llmClient, artifactReader)
+
+	h := httpadapter.NewHandlerWithOptions(runner, httpadapter.HandlerOptions{
+		MaxRequestBytes:  cfg.maxRequestBytes,
+		MaxResponseBytes: cfg.maxResponseBytes,
+	})
 	srv := &http.Server{
 		Addr:              cfg.addr,
 		Handler:           h,
@@ -299,6 +298,10 @@ func parseServeArgs(args []string) (*serveConfig, error) {
 	fs.StringVar(&cfg.promptDir, "prompt-dir", "", "directory containing prompt definition YAML files")
 	fs.StringVar(&cfg.profileDir, "profile-dir", "", "directory containing execution profile YAML files")
 	fs.StringVar(&cfg.schemaDir, "schema-dir", "", "base directory for validation schemas")
+	fs.StringVar(&cfg.artifactRoot, "artifact-root", "", "base directory for HTTP file input artifacts")
+	fs.Int64Var(&cfg.maxRequestBytes, "max-request-bytes", 0, "maximum HTTP request body bytes; 0 disables the limit")
+	fs.Int64Var(&cfg.maxArtifactBytes, "max-artifact-bytes", 0, "maximum HTTP file artifact bytes; 0 disables the limit")
+	fs.Int64Var(&cfg.maxResponseBytes, "max-response-bytes", 0, "maximum HTTP response body bytes; 0 disables the limit")
 
 	if err := fs.Parse(args); err != nil {
 		return nil, err
@@ -307,31 +310,41 @@ 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),
 		ServerAddr:       cfg.addrIfSet(fs),
+		ArtifactRoot:     cfg.artifactRootIfSet(fs),
+		MaxRequestBytes:  cfg.maxRequestBytesIfSet(fs),
+		MaxArtifactBytes: cfg.maxArtifactBytesIfSet(fs),
+		MaxResponseBytes: cfg.maxResponseBytesIfSet(fs),
 	})
 	if err != nil {
 		return nil, err
 	}
 
-	cfg.promptDir = settings.PromptDir
-	cfg.profileDir = settings.ProfileDir
-	cfg.schemaDir = settings.SchemaDir
-	cfg.addr = settings.ServerAddr
+	cfg.promptDir = settings.promptDir
+	cfg.profileDir = settings.profileDir
+	cfg.schemaDir = settings.schemaDir
+	cfg.addr = settings.serverAddr
+	cfg.artifactRoot = settings.artifactRoot
+	cfg.maxRequestBytes = settings.maxRequestBytes
+	cfg.maxArtifactBytes = settings.maxArtifactBytes
+	cfg.maxResponseBytes = settings.maxResponseBytes
 
-	if strings.TrimSpace(cfg.promptDir) == "" {
-		return nil, errors.New(errPromptDirRequired)
-	}
-	if strings.TrimSpace(cfg.profileDir) == "" {
-		return nil, errors.New(errProfileDirRequired)
+	if err := validateRequiredLibraryDirs(cfg.promptDir); err != nil {
+		return nil, err
 	}
 
 	cfg.promptDir = filepath.Clean(cfg.promptDir)
+	if strings.TrimSpace(cfg.profileDir) != "" {
 		cfg.profileDir = filepath.Clean(cfg.profileDir)
+	}
 	cfg.schemaDir = filepath.Clean(cfg.schemaDir)
+	if strings.TrimSpace(cfg.artifactRoot) != "" {
+		cfg.artifactRoot = filepath.Clean(cfg.artifactRoot)
+	}
 	return cfg, nil
 }
 
@@ -359,7 +372,7 @@ 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 +381,13 @@ func finalizeExecutionRequestConfig(fs *flag.FlagSet, cfg *runConfig) error {
 		return err
 	}
 
-	cfg.promptDir = settings.PromptDir
-	cfg.profileDir = settings.ProfileDir
-	cfg.schemaDir = settings.SchemaDir
-	cfg.defaultRenderFormat = settings.DefaultRenderFormat
+	cfg.promptDir = settings.promptDir
+	cfg.profileDir = settings.profileDir
+	cfg.schemaDir = settings.schemaDir
+	cfg.defaultRenderFormat = settings.defaultRenderFormat
 
-	if strings.TrimSpace(cfg.promptDir) == "" {
-		return errors.New(errPromptDirRequired)
-	}
-	if strings.TrimSpace(cfg.profileDir) == "" {
-		return errors.New(errProfileDirRequired)
+	if err := validateRequiredLibraryDirs(cfg.promptDir); err != nil {
+		return err
 	}
 	if strings.TrimSpace(cfg.promptID) == "" {
 		return errors.New("--prompt is required")
@@ -386,7 +396,9 @@ func finalizeExecutionRequestConfig(fs *flag.FlagSet, cfg *runConfig) error {
 		return errors.New("at least one --input is required")
 	}
 	cfg.promptDir = filepath.Clean(cfg.promptDir)
+	if strings.TrimSpace(cfg.profileDir) != "" {
 		cfg.profileDir = filepath.Clean(cfg.profileDir)
+	}
 	if cfg.outputPath != "" {
 		cfg.outputPath = filepath.Clean(cfg.outputPath)
 	}
@@ -449,6 +461,34 @@ func (c *serveConfig) addrIfSet(fs *flag.FlagSet) string {
 	return ""
 }
 
+func (c *serveConfig) artifactRootIfSet(fs *flag.FlagSet) string {
+	if flagWasSet(fs, "artifact-root") {
+		return c.artifactRoot
+	}
+	return ""
+}
+
+func (c *serveConfig) maxRequestBytesIfSet(fs *flag.FlagSet) *int64 {
+	if flagWasSet(fs, "max-request-bytes") {
+		return &c.maxRequestBytes
+	}
+	return nil
+}
+
+func (c *serveConfig) maxArtifactBytesIfSet(fs *flag.FlagSet) *int64 {
+	if flagWasSet(fs, "max-artifact-bytes") {
+		return &c.maxArtifactBytes
+	}
+	return nil
+}
+
+func (c *serveConfig) maxResponseBytesIfSet(fs *flag.FlagSet) *int64 {
+	if flagWasSet(fs, "max-response-bytes") {
+		return &c.maxResponseBytes
+	}
+	return nil
+}
+
 func registerConfigPathFlag(fs *flag.FlagSet, target *string) {
 	fs.StringVar(
 		target,
@@ -476,6 +516,55 @@ 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,
+		artifactRoot:        settings.ArtifactRoot,
+		maxRequestBytes:     settings.MaxRequestBytes,
+		maxArtifactBytes:    settings.MaxArtifactBytes,
+		maxResponseBytes:    settings.MaxResponseBytes,
+		defaultRenderFormat: settings.DefaultRenderFormat,
+	}, nil
+}
+
+func validateRequiredLibraryDirs(promptDir string) error {
+	if strings.TrimSpace(promptDir) == "" {
+		return errors.New(errPromptDirRequired)
+	}
+	return nil
+}
+
+func newRunner(promptDir, profileDir, schemaDir string, llmClient llm.Client) *usecase.Runner {
+	return newRunnerWithArtifactReader(promptDir, profileDir, schemaDir, llmClient, artifactadapter.NewCompositeReader())
+}
+
+func newRunnerWithArtifactReader(promptDir, profileDir, schemaDir string, llmClient llm.Client, artifactReader artifactadapter.Reader) *usecase.Runner {
+	if artifactReader == nil {
+		artifactReader = artifactadapter.NewCompositeReader()
+	}
+	return usecase.NewRunner(
+		promptdef.NewFilesystemRepository(promptDir),
+		builtin.NewRepositoryWithDirectory(profileDir),
+		artifactReader,
+		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 {
@@ -495,18 +584,25 @@ func buildRunRequestFromConfig(cfg *runConfig) (domain.RunRequest, error) {
 		inputs[name] = domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: path}
 	}
 
-	var modelOverride *domain.ExecutionTarget
+	var modelOverride *domain.ExecutionTargetOverride
 	if cfg.llmBaseURLSet || cfg.modelSet || cfg.temperatureSet || cfg.maxTokensSet || cfg.topPSet || cfg.apiKeyEnvSet || cfg.timeoutSet {
-		modelOverride = &domain.ExecutionTarget{
+		modelOverride = &domain.ExecutionTargetOverride{
 			Endpoint:  cfg.llmBaseURL,
 			Model:     cfg.model,
-			Temperature: cfg.temperature,
-			MaxTokens:   cfg.maxTokens,
-			TopP:        cfg.topP,
 			APIKeyEnv: cfg.apiKeyEnv,
 		}
+		if cfg.temperatureSet {
+			modelOverride.Temperature = &cfg.temperature
+		}
+		if cfg.maxTokensSet {
+			modelOverride.MaxTokens = &cfg.maxTokens
+		}
+		if cfg.topPSet {
+			modelOverride.TopP = &cfg.topP
+		}
 		if cfg.timeoutSet {
-			modelOverride.TimeoutSeconds = int(cfg.timeout.Seconds())
+			timeoutSeconds := int(cfg.timeout.Seconds())
+			modelOverride.TimeoutSeconds = &timeoutSeconds
 		}
 	}
 
@@ -588,7 +684,7 @@ func printSummary(stderr io.Writer, res *domain.RunResult) {
 	if res == nil {
 		return
 	}
-	fmt.Fprintf(stderr, "prompt=%s@%s selected_profile=%s model=%s validation=%s mode=%s validation_errors=%d prompt_hash=%s inputs=%d usage=%d/%d/%d\n",
+	fmt.Fprintf(stderr, "prompt=%s@%s selected_profile=%s model=%s validation=%s mode=%s validation_errors=%d prompt_hash=%s inputs=%d usage=%d/%d/%d",
 		res.PromptID,
 		res.PromptVersion,
 		res.SelectedProfileID,
@@ -602,11 +698,15 @@ func printSummary(stderr io.Writer, res *domain.RunResult) {
 		res.Usage.CompletionTokens,
 		res.Usage.TotalTokens,
 	)
+	if res.Usage.CachedTokens != 0 || res.Usage.CacheWriteTokens != 0 {
+		fmt.Fprintf(stderr, " cached_tokens=%d cache_write_tokens=%d", res.Usage.CachedTokens, res.Usage.CacheWriteTokens)
+	}
+	fmt.Fprintln(stderr)
 }
 
 func printUsage(w io.Writer) {
 	fmt.Fprintln(w, "usage: scriptorium <run|render|serve> ...")
 	fmt.Fprintln(w, "  run:   scriptorium run [--config PATH] [--prompt-dir DIR] [--profile-dir DIR] --prompt ID --input name=path [--input ...] [--profile ID] [--llm-base-url URL] [--model NAME] [--api-key-env ENV] [--temperature N] [--max-tokens N] [--top-p N] [--var k=v] [--out path] [--timeout 10m]")
 	fmt.Fprintln(w, "  render: scriptorium render [--config PATH] [--prompt-dir DIR] [--profile-dir DIR] --prompt ID --input name=path [--input ...] [--profile ID] [--llm-base-url URL] [--model NAME] [--api-key-env ENV] [--temperature N] [--max-tokens N] [--top-p N] [--var k=v] [--format text|json] [--out path] [--timeout 10m]")
-	fmt.Fprintf(w, "  serve: scriptorium serve [--config PATH] [--addr %s] [--prompt-dir DIR] [--profile-dir DIR] [--schema-dir DIR]\n", defaults.HTTPAddrDefault)
+	fmt.Fprintf(w, "  serve: scriptorium serve [--config PATH] [--addr %s] [--prompt-dir DIR] [--profile-dir DIR] [--schema-dir DIR] [--artifact-root DIR] [--max-request-bytes N] [--max-artifact-bytes N] [--max-response-bytes N]\n", defaults.HTTPAddrDefault)
 }

internal/adapter/http/dto.go

@@ -23,13 +23,14 @@ type inputRefDTO struct {
 type modelOverrideRequestDTO struct {
 	Endpoint        string         `json:"endpoint,omitempty"`
 	Model           string         `json:"model,omitempty"`
-	Temperature     float64           `json:"temperature,omitempty"`
-	MaxTokens       int               `json:"max_tokens,omitempty"`
-	TopP            float64           `json:"top_p,omitempty"`
-	TimeoutSeconds  int               `json:"timeout_seconds,omitempty"`
+	Temperature     *float64       `json:"temperature,omitempty"`
+	MaxTokens       *int           `json:"max_tokens,omitempty"`
+	TopP            *float64       `json:"top_p,omitempty"`
+	TimeoutSeconds  *int           `json:"timeout_seconds,omitempty"`
+	ServiceTier     string         `json:"service_tier,omitempty"`
 	ReasoningEffort string         `json:"reasoning_effort,omitempty"`
 	APIKeyEnv       string         `json:"api_key_env,omitempty"`
-	ExtraParams     map[string]string `json:"extra_params,omitempty"`
+	ExtraParams     map[string]any `json:"extra_params,omitempty"`
 }
 
 type runResponseDTO struct {
@@ -75,15 +76,18 @@ type modelParamsDTO struct {
 	MaxTokens       int            `json:"max_tokens"`
 	TopP            float64        `json:"top_p"`
 	TimeoutSeconds  int            `json:"timeout_seconds"`
+	ServiceTier     string         `json:"service_tier,omitempty"`
 	ReasoningEffort string         `json:"reasoning_effort,omitempty"`
 	APIKeyEnv       string         `json:"api_key_env,omitempty"`
-	ExtraParams     map[string]string `json:"extra_params,omitempty"`
+	ExtraParams     map[string]any `json:"extra_params,omitempty"`
 }
 
 type tokenUsageDTO struct {
 	PromptTokens     int `json:"prompt_tokens"`
 	CompletionTokens int `json:"completion_tokens"`
 	TotalTokens      int `json:"total_tokens"`
+	CachedTokens     int `json:"cached_tokens"`
+	CacheWriteTokens int `json:"cache_write_tokens"`
 }
 
 type validationDTO struct {

internal/adapter/http/handler.go

@@ -4,9 +4,12 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"io"
 	"net/http"
 	"strings"
 
+	"gitea.maximumdirect.net/eric/scriptorium/internal/artifact"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/defaults"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/promptdef"
@@ -19,10 +22,23 @@ type Runner interface {
 
 type Handler struct {
 	runner  Runner
+	options HandlerOptions
+}
+
+type HandlerOptions struct {
+	MaxRequestBytes  int64
+	MaxResponseBytes int64
 }
 
 func NewHandler(runner Runner) *Handler {
-	return &Handler{runner: runner}
+	return NewHandlerWithOptions(runner, HandlerOptions{
+		MaxRequestBytes:  defaults.HTTPMaxRequestBytesDefault,
+		MaxResponseBytes: defaults.HTTPMaxResponseBytesDefault,
+	})
+}
+
+func NewHandlerWithOptions(runner Runner, options HandlerOptions) *Handler {
+	return &Handler{runner: runner, options: options}
 }
 
 func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
@@ -36,9 +52,26 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	}
 
 	var req runRequestDTO
-	dec := json.NewDecoder(r.Body)
+	body := r.Body
+	if h.options.MaxRequestBytes > 0 {
+		body = http.MaxBytesReader(w, r.Body, h.options.MaxRequestBytes)
+	}
+	dec := json.NewDecoder(body)
 	dec.DisallowUnknownFields()
 	if err := dec.Decode(&req); err != nil {
+		if isRequestTooLarge(err) {
+			writeError(w, http.StatusRequestEntityTooLarge, "request_too_large", "request body is too large")
+			return
+		}
+		writeError(w, http.StatusBadRequest, "invalid_json", "invalid JSON request body")
+		return
+	}
+	var trailing any
+	if err := dec.Decode(&trailing); err != io.EOF {
+		if isRequestTooLarge(err) {
+			writeError(w, http.StatusRequestEntityTooLarge, "request_too_large", "request body is too large")
+			return
+		}
 		writeError(w, http.StatusBadRequest, "invalid_json", "invalid JSON request body")
 		return
 	}
@@ -61,19 +94,9 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 
-	var model *domain.ExecutionTarget
+	var model *domain.ExecutionTargetOverride
 	if req.Model != nil {
-		model = &domain.ExecutionTarget{
-			Endpoint:        req.Model.Endpoint,
-			Model:           req.Model.Model,
-			Temperature:     req.Model.Temperature,
-			MaxTokens:       req.Model.MaxTokens,
-			TopP:            req.Model.TopP,
-			TimeoutSeconds:  req.Model.TimeoutSeconds,
-			ReasoningEffort: req.Model.ReasoningEffort,
-			APIKeyEnv:       req.Model.APIKeyEnv,
-			ExtraParams:     req.Model.ExtraParams,
-		}
+		model = executionTargetOverrideFromModelOverrideDTO(req.Model)
 	}
 
 	res, err := h.runner.Run(r.Context(), domain.RunRequest{
@@ -109,22 +132,14 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 			SelectedProfileID:  res.SelectedProfileID,
 			ModelName:          res.ModelName,
 			Endpoint:           res.Endpoint,
-			ModelParams: modelParamsDTO{
-				Endpoint:        res.EffectiveModelParams.Endpoint,
-				Model:           res.EffectiveModelParams.Model,
-				Temperature:     res.EffectiveModelParams.Temperature,
-				MaxTokens:       res.EffectiveModelParams.MaxTokens,
-				TopP:            res.EffectiveModelParams.TopP,
-				TimeoutSeconds:  res.EffectiveModelParams.TimeoutSeconds,
-				ReasoningEffort: res.EffectiveModelParams.ReasoningEffort,
-				APIKeyEnv:       res.EffectiveModelParams.APIKeyEnv,
-				ExtraParams:     res.EffectiveModelParams.ExtraParams,
-			},
+			ModelParams:        modelParamsDTOFromExecutionTarget(res.EffectiveModelParams),
 			InputHashes:        res.InputHashes,
 			Usage: tokenUsageDTO{
 				PromptTokens:     res.Usage.PromptTokens,
 				CompletionTokens: res.Usage.CompletionTokens,
 				TotalTokens:      res.Usage.TotalTokens,
+				CachedTokens:     res.Usage.CachedTokens,
+				CacheWriteTokens: res.Usage.CacheWriteTokens,
 			},
 			StartTime:          res.StartTime,
 			EndTime:            res.EndTime,
@@ -138,7 +153,40 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		raw := res.RawOutput
 		resp.RawModelOutput = &raw
 	}
-	writeJSON(w, http.StatusOK, resp)
+	writeLimitedJSON(w, http.StatusOK, resp, h.options.MaxResponseBytes)
+}
+
+func executionTargetOverrideFromModelOverrideDTO(dto *modelOverrideRequestDTO) *domain.ExecutionTargetOverride {
+	if dto == nil {
+		return nil
+	}
+	return &domain.ExecutionTargetOverride{
+		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 {
@@ -160,16 +208,22 @@ func mapRunError(err error) (int, string, string) {
 		return http.StatusNotFound, "profile_not_found", "execution profile not found"
 	case errors.Is(err, promptdef.ErrInvalidYAML), errors.Is(err, promptdef.ErrInvalidPromptDefinition):
 		return http.StatusBadRequest, "prompt_load_failed", "failed to load prompt definition"
-	case errors.Is(err, profile.ErrInvalidYAML), errors.Is(err, profile.ErrInvalidProfile):
+	case errors.Is(err, profile.ErrInvalidYAML), errors.Is(err, profile.ErrInvalidProfile), errors.Is(err, profile.ErrRawAPIKeyNotAllowed):
 		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"
-	case errors.Is(err, usecase.ErrProfileLoad):
+	case errors.Is(err, usecase.ErrPromptLoad):
 		return http.StatusBadRequest, "prompt_load_failed", "failed to load prompt definition"
+	case errors.Is(err, usecase.ErrProfileLoad):
+		return http.StatusBadRequest, "profile_load_failed", "failed to load execution profile"
+	case errors.Is(err, artifact.ErrFileNotAllowed), errors.Is(err, artifact.ErrFileOutsideRoot):
+		return http.StatusBadRequest, "artifact_not_allowed", "file input artifact is not allowed"
+	case errors.Is(err, artifact.ErrFileTooLarge):
+		return http.StatusRequestEntityTooLarge, "artifact_too_large", "file input artifact is too large"
 	case errors.Is(err, usecase.ErrArtifactLoad):
 		return http.StatusBadRequest, "artifact_read_failed", "failed to read input artifact"
 	case errors.Is(err, usecase.ErrPromptRender):
@@ -184,9 +238,23 @@ func mapRunError(err error) (int, string, string) {
 }
 
 func writeJSON(w http.ResponseWriter, status int, v any) {
+	writeLimitedJSON(w, status, v, 0)
+}
+
+func writeLimitedJSON(w http.ResponseWriter, status int, v any, maxBytes int64) {
+	data, err := json.Marshal(v)
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, "internal_error", "internal server error")
+		return
+	}
+	data = append(data, '\n')
+	if maxBytes > 0 && int64(len(data)) > maxBytes {
+		writeError(w, http.StatusRequestEntityTooLarge, "response_too_large", "response body is too large")
+		return
+	}
 	w.Header().Set("Content-Type", "application/json")
 	w.WriteHeader(status)
-	_ = json.NewEncoder(w).Encode(v)
+	_, _ = w.Write(data)
 }
 
 func writeError(w http.ResponseWriter, status int, code, message string) {
@@ -197,3 +265,8 @@ func writeError(w http.ResponseWriter, status int, code, message string) {
 		},
 	})
 }
+
+func isRequestTooLarge(err error) bool {
+	var maxBytesErr *http.MaxBytesError
+	return errors.As(err, &maxBytesErr)
+}

internal/adapter/http/handler_test.go

@@ -4,15 +4,19 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"net/http"
 	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"reflect"
 	"strings"
 	"testing"
 	"time"
 
+	"gitea.maximumdirect.net/eric/scriptorium/internal/artifact"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/llm"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/promptdef"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/usecase"
@@ -32,6 +36,72 @@ func (f *fakeRunner) Run(ctx context.Context, req domain.RunRequest) (*domain.Ru
 	return f.result, nil
 }
 
+func TestMaintainedHTTPRunExampleMatchesRequestContract(t *testing.T) {
+	body, err := os.ReadFile(filepath.Join("..", "..", "..", "examples", "http-run.json"))
+	if err != nil {
+		t.Fatalf("read maintained HTTP request example: %v", err)
+	}
+
+	runner := &fakeRunner{result: &domain.RunResult{}}
+	h := NewHandler(runner)
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewReader(body))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Fatalf("expected maintained HTTP request example to be accepted, got %d: %s", w.Code, w.Body.String())
+	}
+
+	var invalidExample map[string]json.RawMessage
+	if err := json.Unmarshal(body, &invalidExample); err != nil {
+		t.Fatalf("decode maintained HTTP request example: %v", err)
+	}
+	invalidExample["unexpected"] = json.RawMessage(`true`)
+	invalidBody, err := json.Marshal(invalidExample)
+	if err != nil {
+		t.Fatalf("encode structurally invalid request example: %v", err)
+	}
+	invalidReq := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewReader(invalidBody))
+	invalidW := httptest.NewRecorder()
+	h.ServeHTTP(invalidW, invalidReq)
+	assertHTTPErrorCode(t, invalidW, http.StatusBadRequest, "invalid_json")
+}
+
+type handlerPromptRepo struct {
+	def *domain.PromptDefinition
+}
+
+func (r handlerPromptRepo) GetPromptDefinition(ctx context.Context, id string, version string) (*domain.PromptDefinition, error) {
+	return r.def, nil
+}
+
+type handlerProfileRepo struct {
+	profile *domain.ExecutionProfile
+}
+
+func (r handlerProfileRepo) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	return r.profile, nil
+}
+
+type handlerArtifactReader struct{}
+
+func (handlerArtifactReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error) {
+	return &domain.Artifact{Name: "input", Body: []byte("input"), Hash: "hash"}, nil
+}
+
+type handlerRenderer struct{}
+
+func (handlerRenderer) Render(ctx context.Context, definition *domain.PromptDefinition, inputs map[string]*domain.Artifact, vars map[string]string) (*domain.RenderedPrompt, error) {
+	return &domain.RenderedPrompt{Messages: []domain.RenderedMessage{{Role: "user", Content: "hi"}}}, nil
+}
+
+type handlerLLMClient struct{}
+
+func (handlerLLMClient) Generate(ctx context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
+	return &domain.GenerateResponse{Content: "ok"}, nil
+}
+
 func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
 	start := time.Now().UTC()
 	end := start.Add(2 * time.Second)
@@ -62,10 +132,17 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
 			MaxTokens:      42,
 			TopP:           0.9,
 			TimeoutSeconds: 120,
+			ServiceTier:    "priority",
 			APIKeyEnv:      envName,
 		},
 		InputHashes: map[string]string{"transcript": "h1"},
-		Usage:       domain.TokenUsage{PromptTokens: 1, CompletionTokens: 2, TotalTokens: 3},
+		Usage: domain.TokenUsage{
+			PromptTokens:     1,
+			CompletionTokens: 2,
+			TotalTokens:      3,
+			CachedTokens:     4,
+			CacheWriteTokens: 5,
+		},
 		StartTime: start,
 		EndTime:   end,
 		Duration:  2 * time.Second,
@@ -81,7 +158,7 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
     "transcript": {"type": "file", "uri": "./t.md"}
   },
   "vars": {"k": "v"},
-  "model": {"model": "gpt-x", "timeout_seconds": 120, "api_key_env": "SCRIPTORIUM_API_KEY"}
+  "model": {"model": "gpt-x", "timeout_seconds": 120, "service_tier": "flex", "api_key_env": "SCRIPTORIUM_API_KEY"}
 }`)
 	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewReader(body))
 	w := httptest.NewRecorder()
@@ -110,10 +187,20 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
 	if metadata["model_name"] != "m1" || metadata["endpoint"] != "http://llm/v1" {
 		t.Fatalf("unexpected model metadata: name=%#v endpoint=%#v", metadata["model_name"], metadata["endpoint"])
 	}
+	usage := metadata["usage"].(map[string]any)
+	if usage["prompt_tokens"] != float64(1) || usage["completion_tokens"] != float64(2) || usage["total_tokens"] != float64(3) {
+		t.Fatalf("unexpected base usage metadata: %#v", usage)
+	}
+	if usage["cached_tokens"] != float64(4) || usage["cache_write_tokens"] != float64(5) {
+		t.Fatalf("unexpected cache usage metadata: %#v", usage)
+	}
 	modelParams := metadata["model_params"].(map[string]any)
 	if modelParams["api_key_env"] != envName {
 		t.Fatalf("expected model_params.api_key_env=%q, got %#v", envName, modelParams["api_key_env"])
 	}
+	if modelParams["service_tier"] != "priority" {
+		t.Fatalf("expected model_params.service_tier=priority, got %#v", modelParams["service_tier"])
+	}
 	if strings.Contains(w.Body.String(), secret) {
 		t.Fatalf("response leaked raw API key value: %s", w.Body.String())
 	}
@@ -130,9 +217,111 @@ func TestHandlerPostRunsSuccessWithExplicitProfileID(t *testing.T) {
 	if r.last.Execution == nil || r.last.Execution.Model != "gpt-x" {
 		t.Fatalf("expected model override, got %#v", r.last.Execution)
 	}
-	if r.last.Execution.TimeoutSeconds != 120 {
+	if r.last.Execution.TimeoutSeconds == nil || *r.last.Execution.TimeoutSeconds != 120 {
 		t.Fatalf("expected timeout_seconds override 120, got %#v", r.last.Execution)
 	}
+	if r.last.Execution.ServiceTier != "flex" {
+		t.Fatalf("expected service_tier override flex, got %#v", r.last.Execution)
+	}
+}
+
+func TestHandlerInlineRefsWorkWithoutArtifactRoot(t *testing.T) {
+	h := newArtifactRootHandler(t, "")
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"inline","body":"inline body"}}
+	}`))
+	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())
+	}
+}
+
+func TestHandlerFileRefsWithoutArtifactRootAreRejected(t *testing.T) {
+	h := newArtifactRootHandler(t, "")
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"file","uri":"input.txt"}}
+	}`))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	assertHTTPErrorCode(t, w, http.StatusBadRequest, "artifact_not_allowed")
+}
+
+func TestHandlerFileRefsUnderArtifactRootWork(t *testing.T) {
+	root := t.TempDir()
+	if err := os.WriteFile(filepath.Join(root, "input.txt"), []byte("allowed"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	h := newArtifactRootHandler(t, root)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"file","uri":"input.txt"}}
+	}`))
+	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())
+	}
+}
+
+func TestHandlerFileRefsAboveArtifactLimitAreRejected(t *testing.T) {
+	root := t.TempDir()
+	if err := os.WriteFile(filepath.Join(root, "large.txt"), []byte("123456"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	h := newArtifactRootHandlerWithLimit(t, root, 5)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"file","uri":"large.txt"}}
+	}`))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	assertHTTPErrorCode(t, w, http.StatusRequestEntityTooLarge, "artifact_too_large")
+}
+
+func TestHandlerFileRefsOutsideArtifactRootAreRejected(t *testing.T) {
+	root := t.TempDir()
+	outside := t.TempDir()
+	if err := os.WriteFile(filepath.Join(outside, "secret.txt"), []byte("denied"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	h := newArtifactRootHandler(t, root)
+
+	tests := []struct {
+		name string
+		uri  string
+	}{
+		{name: "relative traversal", uri: filepath.Join("..", filepath.Base(outside), "secret.txt")},
+		{name: "absolute outside root", uri: filepath.Join(outside, "secret.txt")},
+	}
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			body := fmt.Sprintf(`{
+				"prompt_id":"p",
+				"inputs":{"x":{"type":"file","uri":%q}}
+			}`, tc.uri)
+			req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(body))
+			w := httptest.NewRecorder()
+
+			h.ServeHTTP(w, req)
+
+			assertHTTPErrorCode(t, w, http.StatusBadRequest, "artifact_not_allowed")
+		})
+	}
 }
 
 func TestHandlerPostRunsSuccessUsingPromptDefaultProfile(t *testing.T) {
@@ -164,6 +353,265 @@ func TestHandlerPostRunsSuccessUsingPromptDefaultProfile(t *testing.T) {
 	if metadata["selected_profile_id"] != "prompt-default" {
 		t.Fatalf("expected selected_profile_id from result, got %#v", metadata["selected_profile_id"])
 	}
+	usage := metadata["usage"].(map[string]any)
+	if usage["cached_tokens"] != float64(0) || usage["cache_write_tokens"] != float64(0) {
+		t.Fatalf("expected zero cache usage fields to be included, got %#v", usage)
+	}
+}
+
+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.ServiceTier != "flex" ||
+		got.ReasoningEffort != "medium" ||
+		got.APIKeyEnv != "SCRIPTORIUM_API_KEY" {
+		t.Fatalf("unexpected mapped execution target: %+v", got)
+	}
+	if got.Temperature == nil || *got.Temperature != 0.6 {
+		t.Fatalf("unexpected mapped temperature: %#v", got.Temperature)
+	}
+	if got.MaxTokens == nil || *got.MaxTokens != 250 {
+		t.Fatalf("unexpected mapped max_tokens: %#v", got.MaxTokens)
+	}
+	if got.TopP == nil || *got.TopP != 0.85 {
+		t.Fatalf("unexpected mapped top_p: %#v", got.TopP)
+	}
+	if got.TimeoutSeconds == nil || *got.TimeoutSeconds != 33 {
+		t.Fatalf("unexpected mapped timeout_seconds: %#v", got.TimeoutSeconds)
+	}
+	if !reflect.DeepEqual(got.ExtraParams, map[string]any{"provider_option": "on"}) {
+		t.Fatalf("unexpected mapped extra_params: %#v", got.ExtraParams)
+	}
+}
+
+func TestHandlerModelOverrideAcceptsJSONCompatibleExtraParams(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": {
+			"extra_params": {
+				"string_value": "enabled",
+				"number_value": 42,
+				"boolean_value": true,
+				"object_value": {"nested": "value", "count": 2},
+				"array_value": ["first", 3, false]
+			}
+		}
+	}`
+	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.Fatal("expected execution override in run request")
+	}
+	want := map[string]any{
+		"string_value":  "enabled",
+		"number_value":  float64(42),
+		"boolean_value": true,
+		"object_value":  map[string]any{"nested": "value", "count": float64(2)},
+		"array_value":   []any{"first", float64(3), false},
+	}
+	if !reflect.DeepEqual(r.last.Execution.ExtraParams, want) {
+		t.Fatalf("unexpected mapped extra_params:\ngot=%#v\nwant=%#v", r.last.Execution.ExtraParams, want)
+	}
+}
+
+func TestHandlerModelOverrideExplicitZeroTemperatureMapsAsPresent(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", Temperature: 0},
+	}}
+	h := NewHandler(r)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id": "prompt-1",
+		"inputs": {"transcript": {"type": "file", "uri": "./t.md"}},
+		"model": {"temperature": 0}
+	}`))
+	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 || r.last.Execution.Temperature == nil {
+		t.Fatalf("expected temperature override to be present, got %#v", r.last.Execution)
+	}
+	if *r.last.Execution.Temperature != 0 {
+		t.Fatalf("expected zero temperature override, got %v", *r.last.Execution.Temperature)
+	}
+}
+
+func TestHandlerModelOverrideOmittedTemperatureMapsAsAbsent(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", Temperature: 0.7},
+	}}
+	h := NewHandler(r)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id": "prompt-1",
+		"inputs": {"transcript": {"type": "file", "uri": "./t.md"}},
+		"model": {"model": "override-model"}
+	}`))
+	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.Fatal("expected model override")
+	}
+	if r.last.Execution.Temperature != nil {
+		t.Fatalf("expected omitted temperature to remain absent, got %#v", r.last.Execution.Temperature)
+	}
+
+	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["temperature"] != 0.7 {
+		t.Fatalf("expected effective profile/default temperature in response, got %#v", params["temperature"])
+	}
+}
+
+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]any{
+				"provider_option": "on",
+				"number_value":    42,
+				"object_value":    map[string]any{"nested": "value"},
+			},
+		},
+	}}
+	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"])
+	}
+	if extraParams["number_value"] != float64(42) {
+		t.Fatalf("unexpected extra_params.number_value: %#v", extraParams["number_value"])
+	}
+	objectValue, ok := extraParams["object_value"].(map[string]any)
+	if !ok || objectValue["nested"] != "value" {
+		t.Fatalf("unexpected extra_params.object_value: %#v", extraParams["object_value"])
+	}
 }
 
 func TestHandlerInvalidJSON(t *testing.T) {
@@ -178,6 +626,69 @@ func TestHandlerInvalidJSON(t *testing.T) {
 	}
 }
 
+func TestHandlerRejectsTrailingJSON(t *testing.T) {
+	h := NewHandler(&fakeRunner{})
+	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)
+
+	assertHTTPErrorCode(t, w, http.StatusBadRequest, "invalid_json")
+}
+
+func TestHandlerRequestTooLarge(t *testing.T) {
+	h := NewHandlerWithOptions(&fakeRunner{}, HandlerOptions{MaxRequestBytes: 12})
+	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)
+
+	assertHTTPErrorCode(t, w, http.StatusRequestEntityTooLarge, "request_too_large")
+}
+
+func TestHandlerMalformedJSONBelowLimitStillBadRequest(t *testing.T) {
+	h := NewHandlerWithOptions(&fakeRunner{}, HandlerOptions{MaxRequestBytes: 1024})
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString("{"))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	assertHTTPErrorCode(t, w, http.StatusBadRequest, "invalid_json")
+}
+
+func TestHandlerResponseTooLarge(t *testing.T) {
+	h := NewHandlerWithOptions(&fakeRunner{result: &domain.RunResult{
+		Artifact:             domain.Artifact{Body: []byte(strings.Repeat("x", 128))},
+		Validation:           domain.ValidationResult{Status: domain.ValidationPassed, Mode: domain.ValidationBasic, IsValid: true},
+		EffectiveModelParams: domain.ExecutionTarget{Endpoint: "http://llm/v1", Model: "m1"},
+	}}, HandlerOptions{MaxRequestBytes: 1024, MaxResponseBytes: 64})
+	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)
+
+	assertHTTPErrorCode(t, w, http.StatusRequestEntityTooLarge, "response_too_large")
+}
+
+func TestHandlerRawOutputDoesNotBypassResponseLimit(t *testing.T) {
+	h := NewHandlerWithOptions(&fakeRunner{result: &domain.RunResult{
+		Artifact:             domain.Artifact{Body: []byte("ok")},
+		RawOutput:            strings.Repeat("raw", 80),
+		Validation:           domain.ValidationResult{Status: domain.ValidationPassed, Mode: domain.ValidationBasic, IsValid: true},
+		EffectiveModelParams: domain.ExecutionTarget{Endpoint: "http://llm/v1", Model: "m1"},
+	}}, HandlerOptions{MaxRequestBytes: 1024, MaxResponseBytes: 128})
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"file","uri":"a"}},
+		"include_raw_output":true
+	}`))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	assertHTTPErrorCode(t, w, http.StatusRequestEntityTooLarge, "response_too_large")
+}
+
 func TestHandlerMissingPromptID(t *testing.T) {
 	h := NewHandler(&fakeRunner{})
 	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{"inputs":{"x":{"type":"file","uri":"a"}}}`))
@@ -198,6 +709,54 @@ func TestHandlerMissingPromptID(t *testing.T) {
 	}
 }
 
+func TestHandlerReservedExtraParamsThroughRunnerMapsToInvalidRequest(t *testing.T) {
+	llmClient, err := llm.NewOpenAICompatibleClient(llm.OpenAICompatibleConfig{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	runner := usecase.NewRunner(
+		handlerPromptRepo{def: &domain.PromptDefinition{
+			ID:             "p",
+			Version:        "1",
+			DefaultProfile: "exec",
+			Templates:      []domain.PromptMessageTemplate{{Role: "user", Content: "hi"}},
+			OutputFormat:   domain.FormatText,
+			Validation:     domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationNone},
+		}},
+		handlerProfileRepo{profile: &domain.ExecutionProfile{
+			ID:       "exec",
+			Endpoint: "http://example.invalid/v1",
+			Model:    "model",
+		}},
+		handlerArtifactReader{},
+		handlerRenderer{},
+		llmClient,
+		nil,
+	)
+	h := NewHandler(runner)
+
+	req := httptest.NewRequest(http.MethodPost, "/v1/runs", bytes.NewBufferString(`{
+		"prompt_id":"p",
+		"inputs":{"x":{"type":"file","uri":"a"}},
+		"model":{"extra_params":{"model":"collision"}}
+	}`))
+	w := httptest.NewRecorder()
+
+	h.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Fatalf("expected 400, 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)
+	}
+	errBody := resp["error"].(map[string]any)
+	if errBody["code"] != "invalid_request" {
+		t.Fatalf("expected invalid_request code, got %#v", errBody["code"])
+	}
+}
+
 func TestHandlerUsecaseErrorMapping(t *testing.T) {
 	tests := []struct {
 		name       string
@@ -207,12 +766,14 @@ func TestHandlerUsecaseErrorMapping(t *testing.T) {
 		message    string
 		avoidCause string
 	}{
-		{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: "prompt not found", err: wrap(usecase.ErrPromptLoad, promptdef.ErrPromptDefinitionNotFound), status: http.StatusNotFound, code: "prompt_not_found", message: "prompt definition not found"},
+		{name: "prompt load invalid", err: wrap(usecase.ErrPromptLoad, promptdef.ErrInvalidPromptDefinition), status: http.StatusBadRequest, code: "prompt_load_failed", message: "failed to load prompt definition"},
+		{name: "prompt load generic", err: wrap(usecase.ErrPromptLoad, fmt.Errorf("read failed")), status: http.StatusBadRequest, code: "prompt_load_failed", message: "failed to load prompt definition", avoidCause: "read failed"},
+		{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: "profile load generic", err: wrap(usecase.ErrProfileLoad, fmt.Errorf("read failed")), status: http.StatusBadRequest, code: "profile_load_failed", message: "failed to load execution profile", avoidCause: "read failed"},
+		{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"},
@@ -318,3 +879,54 @@ func TestHandlerValidationFailureStillSuccessAndRawOutputOptIn(t *testing.T) {
 func wrap(stage error, cause error) error {
 	return fmt.Errorf("%w: %w", stage, cause)
 }
+
+func newArtifactRootHandler(t *testing.T, root string) *Handler {
+	t.Helper()
+
+	return newArtifactRootHandlerWithLimit(t, root, 0)
+}
+
+func newArtifactRootHandlerWithLimit(t *testing.T, root string, maxArtifactBytes int64) *Handler {
+	t.Helper()
+
+	reader, err := artifact.NewRestrictedCompositeReaderWithLimit(root, maxArtifactBytes)
+	if err != nil {
+		t.Fatalf("expected restricted artifact reader: %v", err)
+	}
+	runner := usecase.NewRunner(
+		handlerPromptRepo{def: &domain.PromptDefinition{
+			ID:             "p",
+			Version:        "1",
+			DefaultProfile: "exec",
+			Templates:      []domain.PromptMessageTemplate{{Role: "user", Content: "hi"}},
+			OutputFormat:   domain.FormatText,
+			Validation:     domain.OutputContract{Format: domain.FormatText, ValidationMode: domain.ValidationNone},
+		}},
+		handlerProfileRepo{profile: &domain.ExecutionProfile{
+			ID:       "exec",
+			Endpoint: "http://example.invalid/v1",
+			Model:    "model",
+		}},
+		reader,
+		handlerRenderer{},
+		handlerLLMClient{},
+		nil,
+	)
+	return NewHandler(runner)
+}
+
+func assertHTTPErrorCode(t *testing.T, w *httptest.ResponseRecorder, status int, code string) {
+	t.Helper()
+
+	if w.Code != status {
+		t.Fatalf("expected %d, got %d body=%s", status, 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)
+	}
+	errBody := resp["error"].(map[string]any)
+	if errBody["code"] != code {
+		t.Fatalf("expected code %q, got %#v", code, errBody["code"])
+	}
+}

internal/artifact/reader.go

@@ -7,15 +7,20 @@ import (
 	"fmt"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/defaults"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"io"
 	"mime"
 	"os"
 	"path/filepath"
+	"strings"
 )
 
 var (
 	ErrUnsupportedRefType = errors.New("unsupported artifact reference type")
 	ErrMissingInlineBody  = errors.New("missing body for inline artifact")
 	ErrMissingFilePath    = errors.New("missing file path for file artifact")
+	ErrFileNotAllowed     = errors.New("file artifact references are not allowed")
+	ErrFileOutsideRoot    = errors.New("file artifact path is outside artifact root")
+	ErrFileTooLarge       = errors.New("file artifact exceeds size limit")
 )
 
 // Reader resolves artifact references into actual artifacts.
@@ -26,7 +31,7 @@ type Reader interface {
 // CompositeReader routes artifact resolution based on the reference type.
 type CompositeReader struct {
 	inlineReader *inlineReader
-	fileReader   *fileReader
+	fileReader   Reader
 }
 
 func NewCompositeReader() Reader {
@@ -36,6 +41,21 @@ func NewCompositeReader() Reader {
 	}
 }
 
+func NewRestrictedCompositeReader(root string) (Reader, error) {
+	return NewRestrictedCompositeReaderWithLimit(root, 0)
+}
+
+func NewRestrictedCompositeReaderWithLimit(root string, maxBytes int64) (Reader, error) {
+	fileReader, err := newRestrictedFileReader(root, maxBytes)
+	if err != nil {
+		return nil, err
+	}
+	return &CompositeReader{
+		inlineReader: &inlineReader{},
+		fileReader:   fileReader,
+	}, nil
+}
+
 func (c *CompositeReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error) {
 	select {
 	case <-ctx.Done():
@@ -89,21 +109,133 @@ func (r *fileReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.
 		return nil, ErrMissingFilePath
 	}
 
-	data, err := os.ReadFile(ref.URI)
-	if err != nil {
-		return nil, fmt.Errorf("failed to read file %s: %w", ref.URI, err)
+	return readFileArtifact(ref.URI)
+}
+
+type deniedFileReader struct{}
+
+func (r deniedFileReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error) {
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	default:
 	}
 
-	contentType := mime.TypeByExtension(filepath.Ext(ref.URI))
+	if ref.URI == "" {
+		return nil, ErrMissingFilePath
+	}
+	return nil, ErrFileNotAllowed
+}
+
+type restrictedFileReader struct {
+	root     string
+	maxBytes int64
+}
+
+func newRestrictedFileReader(root string, maxBytes int64) (Reader, error) {
+	if maxBytes < 0 {
+		return nil, fmt.Errorf("artifact size limit must be greater than or equal to 0")
+	}
+	cleanRoot := strings.TrimSpace(root)
+	if cleanRoot == "" {
+		return deniedFileReader{}, nil
+	}
+	absRoot, err := filepath.Abs(filepath.Clean(cleanRoot))
+	if err != nil {
+		return nil, fmt.Errorf("resolve artifact root: %w", err)
+	}
+	return &restrictedFileReader{root: absRoot, maxBytes: maxBytes}, nil
+}
+
+func (r *restrictedFileReader) Read(ctx context.Context, ref domain.ArtifactRef) (*domain.Artifact, error) {
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	default:
+	}
+
+	if ref.URI == "" {
+		return nil, ErrMissingFilePath
+	}
+
+	path, err := r.resolveLexicalPath(ref.URI)
+	if err != nil {
+		return nil, err
+	}
+	return readFileArtifactWithLimit(path, r.maxBytes)
+}
+
+// resolveLexicalPath checks cleaned path containment without resolving symlinks.
+func (r *restrictedFileReader) resolveLexicalPath(rawPath string) (string, error) {
+	cleanPath := filepath.Clean(strings.TrimSpace(rawPath))
+	var candidate string
+	if filepath.IsAbs(cleanPath) {
+		candidate = cleanPath
+	} else {
+		candidate = filepath.Join(r.root, cleanPath)
+	}
+
+	absCandidate, err := filepath.Abs(candidate)
+	if err != nil {
+		return "", fmt.Errorf("resolve artifact path: %w", err)
+	}
+	absCandidate = filepath.Clean(absCandidate)
+
+	rel, err := filepath.Rel(r.root, absCandidate)
+	if err != nil {
+		return "", fmt.Errorf("compare artifact path to root: %w", err)
+	}
+	if rel == ".." || strings.HasPrefix(rel, ".."+string(filepath.Separator)) || filepath.IsAbs(rel) {
+		return "", ErrFileOutsideRoot
+	}
+	return absCandidate, nil
+}
+
+func readFileArtifact(path string) (*domain.Artifact, error) {
+	return readFileArtifactWithLimit(path, 0)
+}
+
+func readFileArtifactWithLimit(path string, maxBytes int64) (*domain.Artifact, error) {
+	if maxBytes < 0 {
+		return nil, fmt.Errorf("file size limit must be greater than or equal to 0")
+	}
+
+	file, err := os.Open(path)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+	}
+	defer file.Close()
+
+	info, err := file.Stat()
+	if err != nil {
+		return nil, fmt.Errorf("failed to stat file %s: %w", path, err)
+	}
+	if maxBytes > 0 && info.Size() > maxBytes {
+		return nil, ErrFileTooLarge
+	}
+
+	var reader io.Reader = file
+	if maxBytes > 0 {
+		reader = io.LimitReader(file, maxBytes+1)
+	}
+	data, err := io.ReadAll(reader)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read file %s: %w", path, err)
+	}
+	if maxBytes > 0 && int64(len(data)) > maxBytes {
+		return nil, ErrFileTooLarge
+	}
+
+	contentType := mime.TypeByExtension(filepath.Ext(path))
 	if contentType == "" {
 		contentType = defaults.ContentTypeTextPlain
 	}
 
 	return &domain.Artifact{
-		Name:        filepath.Base(ref.URI),
+		Name:        filepath.Base(path),
 		ContentType: contentType,
 		Body:        data,
-		URI:         ref.URI,
+		URI:         path,
 		Size:        int64(len(data)),
 		Hash:        fmt.Sprintf("%x", sha256.Sum256(data)),
 	}, nil

internal/artifact/reader_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"os"
+	"path/filepath"
 	"testing"
 
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
@@ -46,8 +47,8 @@ func TestCompositeReader_Read(t *testing.T) {
 
 	t.Run("unsupported ref type", func(t *testing.T) {
 		ref := domain.ArtifactRef{
-			Type: domain.ArtifactRefS3,
-			URI:  "s3://bucket/key",
+			Type: domain.ArtifactRefType("unsupported"),
+			URI:  "unsupported://bucket/key",
 		}
 		_, err := reader.Read(ctx, ref)
 		if !errors.Is(err, ErrUnsupportedRefType) {
@@ -56,6 +57,157 @@ func TestCompositeReader_Read(t *testing.T) {
 	})
 }
 
+func TestRestrictedCompositeReader(t *testing.T) {
+	ctx := context.Background()
+	root := t.TempDir()
+	outside := t.TempDir()
+
+	if err := os.WriteFile(filepath.Join(root, "input.txt"), []byte("allowed"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Mkdir(filepath.Join(root, "nested"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(outside, "secret.txt"), []byte("denied"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	reader, err := NewRestrictedCompositeReader(root)
+	if err != nil {
+		t.Fatalf("expected restricted reader construction, got %v", err)
+	}
+
+	t.Run("accepts relative contained path", func(t *testing.T) {
+		art, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "nested/../input.txt"})
+		if err != nil {
+			t.Fatalf("expected contained relative path to succeed, got %v", err)
+		}
+		if string(art.Body) != "allowed" {
+			t.Fatalf("unexpected artifact body: %q", string(art.Body))
+		}
+	})
+
+	t.Run("accepts absolute contained path", func(t *testing.T) {
+		art, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filepath.Join(root, "input.txt")})
+		if err != nil {
+			t.Fatalf("expected contained absolute path to succeed, got %v", err)
+		}
+		if art.Name != "input.txt" {
+			t.Fatalf("unexpected artifact name: %q", art.Name)
+		}
+	})
+
+	t.Run("rejects relative traversal outside root", func(t *testing.T) {
+		_, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filepath.Join("..", filepath.Base(outside), "secret.txt")})
+		if !errors.Is(err, ErrFileOutsideRoot) {
+			t.Fatalf("expected ErrFileOutsideRoot, got %v", err)
+		}
+	})
+
+	t.Run("rejects absolute path outside root", func(t *testing.T) {
+		_, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: filepath.Join(outside, "secret.txt")})
+		if !errors.Is(err, ErrFileOutsideRoot) {
+			t.Fatalf("expected ErrFileOutsideRoot, got %v", err)
+		}
+	})
+}
+
+func TestRestrictedCompositeReaderFollowsSymlinkInsideRoot(t *testing.T) {
+	ctx := context.Background()
+	root := t.TempDir()
+	outside := t.TempDir()
+
+	target := filepath.Join(outside, "linked.txt")
+	if err := os.WriteFile(target, []byte("linked outside root"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	link := filepath.Join(root, "linked.txt")
+	if err := os.Symlink(target, link); err != nil {
+		t.Skipf("symlink creation unavailable: %v", err)
+	}
+
+	reader, err := NewRestrictedCompositeReader(root)
+	if err != nil {
+		t.Fatalf("expected restricted reader construction, got %v", err)
+	}
+
+	art, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "linked.txt"})
+	if err != nil {
+		t.Fatalf("expected symlink inside root to be followed, got %v", err)
+	}
+	if string(art.Body) != "linked outside root" {
+		t.Fatalf("unexpected artifact body: %q", string(art.Body))
+	}
+}
+
+func TestRestrictedCompositeReaderWithoutRootDeniesFileRefs(t *testing.T) {
+	reader, err := NewRestrictedCompositeReader("")
+	if err != nil {
+		t.Fatalf("expected restricted reader construction, got %v", err)
+	}
+
+	art, err := reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: "inline"})
+	if err != nil {
+		t.Fatalf("expected inline ref to work without artifact root, got %v", err)
+	}
+	if string(art.Body) != "inline" {
+		t.Fatalf("unexpected inline body: %q", string(art.Body))
+	}
+
+	_, err = reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "input.txt"})
+	if !errors.Is(err, ErrFileNotAllowed) {
+		t.Fatalf("expected ErrFileNotAllowed, got %v", err)
+	}
+}
+
+func TestRestrictedCompositeReaderFileSizeLimit(t *testing.T) {
+	ctx := context.Background()
+	root := t.TempDir()
+	if err := os.WriteFile(filepath.Join(root, "exact.txt"), []byte("12345"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "large.txt"), []byte("123456"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	reader, err := NewRestrictedCompositeReaderWithLimit(root, 5)
+	if err != nil {
+		t.Fatalf("expected restricted reader construction, got %v", err)
+	}
+
+	art, err := reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "exact.txt"})
+	if err != nil {
+		t.Fatalf("expected file at limit to succeed, got %v", err)
+	}
+	if string(art.Body) != "12345" {
+		t.Fatalf("unexpected artifact body: %q", string(art.Body))
+	}
+
+	_, err = reader.Read(ctx, domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "large.txt"})
+	if !errors.Is(err, ErrFileTooLarge) {
+		t.Fatalf("expected ErrFileTooLarge, got %v", err)
+	}
+}
+
+func TestRestrictedCompositeReaderFileSizeLimitZeroDisablesLimit(t *testing.T) {
+	root := t.TempDir()
+	if err := os.WriteFile(filepath.Join(root, "large.txt"), []byte("123456"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	reader, err := NewRestrictedCompositeReaderWithLimit(root, 0)
+	if err != nil {
+		t.Fatalf("expected restricted reader construction, got %v", err)
+	}
+	art, err := reader.Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefFile, URI: "large.txt"})
+	if err != nil {
+		t.Fatalf("expected unlimited reader to succeed, got %v", err)
+	}
+	if string(art.Body) != "123456" {
+		t.Fatalf("unexpected artifact body: %q", string(art.Body))
+	}
+}
+
 func TestFileReader_Read(t *testing.T) {
 	content := []byte("test file content")
 	tmpFile, err := os.CreateTemp("", "artifact_test_*.txt")

internal/config/config.go

@@ -41,6 +41,10 @@ type Config struct {
 
 type ServerConfig struct {
 	Addr             string `yaml:"addr"`
+	ArtifactRoot     string `yaml:"artifact_root"`
+	MaxRequestBytes  *int64 `yaml:"max_request_bytes"`
+	MaxArtifactBytes *int64 `yaml:"max_artifact_bytes"`
+	MaxResponseBytes *int64 `yaml:"max_response_bytes"`
 }
 
 type DefaultsConfig struct {
@@ -53,6 +57,10 @@ type AppSettings struct {
 	ProfileDir          string
 	SchemaDir           string
 	ServerAddr          string
+	ArtifactRoot        string
+	MaxRequestBytes     int64
+	MaxArtifactBytes    int64
+	MaxResponseBytes    int64
 	DefaultRenderFormat renderformat.PreparedRunOutputFormat
 }
 
@@ -62,6 +70,10 @@ type CLIOverrides struct {
 	ProfileDir       string
 	SchemaDir        string
 	ServerAddr       string
+	ArtifactRoot     string
+	MaxRequestBytes  *int64
+	MaxArtifactBytes *int64
+	MaxResponseBytes *int64
 	RenderFormat     string
 }
 
@@ -70,6 +82,9 @@ func BuiltInDefaults() AppSettings {
 	return AppSettings{
 		SchemaDir:           defaults.SchemaDirDefault,
 		ServerAddr:          defaults.HTTPAddrDefault,
+		MaxRequestBytes:     defaults.HTTPMaxRequestBytesDefault,
+		MaxArtifactBytes:    defaults.HTTPMaxArtifactBytesDefault,
+		MaxResponseBytes:    defaults.HTTPMaxResponseBytesDefault,
 		DefaultRenderFormat: renderformat.DefaultPreparedRunOutputFormat,
 	}
 }
@@ -142,6 +157,27 @@ func ApplyCLIOverrides(base AppSettings, overrides CLIOverrides) (AppSettings, e
 	if v := strings.TrimSpace(overrides.ServerAddr); v != "" {
 		out.ServerAddr = v
 	}
+	if v := strings.TrimSpace(overrides.ArtifactRoot); v != "" {
+		out.ArtifactRoot = filepath.Clean(v)
+	}
+	if overrides.MaxRequestBytes != nil {
+		if *overrides.MaxRequestBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_request_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxRequestBytes = *overrides.MaxRequestBytes
+	}
+	if overrides.MaxArtifactBytes != nil {
+		if *overrides.MaxArtifactBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_artifact_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxArtifactBytes = *overrides.MaxArtifactBytes
+	}
+	if overrides.MaxResponseBytes != nil {
+		if *overrides.MaxResponseBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_response_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxResponseBytes = *overrides.MaxResponseBytes
+	}
 	if rawFormat := strings.TrimSpace(overrides.RenderFormat); rawFormat != "" {
 		parsed, err := renderformat.ParsePreparedRunOutputFormat(rawFormat)
 		if err != nil {
@@ -181,6 +217,27 @@ func applyConfig(base AppSettings, cfg Config) (AppSettings, error) {
 	if v := strings.TrimSpace(cfg.Server.Addr); v != "" {
 		out.ServerAddr = v
 	}
+	if v := strings.TrimSpace(cfg.Server.ArtifactRoot); v != "" {
+		out.ArtifactRoot = filepath.Clean(v)
+	}
+	if cfg.Server.MaxRequestBytes != nil {
+		if *cfg.Server.MaxRequestBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_request_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxRequestBytes = *cfg.Server.MaxRequestBytes
+	}
+	if cfg.Server.MaxArtifactBytes != nil {
+		if *cfg.Server.MaxArtifactBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_artifact_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxArtifactBytes = *cfg.Server.MaxArtifactBytes
+	}
+	if cfg.Server.MaxResponseBytes != nil {
+		if *cfg.Server.MaxResponseBytes < 0 {
+			return AppSettings{}, fmt.Errorf("%w: server.max_response_bytes must be greater than or equal to 0", ErrInvalidConfig)
+		}
+		out.MaxResponseBytes = *cfg.Server.MaxResponseBytes
+	}
 	if rawFormat := strings.TrimSpace(cfg.Defaults.RenderFormat); rawFormat != "" {
 		parsed, err := renderformat.ParsePreparedRunOutputFormat(rawFormat)
 		if err != nil {

internal/config/config_test.go

@@ -6,6 +6,7 @@ import (
 	"path/filepath"
 	"testing"
 
+	"gitea.maximumdirect.net/eric/scriptorium/internal/defaults"
 	renderformat "gitea.maximumdirect.net/eric/scriptorium/internal/format"
 )
 
@@ -24,6 +25,20 @@ func TestLoadConfigMissingImplicitPathUsesBuiltInDefaults(t *testing.T) {
 	}
 }
 
+func TestBuiltInDefaultsIncludeHTTPSizeLimits(t *testing.T) {
+	got := BuiltInDefaults()
+
+	if got.MaxRequestBytes != defaults.HTTPMaxRequestBytesDefault {
+		t.Fatalf("unexpected max request bytes: %d", got.MaxRequestBytes)
+	}
+	if got.MaxArtifactBytes != defaults.HTTPMaxArtifactBytesDefault {
+		t.Fatalf("unexpected max artifact bytes: %d", got.MaxArtifactBytes)
+	}
+	if got.MaxResponseBytes != defaults.HTTPMaxResponseBytesDefault {
+		t.Fatalf("unexpected max response bytes: %d", got.MaxResponseBytes)
+	}
+}
+
 func TestLoadConfigMissingExplicitPathReturnsError(t *testing.T) {
 	tmp := t.TempDir()
 	missing := filepath.Join(tmp, "missing.yml")
@@ -92,6 +107,10 @@ profile_dir: ./profiles
 schema_dir: ./schemas
 server:
   addr: 127.0.0.1:9090
+  artifact_root: ./artifacts
+  max_request_bytes: 1024
+  max_artifact_bytes: 2048
+  max_response_bytes: 4096
 defaults:
   render_format: json
 `)
@@ -113,11 +132,61 @@ defaults:
 	if got.ServerAddr != "127.0.0.1:9090" {
 		t.Fatalf("unexpected server.addr: %q", got.ServerAddr)
 	}
+	if got.ArtifactRoot != filepath.Clean("./artifacts") {
+		t.Fatalf("unexpected server.artifact_root: %q", got.ArtifactRoot)
+	}
+	if got.MaxRequestBytes != 1024 {
+		t.Fatalf("unexpected server.max_request_bytes: %d", got.MaxRequestBytes)
+	}
+	if got.MaxArtifactBytes != 2048 {
+		t.Fatalf("unexpected server.max_artifact_bytes: %d", got.MaxArtifactBytes)
+	}
+	if got.MaxResponseBytes != 4096 {
+		t.Fatalf("unexpected server.max_response_bytes: %d", got.MaxResponseBytes)
+	}
 	if got.DefaultRenderFormat != renderformat.PreparedRunFormatJSON {
 		t.Fatalf("unexpected defaults.render_format: %q", got.DefaultRenderFormat)
 	}
 }
 
+func TestLoadConfigAcceptsZeroHTTPSizeLimits(t *testing.T) {
+	path := writeConfigFile(t, "config.yml", `
+server:
+  max_request_bytes: 0
+  max_artifact_bytes: 0
+  max_response_bytes: 0
+`)
+
+	got, err := LoadConfig(path, true)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if got.MaxRequestBytes != 0 || got.MaxArtifactBytes != 0 || got.MaxResponseBytes != 0 {
+		t.Fatalf("expected zero limits to be preserved, got request=%d artifact=%d response=%d", got.MaxRequestBytes, got.MaxArtifactBytes, got.MaxResponseBytes)
+	}
+}
+
+func TestLoadConfigRejectsNegativeHTTPSizeLimits(t *testing.T) {
+	tests := []struct {
+		name string
+		body string
+	}{
+		{name: "request", body: "server:\n  max_request_bytes: -1\n"},
+		{name: "artifact", body: "server:\n  max_artifact_bytes: -1\n"},
+		{name: "response", body: "server:\n  max_response_bytes: -1\n"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			path := writeConfigFile(t, "config.yml", tc.body)
+			_, err := LoadConfig(path, true)
+			if !errors.Is(err, ErrInvalidConfig) {
+				t.Fatalf("expected ErrInvalidConfig, got %v", err)
+			}
+		})
+	}
+}
+
 func TestLoadConfigEmptyFileResolvesToBuiltInDefaults(t *testing.T) {
 	path := writeConfigFile(t, "config.yml", "")
 
@@ -185,14 +254,25 @@ func TestApplyCLIOverridesAppliesPrecedence(t *testing.T) {
 		ProfileDir:          "/from/config/profiles",
 		SchemaDir:           "/from/config/schemas",
 		ServerAddr:          ":1234",
+		ArtifactRoot:        "/from/config/artifacts",
+		MaxRequestBytes:     111,
+		MaxArtifactBytes:    222,
+		MaxResponseBytes:    333,
 		DefaultRenderFormat: renderformat.PreparedRunFormatJSON,
 	}
+	maxRequestBytes := int64(0)
+	maxArtifactBytes := int64(444)
+	maxResponseBytes := int64(555)
 
 	got, err := ApplyCLIOverrides(base, CLIOverrides{
 		PromptDir:        "./prompts-cli",
 		ProfileDir:       "./profiles-cli",
 		SchemaDir:        "./schemas-cli",
 		ServerAddr:       ":8081",
+		ArtifactRoot:     "./artifacts-cli",
+		MaxRequestBytes:  &maxRequestBytes,
+		MaxArtifactBytes: &maxArtifactBytes,
+		MaxResponseBytes: &maxResponseBytes,
 		RenderFormat:     "text",
 	})
 	if err != nil {
@@ -211,11 +291,45 @@ func TestApplyCLIOverridesAppliesPrecedence(t *testing.T) {
 	if got.ServerAddr != ":8081" {
 		t.Fatalf("unexpected server addr: %q", got.ServerAddr)
 	}
+	if got.ArtifactRoot != filepath.Clean("./artifacts-cli") {
+		t.Fatalf("unexpected artifact root: %q", got.ArtifactRoot)
+	}
+	if got.MaxRequestBytes != 0 {
+		t.Fatalf("unexpected max request bytes: %d", got.MaxRequestBytes)
+	}
+	if got.MaxArtifactBytes != 444 {
+		t.Fatalf("unexpected max artifact bytes: %d", got.MaxArtifactBytes)
+	}
+	if got.MaxResponseBytes != 555 {
+		t.Fatalf("unexpected max response bytes: %d", got.MaxResponseBytes)
+	}
 	if got.DefaultRenderFormat != renderformat.PreparedRunFormatText {
 		t.Fatalf("unexpected render format: %q", got.DefaultRenderFormat)
 	}
 }
 
+func TestApplyCLIOverridesRejectsNegativeHTTPSizeLimits(t *testing.T) {
+	negative := int64(-1)
+
+	tests := []struct {
+		name      string
+		overrides CLIOverrides
+	}{
+		{name: "request", overrides: CLIOverrides{MaxRequestBytes: &negative}},
+		{name: "artifact", overrides: CLIOverrides{MaxArtifactBytes: &negative}},
+		{name: "response", overrides: CLIOverrides{MaxResponseBytes: &negative}},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			_, err := ApplyCLIOverrides(BuiltInDefaults(), tc.overrides)
+			if !errors.Is(err, ErrInvalidConfig) {
+				t.Fatalf("expected ErrInvalidConfig, got %v", err)
+			}
+		})
+	}
+}
+
 func TestApplyCLIOverridesInvalidRenderFormatReturnsError(t *testing.T) {
 	_, err := ApplyCLIOverrides(BuiltInDefaults(), CLIOverrides{RenderFormat: "yaml"})
 	if err == nil {

internal/defaults/defaults.go

@@ -14,6 +14,9 @@ const (
 	ContentTypeTextMarkdown     = "text/markdown"
 	ContentTypeApplicationJSON  = "application/json"
 	OpenAIChatCompletionsPath   = "/chat/completions"
+	HTTPMaxRequestBytesDefault  = 16 * 1024 * 1024
+	HTTPMaxArtifactBytesDefault = 16 * 1024 * 1024
+	HTTPMaxResponseBytesDefault = 16 * 1024 * 1024
 
 	ExecutionDefaultTemperature    = 0.0
 	ExecutionDefaultMaxTokens      = 0

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.
@@ -41,14 +40,33 @@ const (
 	ValidationSkipped ValidationStatus = "skipped"
 )
 
+// CacheControlType defines provider cache behavior for prompt content.
+type CacheControlType string
+
+const (
+	CacheControlEphemeral CacheControlType = "ephemeral"
+)
+
+const (
+	// SessionIDMaxLength is OpenRouter's documented maximum session_id length.
+	SessionIDMaxLength = 256
+)
+
+// CacheControl describes provider cache metadata attached to prompt content.
+type CacheControl struct {
+	Type CacheControlType `yaml:"type" json:"type"`
+	TTL  string           `yaml:"ttl,omitempty" json:"ttl,omitempty"`
+}
+
 // RunRequest represents a request to generate a single artifact.
 type RunRequest struct {
 	PromptID      string
 	PromptVersion string
 	ProfileID     string
+	APIKey        string `json:"-" yaml:"-"`
 	Inputs        map[string]ArtifactRef
 	Vars          map[string]string
-	Execution     *ExecutionTarget
+	Execution     *ExecutionTargetOverride
 	Validation    *OutputContract
 	Metadata      map[string]string
 }
@@ -72,7 +90,6 @@ type RunResult struct {
 	StartTime            time.Time
 	EndTime              time.Time
 	Duration             time.Duration
-	Error                error
 }
 
 // PreparedRun contains pre-LLM execution state from the prepare/render phase.
@@ -83,9 +100,11 @@ type PreparedRun struct {
 	PromptHash           string                  `json:"prompt_hash,omitempty"`
 	SelectedProfileID    string                  `json:"selected_profile_id"`
 	EffectiveModelParams ExecutionTarget         `json:"effective_model_params"`
+	TargetPresence       ExecutionTargetPresence `json:"-"`
 	OutputContract       OutputContract          `json:"output_contract"`
 	StructuredOutput     *StructuredOutputSpec   `json:"structured_output,omitempty"`
 	InputHashes          map[string]string       `json:"input_hashes,omitempty"`
+	SessionID            string                  `json:"session_id,omitempty"`
 	RenderedPromptHash   string                  `json:"rendered_prompt_hash"`
 	Messages             []RenderedMessage       `json:"messages"`
 	StartTime            time.Time               `json:"start_time,omitempty"`
@@ -116,6 +135,7 @@ type PromptDefinition struct {
 	Version        string                  `yaml:"version"`
 	DefaultProfile string                  `yaml:"default_profile"`
 	Description    string                  `yaml:"description"`
+	SessionID      string                  `yaml:"session_id" json:"session_id,omitempty"`
 	Inputs         []PromptInput           `yaml:"inputs"`
 	Templates      []PromptMessageTemplate `yaml:"templates"`
 	OutputFormat   OutputFormat            `yaml:"output_format"`
@@ -135,6 +155,7 @@ type PromptMessageTemplate struct {
 	Role         string        `yaml:"role"`
 	Content      string        `yaml:"content"`
 	ContentFile  string        `yaml:"content_file"`
+	CacheControl *CacheControl `yaml:"cache_control,omitempty" json:"cache_control,omitempty"`
 }
 
 // ExecutionProfile describes how and where to execute a model.
@@ -146,9 +167,34 @@ type ExecutionProfile struct {
 	MaxTokens       int            `yaml:"max_tokens"`
 	TopP            float64        `yaml:"top_p"`
 	TimeoutSeconds  int            `yaml:"timeout_seconds"`
+	ServiceTier     string         `yaml:"service_tier"`
 	ReasoningEffort string         `yaml:"reasoning_effort"`
 	APIKeyEnv       string         `yaml:"api_key_env"`
-	ExtraParams     map[string]string `yaml:"extra_params"`
+	APIKeyRequired  bool           `yaml:"-" json:"-"`
+	ExtraParams     map[string]any `yaml:"extra_params"`
+}
+
+// ExecutionTargetOverride represents per-request runtime setting overrides.
+type ExecutionTargetOverride struct {
+	Endpoint        string         `json:"endpoint,omitempty"`
+	Model           string         `json:"model,omitempty"`
+	Temperature     *float64       `json:"temperature,omitempty"`
+	MaxTokens       *int           `json:"max_tokens,omitempty"`
+	TopP            *float64       `json:"top_p,omitempty"`
+	TimeoutSeconds  *int           `json:"timeout_seconds,omitempty"`
+	ServiceTier     string         `json:"service_tier,omitempty"`
+	ReasoningEffort string         `json:"reasoning_effort,omitempty"`
+	APIKeyEnv       string         `json:"api_key_env,omitempty"`
+	ExtraParams     map[string]any `json:"extra_params,omitempty"`
+}
+
+// ExecutionTargetPresence tracks which effective runtime fields came from an
+// explicit request override even when the resolved value is a zero value.
+type ExecutionTargetPresence struct {
+	Temperature    bool
+	MaxTokens      bool
+	TopP           bool
+	TimeoutSeconds bool
 }
 
 // ExecutionTarget represents effective model runtime settings for a run.
@@ -159,9 +205,12 @@ type ExecutionTarget struct {
 	MaxTokens       int            `yaml:"max_tokens" json:"max_tokens"`
 	TopP            float64        `yaml:"top_p" json:"top_p"`
 	TimeoutSeconds  int            `yaml:"timeout_seconds" json:"timeout_seconds"`
+	ServiceTier     string         `yaml:"service_tier" json:"service_tier"`
 	ReasoningEffort string         `yaml:"reasoning_effort" json:"reasoning_effort"`
 	APIKeyEnv       string         `yaml:"api_key_env" json:"api_key_env"`
-	ExtraParams     map[string]string `yaml:"extra_params" json:"extra_params"`
+	APIKey          string         `yaml:"-" json:"-"`
+	APIKeyRequired  bool           `yaml:"-" json:"-"`
+	ExtraParams     map[string]any `yaml:"extra_params" json:"extra_params"`
 }
 
 // OutputContract defines the requirements for the output artifact.
@@ -174,6 +223,7 @@ type OutputContract struct {
 
 // RenderedPrompt represents the prompt after template application.
 type RenderedPrompt struct {
+	SessionID string            `json:"session_id,omitempty"`
 	Messages  []RenderedMessage `json:"messages"`
 }
 
@@ -181,12 +231,14 @@ type RenderedPrompt struct {
 type RenderedMessage struct {
 	Role         string        `json:"role"`
 	Content      string        `json:"content"`
+	CacheControl *CacheControl `json:"cache_control,omitempty"`
 }
 
 // GenerateRequest is the internal request passed to the LLM client.
 type GenerateRequest struct {
 	Prompt           RenderedPrompt
 	Target           ExecutionTarget
+	TargetPresence   ExecutionTargetPresence
 	StructuredOutput *StructuredOutputSpec
 }
 
@@ -221,6 +273,8 @@ type TokenUsage struct {
 	PromptTokens     int
 	CompletionTokens int
 	TotalTokens      int
+	CachedTokens     int
+	CacheWriteTokens int
 }
 
 // ValidationResult represents the outcome of an output validation.
@@ -232,23 +286,3 @@ type ValidationResult struct {
 	RepairAttempts int
 	IsValid        bool
 }
-
-// RunMetadata contains auditing information for a run.
-type RunMetadata struct {
-	RunID              string
-	PromptID           string
-	PromptVersion      string
-	PromptHash         string
-	RenderedPromptHash string
-	SelectedProfileID  string
-	InputHashes        map[string]string
-	ModelEndpoint      string
-	ModelName          string
-	Params             ExecutionTarget
-	Timestamp          time.Time
-	Duration           time.Duration
-	Usage              TokenUsage
-	ValidationMode     ValidationMode
-	ValidationStatus   ValidationStatus
-	RepairAttempts     int
-}

internal/domain/prepared_run_test.go

@@ -20,6 +20,7 @@ func TestPreparedRunJSONDoesNotIncludeSecretValues(t *testing.T) {
 			Endpoint:  "http://llm/v1",
 			Model:     "gpt-test",
 			APIKeyEnv: envName,
+			APIKey:    secret,
 		},
 		InputHashes:        map[string]string{"transcript": "hash-1"},
 		RenderedPromptHash: "rendered-hash",
@@ -53,3 +54,88 @@ func TestPreparedRunJSONDoesNotIncludeSecretValues(t *testing.T) {
 		}
 	}
 }
+
+func TestPreparedRunJSONIncludesMessageCacheControlOnlyWhenPresent(t *testing.T) {
+	prepared := PreparedRun{
+		PromptID:          "prompt.id",
+		SelectedProfileID: "local-fast",
+		EffectiveModelParams: ExecutionTarget{
+			Endpoint: "http://llm/v1",
+			Model:    "gpt-test",
+		},
+		RenderedPromptHash: "rendered-hash",
+		Messages: []RenderedMessage{
+			{
+				Role:    "system",
+				Content: "You are helpful.",
+				CacheControl: &CacheControl{
+					Type: CacheControlEphemeral,
+					TTL:  "1h",
+				},
+			},
+			{Role: "user", Content: "Summarize this."},
+		},
+	}
+
+	b, err := json.Marshal(prepared)
+	if err != nil {
+		t.Fatalf("marshal failed: %v", err)
+	}
+
+	var decoded struct {
+		Messages []map[string]any `json:"messages"`
+	}
+	if err := json.Unmarshal(b, &decoded); err != nil {
+		t.Fatalf("unmarshal failed: %v", err)
+	}
+	if len(decoded.Messages) != 2 {
+		t.Fatalf("expected 2 messages, got %d", len(decoded.Messages))
+	}
+
+	cacheControl, ok := decoded.Messages[0]["cache_control"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected cache_control on first message, got %#v", decoded.Messages[0])
+	}
+	if cacheControl["type"] != string(CacheControlEphemeral) || cacheControl["ttl"] != "1h" {
+		t.Fatalf("unexpected cache_control payload: %#v", cacheControl)
+	}
+	if _, ok := decoded.Messages[1]["cache_control"]; ok {
+		t.Fatalf("expected second message to omit cache_control, got %#v", decoded.Messages[1])
+	}
+}
+
+func TestPreparedRunJSONIncludesSessionIDOnlyWhenPresent(t *testing.T) {
+	prepared := PreparedRun{
+		PromptID:          "prompt.id",
+		SelectedProfileID: "local-fast",
+		EffectiveModelParams: ExecutionTarget{
+			Endpoint: "http://llm/v1",
+			Model:    "gpt-test",
+		},
+		SessionID:          "session-123",
+		RenderedPromptHash: "rendered-hash",
+		Messages:           []RenderedMessage{{Role: "user", Content: "Summarize this."}},
+	}
+
+	b, err := json.Marshal(prepared)
+	if err != nil {
+		t.Fatalf("marshal failed: %v", err)
+	}
+
+	var decoded map[string]any
+	if err := json.Unmarshal(b, &decoded); err != nil {
+		t.Fatalf("unmarshal failed: %v", err)
+	}
+	if decoded["session_id"] != "session-123" {
+		t.Fatalf("expected session_id in prepared run JSON, got %#v", decoded["session_id"])
+	}
+
+	prepared.SessionID = ""
+	b, err = json.Marshal(prepared)
+	if err != nil {
+		t.Fatalf("marshal failed: %v", err)
+	}
+	if strings.Contains(string(b), "session_id") {
+		t.Fatalf("expected empty session_id to be omitted, got %s", b)
+	}
+}

internal/filecatalog/catalog.go

@@ -0,0 +1,142 @@
+package filecatalog
+
+import (
+	"context"
+	"fmt"
+	"io/fs"
+	"os"
+	"path"
+	"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
+}
+
+// FindFSYAMLFiles returns sorted paths for .yaml and .yml files under root in fsys.
+func FindFSYAMLFiles(ctx context.Context, fsys fs.FS, root string) ([]string, error) {
+	cleanRoot := CleanFSRoot(root)
+	var files []string
+	err := fs.WalkDir(fsys, cleanRoot, func(name string, d fs.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, name)
+		return nil
+	})
+	sort.Strings(files)
+	return files, err
+}
+
+// RelativePath computes a clean relative path from root to path.
+func RelativePath(root string, filePath string) string {
+	rel, err := filepath.Rel(root, filePath)
+	if err != nil {
+		return filepath.Clean(filePath)
+	}
+	return filepath.Clean(rel)
+}
+
+// CleanFSRoot normalizes a root path for use with fs.FS.
+func CleanFSRoot(root string) string {
+	root = strings.TrimSpace(root)
+	if root == "" || root == "." {
+		return "."
+	}
+	return path.Clean(root)
+}
+
+// DisplayPath returns name relative to root for messages about fs.FS paths.
+func DisplayPath(root string, name string) string {
+	cleanRoot := CleanFSRoot(root)
+	cleanName := path.Clean(name)
+	if cleanRoot == "." {
+		return cleanName
+	}
+	prefix := strings.TrimSuffix(cleanRoot, "/") + "/"
+	if strings.HasPrefix(cleanName, prefix) {
+		return strings.TrimPrefix(cleanName, prefix)
+	}
+	return cleanName
+}
+
+// ResolveFSPath resolves userPath from baseDir and keeps it inside root.
+func ResolveFSPath(root string, baseDir string, userPath string) (string, string, error) {
+	cleanRoot := CleanFSRoot(root)
+	cleanBase := path.Clean(strings.TrimSpace(baseDir))
+	if cleanBase == "" {
+		cleanBase = cleanRoot
+	}
+	if !containsFSPath(cleanRoot, cleanBase) {
+		return "", "", fmt.Errorf("base path %q is outside source root %q", cleanBase, cleanRoot)
+	}
+
+	cleanUserPath := strings.TrimSpace(userPath)
+	if cleanUserPath == "" {
+		return "", "", fmt.Errorf("path is required")
+	}
+	cleanUserPath = path.Clean(cleanUserPath)
+	if path.IsAbs(cleanUserPath) {
+		return "", "", fmt.Errorf("path %q must be relative", userPath)
+	}
+
+	resolved := path.Clean(path.Join(cleanBase, cleanUserPath))
+	if !containsFSPath(cleanRoot, resolved) {
+		return "", "", fmt.Errorf("path %q escapes source root %q", userPath, cleanRoot)
+	}
+	return resolved, DisplayPath(cleanRoot, resolved), nil
+}
+
+func containsFSPath(root string, name string) bool {
+	root = CleanFSRoot(root)
+	name = path.Clean(name)
+	if root == "." {
+		return name == "." || (name != ".." && !strings.HasPrefix(name, "../"))
+	}
+	return name == root || strings.HasPrefix(name, strings.TrimSuffix(root, "/")+"/")
+}
+
+// Stem strips .yaml or .yml from a file name.
+func Stem(name string) string {
+	name = strings.TrimSuffix(name, ".yaml")
+	name = strings.TrimSuffix(name, ".yml")
+	return name
+}
+
+func IsYAMLFile(name string) bool {
+	return strings.HasSuffix(name, ".yaml") || strings.HasSuffix(name, ".yml")
+}

internal/filecatalog/catalog_test.go

@@ -0,0 +1,270 @@
+package filecatalog
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+	"testing/fstest"
+)
+
+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 TestFindFSYAMLFilesNestedSortedAndFiltered(t *testing.T) {
+	fsys := fstest.MapFS{
+		"prompts/z/prompt.yml":      &fstest.MapFile{Data: []byte("id: z")},
+		"prompts/a/profile.yaml":    &fstest.MapFile{Data: []byte("id: a")},
+		"prompts/a/ignore.txt":      &fstest.MapFile{Data: []byte("not yaml")},
+		"prompts/b/ignore.yaml.bak": &fstest.MapFile{Data: []byte("not yaml")},
+		"other/ignored.yaml":        &fstest.MapFile{Data: []byte("id: ignored")},
+	}
+
+	got, err := FindFSYAMLFiles(context.Background(), fsys, " prompts ")
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+
+	want := []string{
+		"prompts/a/profile.yaml",
+		"prompts/z/prompt.yml",
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("expected sorted YAML files %v, got %v", want, got)
+	}
+}
+
+func TestFindFSYAMLFilesHonorsContextCancellation(t *testing.T) {
+	fsys := fstest.MapFS{
+		"one.yaml": &fstest.MapFile{Data: []byte("id: one")},
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	_, err := FindFSYAMLFiles(ctx, fsys, ".")
+	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 TestCleanFSRoot(t *testing.T) {
+	tests := []struct {
+		name string
+		root string
+		want string
+	}{
+		{name: "empty", root: "", want: "."},
+		{name: "dot", root: ".", want: "."},
+		{name: "trimmed", root: " prompts/../profiles ", want: "profiles"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := CleanFSRoot(tc.root); got != tc.want {
+				t.Fatalf("expected %q, got %q", tc.want, got)
+			}
+		})
+	}
+}
+
+func TestDisplayPath(t *testing.T) {
+	tests := []struct {
+		name string
+		root string
+		path string
+		want string
+	}{
+		{name: "root dot", root: ".", path: "profiles/local.yaml", want: "profiles/local.yaml"},
+		{name: "nested root", root: "profiles", path: "profiles/local.yaml", want: "local.yaml"},
+		{name: "outside root", root: "profiles", path: "other/local.yaml", want: "other/local.yaml"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := DisplayPath(tc.root, tc.path); got != tc.want {
+				t.Fatalf("expected %q, got %q", tc.want, got)
+			}
+		})
+	}
+}
+
+func TestResolveFSPath(t *testing.T) {
+	tests := []struct {
+		name        string
+		root        string
+		baseDir     string
+		userPath    string
+		wantPath    string
+		wantDisplay string
+		wantErr     string
+	}{
+		{
+			name:        "sibling inside root",
+			root:        "prompts",
+			baseDir:     "prompts/nested",
+			userPath:    "./messages/user.tmpl",
+			wantPath:    "prompts/nested/messages/user.tmpl",
+			wantDisplay: "nested/messages/user.tmpl",
+		},
+		{
+			name:        "parent inside root",
+			root:        "prompts",
+			baseDir:     "prompts/nested",
+			userPath:    "../shared/user.tmpl",
+			wantPath:    "prompts/shared/user.tmpl",
+			wantDisplay: "shared/user.tmpl",
+		},
+		{
+			name:     "escape rejected",
+			root:     "prompts",
+			baseDir:  "prompts/nested",
+			userPath: "../../outside.tmpl",
+			wantErr:  "escapes source root",
+		},
+		{
+			name:     "absolute path rejected",
+			root:     "prompts",
+			baseDir:  "prompts/nested",
+			userPath: "/outside.tmpl",
+			wantErr:  "must be relative",
+		},
+		{
+			name:     "empty path rejected",
+			root:     "prompts",
+			baseDir:  "prompts/nested",
+			userPath: " ",
+			wantErr:  "path is required",
+		},
+		{
+			name:        "dot root allows normal relative path",
+			root:        ".",
+			baseDir:     ".",
+			userPath:    "schemas/events.schema.json",
+			wantPath:    "schemas/events.schema.json",
+			wantDisplay: "schemas/events.schema.json",
+		},
+		{
+			name:     "dot root rejects parent escape",
+			root:     ".",
+			baseDir:  ".",
+			userPath: "../outside.tmpl",
+			wantErr:  "escapes source root",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			gotPath, gotDisplay, err := ResolveFSPath(tc.root, tc.baseDir, tc.userPath)
+			if tc.wantErr != "" {
+				if err == nil {
+					t.Fatalf("expected error containing %q", tc.wantErr)
+				}
+				if !strings.Contains(err.Error(), tc.wantErr) {
+					t.Fatalf("expected error to contain %q, got %v", tc.wantErr, err)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatalf("expected no error, got %v", err)
+			}
+			if gotPath != tc.wantPath || gotDisplay != tc.wantDisplay {
+				t.Fatalf("expected path/display %q/%q, got %q/%q", tc.wantPath, tc.wantDisplay, gotPath, gotDisplay)
+			}
+		})
+	}
+}
+
+func TestStemStripsYAMLExtensions(t *testing.T) {
+	tests := []struct {
+		name string
+		in   string
+		want string
+	}{
+		{name: "yaml", in: "prompt.yaml", want: "prompt"},
+		{name: "yml", in: "profile.yml", want: "profile"},
+		{name: "other", in: "file.txt", want: "file.txt"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := Stem(tc.in); got != tc.want {
+				t.Fatalf("expected %q, got %q", tc.want, got)
+			}
+		})
+	}
+}
+
+func TestIsYAMLFile(t *testing.T) {
+	tests := []struct {
+		name string
+		in   string
+		want bool
+	}{
+		{name: "yaml", in: "prompt.yaml", want: true},
+		{name: "yml", in: "profile.yml", want: true},
+		{name: "backup", in: "profile.yaml.bak", want: false},
+		{name: "uppercase", in: "profile.YAML", want: false},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := IsYAMLFile(tc.in); got != tc.want {
+				t.Fatalf("expected %v, got %v", 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/format/prepared_run.go

@@ -96,6 +96,9 @@ func (textPreparedRunFormatter) Format(prepared *domain.PreparedRun) ([]byte, er
 	if prepared.PromptHash != "" {
 		fmt.Fprintf(&b, "prompt_hash: %s\n", prepared.PromptHash)
 	}
+	if prepared.SessionID != "" {
+		fmt.Fprintf(&b, "session_id: %s\n", prepared.SessionID)
+	}
 	fmt.Fprintf(&b, "rendered_prompt_hash: %s\n", prepared.RenderedPromptHash)
 
 	target := prepared.EffectiveModelParams
@@ -106,6 +109,9 @@ func (textPreparedRunFormatter) Format(prepared *domain.PreparedRun) ([]byte, er
 	fmt.Fprintf(&b, "  max_tokens: %d\n", target.MaxTokens)
 	fmt.Fprintf(&b, "  top_p: %g\n", target.TopP)
 	fmt.Fprintf(&b, "  timeout_seconds: %d\n", target.TimeoutSeconds)
+	if target.ServiceTier != "" {
+		fmt.Fprintf(&b, "  service_tier: %s\n", target.ServiceTier)
+	}
 	if target.ReasoningEffort != "" {
 		fmt.Fprintf(&b, "  reasoning_effort: %s\n", target.ReasoningEffort)
 	}
@@ -120,7 +126,11 @@ func (textPreparedRunFormatter) Format(prepared *domain.PreparedRun) ([]byte, er
 		}
 		sort.Strings(keys)
 		for _, k := range keys {
-			fmt.Fprintf(&b, "    %s: %s\n", k, target.ExtraParams[k])
+			renderedValue, err := formatExtraParamTextValue(target.ExtraParams[k])
+			if err != nil {
+				return nil, fmt.Errorf("failed to format extra_params.%s: %w", k, err)
+			}
+			fmt.Fprintf(&b, "    %s: %s\n", k, renderedValue)
 		}
 	}
 
@@ -148,6 +158,13 @@ func (textPreparedRunFormatter) Format(prepared *domain.PreparedRun) ([]byte, er
 		messages := byRole[role]
 		for i, msg := range messages {
 			fmt.Fprintf(&b, "    - message: %d\n", i+1)
+			if msg.CacheControl != nil {
+				fmt.Fprintf(&b, "      cache_control: %s", msg.CacheControl.Type)
+				if msg.CacheControl.TTL != "" {
+					fmt.Fprintf(&b, " ttl=%s", msg.CacheControl.TTL)
+				}
+				fmt.Fprintln(&b)
+			}
 			fmt.Fprintln(&b, "      content: |")
 			content := msg.Content
 			if content == "" {
@@ -162,3 +179,15 @@ func (textPreparedRunFormatter) Format(prepared *domain.PreparedRun) ([]byte, er
 
 	return b.Bytes(), nil
 }
+
+func formatExtraParamTextValue(value any) (string, error) {
+	if s, ok := value.(string); ok {
+		return s, nil
+	}
+
+	b, err := json.Marshal(value)
+	if err != nil {
+		return "", err
+	}
+	return string(b), nil
+}

internal/format/prepared_run_test.go

@@ -28,6 +28,7 @@ func TestTextFormatterIncludesPreparedRunDetails(t *testing.T) {
 		"max_tokens: 256",
 		"top_p: 0.8",
 		"timeout_seconds: 45",
+		"service_tier: priority",
 		"reasoning_effort: medium",
 		"api_key_env: SCRIPTORIUM_API_KEY",
 		"prompt_hash: prompt-hash",
@@ -48,6 +49,36 @@ func TestTextFormatterIncludesPreparedRunDetails(t *testing.T) {
 	}
 }
 
+func TestTextFormatterRendersExtraParamsDeterministically(t *testing.T) {
+	prepared := samplePreparedRun()
+	prepared.EffectiveModelParams.ExtraParams = map[string]any{
+		"z_string": "enabled",
+		"b_number": 42,
+		"a_object": map[string]any{
+			"nested": "value",
+			"count":  2,
+		},
+		"c_array": []any{"first", 3, false},
+	}
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatText)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	s := string(out)
+
+	want := strings.Join([]string{
+		"  extra_params:",
+		"    a_object: {\"count\":2,\"nested\":\"value\"}",
+		"    b_number: 42",
+		"    c_array: [\"first\",3,false]",
+		"    z_string: enabled",
+	}, "\n")
+	if !strings.Contains(s, want) {
+		t.Fatalf("expected deterministic extra_params block %q, got:\n%s", want, s)
+	}
+}
+
 func TestTextFormatterDoesNotIncludeResolvedAPIKeyValue(t *testing.T) {
 	const secret = "super-secret-api-key"
 	t.Setenv("SCRIPTORIUM_API_KEY", secret)
@@ -61,8 +92,94 @@ func TestTextFormatterDoesNotIncludeResolvedAPIKeyValue(t *testing.T) {
 	}
 }
 
+func TestTextFormatterDoesNotIncludeDirectAPIKeyValue(t *testing.T) {
+	const directKey = "direct-format-key"
+	prepared := samplePreparedRun()
+	prepared.EffectiveModelParams.APIKey = directKey
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatText)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if strings.Contains(string(out), directKey) {
+		t.Fatalf("text output should not include direct api key value: %s", out)
+	}
+}
+
+func TestTextFormatterIncludesMessageCacheControlBeforeContent(t *testing.T) {
+	prepared := samplePreparedRun()
+	prepared.Messages = []domain.RenderedMessage{
+		{
+			Role:    "system",
+			Content: "System guidance.",
+			CacheControl: &domain.CacheControl{
+				Type: domain.CacheControlEphemeral,
+				TTL:  "1h",
+			},
+		},
+		{Role: "user", Content: "Summarize the transcript."},
+	}
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatText)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	s := string(out)
+	if !strings.Contains(s, "  system:\n    - message: 1\n      cache_control: ephemeral ttl=1h\n      content: |") {
+		t.Fatalf("expected system message cache control before content, got:\n%s", s)
+	}
+	if strings.Count(s, "cache_control:") != 1 {
+		t.Fatalf("expected exactly one cache_control line, got:\n%s", s)
+	}
+}
+
+func TestTextFormatterIncludesSessionIDWhenPresent(t *testing.T) {
+	prepared := samplePreparedRun()
+	prepared.SessionID = "session-123"
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatText)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if !strings.Contains(string(out), "session_id: session-123\n") {
+		t.Fatalf("expected session_id in text output, got:\n%s", out)
+	}
+}
+
+func TestTextFormatterOmitsEmptyCacheControlTTL(t *testing.T) {
+	prepared := samplePreparedRun()
+	prepared.Messages = []domain.RenderedMessage{
+		{
+			Role:    "system",
+			Content: "System guidance.",
+			CacheControl: &domain.CacheControl{
+				Type: domain.CacheControlEphemeral,
+			},
+		},
+	}
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatText)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	s := string(out)
+	if !strings.Contains(s, "      cache_control: ephemeral\n") {
+		t.Fatalf("expected cache_control line without ttl, got:\n%s", s)
+	}
+	if strings.Contains(s, "ttl=") {
+		t.Fatalf("expected empty ttl to be omitted, got:\n%s", s)
+	}
+}
+
 func TestJSONFormatterEmitsValidJSONAndIncludesPreparedRunFields(t *testing.T) {
 	prepared := samplePreparedRun()
+	prepared.SessionID = "session-123"
+	prepared.EffectiveModelParams.ExtraParams = map[string]any{
+		"number": 42,
+		"nested": map[string]any{
+			"enabled": true,
+		},
+	}
 
 	out, err := FormatPreparedRun(prepared, PreparedRunFormatJSON)
 	if err != nil {
@@ -86,9 +203,24 @@ func TestJSONFormatterEmitsValidJSONAndIncludesPreparedRunFields(t *testing.T) {
 	if decoded["rendered_prompt_hash"] != "rendered-hash" {
 		t.Fatalf("expected rendered_prompt_hash in json output, got %#v", decoded["rendered_prompt_hash"])
 	}
-	if _, ok := decoded["effective_model_params"]; !ok {
+	if decoded["session_id"] != "session-123" {
+		t.Fatalf("expected session_id in json output, got %#v", decoded["session_id"])
+	}
+	modelParams, ok := decoded["effective_model_params"].(map[string]any)
+	if !ok {
 		t.Fatalf("expected effective_model_params in json output, got %#v", decoded)
 	}
+	extraParams, ok := modelParams["extra_params"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected extra_params in json output, got %#v", modelParams["extra_params"])
+	}
+	if extraParams["number"] != float64(42) {
+		t.Fatalf("unexpected numeric extra param in json output: %#v", extraParams["number"])
+	}
+	nested, ok := extraParams["nested"].(map[string]any)
+	if !ok || nested["enabled"] != true {
+		t.Fatalf("unexpected nested extra param in json output: %#v", extraParams["nested"])
+	}
 	if _, ok := decoded["input_hashes"]; !ok {
 		t.Fatalf("expected input_hashes in json output, got %#v", decoded)
 	}
@@ -97,6 +229,47 @@ func TestJSONFormatterEmitsValidJSONAndIncludesPreparedRunFields(t *testing.T) {
 	}
 }
 
+func TestJSONFormatterIncludesMessageCacheControlOnlyWhenPresent(t *testing.T) {
+	prepared := samplePreparedRun()
+	prepared.Messages = []domain.RenderedMessage{
+		{
+			Role:    "system",
+			Content: "System guidance.",
+			CacheControl: &domain.CacheControl{
+				Type: domain.CacheControlEphemeral,
+				TTL:  "1h",
+			},
+		},
+		{Role: "user", Content: "Summarize the transcript."},
+	}
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatJSON)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+
+	var decoded struct {
+		Messages []map[string]any `json:"messages"`
+	}
+	if err := json.Unmarshal(out, &decoded); err != nil {
+		t.Fatalf("expected valid json output, got %v", err)
+	}
+	if len(decoded.Messages) != 2 {
+		t.Fatalf("expected 2 messages, got %d", len(decoded.Messages))
+	}
+
+	cacheControl, ok := decoded.Messages[0]["cache_control"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected first message cache_control, got %#v", decoded.Messages[0])
+	}
+	if cacheControl["type"] != string(domain.CacheControlEphemeral) || cacheControl["ttl"] != "1h" {
+		t.Fatalf("unexpected cache_control payload: %#v", cacheControl)
+	}
+	if _, ok := decoded.Messages[1]["cache_control"]; ok {
+		t.Fatalf("expected second message to omit cache_control, got %#v", decoded.Messages[1])
+	}
+}
+
 func TestJSONFormatterDoesNotIncludeResolvedAPIKeyValue(t *testing.T) {
 	const secret = "super-secret-api-key"
 	t.Setenv("SCRIPTORIUM_API_KEY", secret)
@@ -110,6 +283,20 @@ func TestJSONFormatterDoesNotIncludeResolvedAPIKeyValue(t *testing.T) {
 	}
 }
 
+func TestJSONFormatterDoesNotIncludeDirectAPIKeyValue(t *testing.T) {
+	const directKey = "direct-format-key"
+	prepared := samplePreparedRun()
+	prepared.EffectiveModelParams.APIKey = directKey
+
+	out, err := FormatPreparedRun(prepared, PreparedRunFormatJSON)
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if strings.Contains(string(out), directKey) {
+		t.Fatalf("json output should not include direct api key value: %s", out)
+	}
+}
+
 func TestParsePreparedRunOutputFormatRecognizesSupportedNames(t *testing.T) {
 	tests := []struct {
 		name  string
@@ -168,6 +355,7 @@ func samplePreparedRun() *domain.PreparedRun {
 			MaxTokens:       256,
 			TopP:            0.8,
 			TimeoutSeconds:  45,
+			ServiceTier:     "priority",
 			ReasoningEffort: "medium",
 			APIKeyEnv:       "SCRIPTORIUM_API_KEY",
 		},

internal/llm/openai_compatible_client.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"strings"
 	"time"
+	"unicode/utf8"
 
 	"gitea.maximumdirect.net/eric/scriptorium/internal/defaults"
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
@@ -35,7 +36,6 @@ type OpenAICompatibleConfig struct {
 type OpenAICompatibleClient struct {
 	baseURL      string
 	defaultModel string
-	timeout      time.Duration
 	httpClient   *http.Client
 }
 
@@ -54,10 +54,11 @@ func NewOpenAICompatibleClient(cfg OpenAICompatibleConfig) (*OpenAICompatibleCli
 
 	var client *http.Client
 	if cfg.HTTPClient != nil {
-		client = cfg.HTTPClient
-		if client.Timeout == 0 {
-			client.Timeout = timeout
+		cloned := *cfg.HTTPClient
+		if cloned.Timeout <= 0 {
+			cloned.Timeout = timeout
 		}
+		client = &cloned
 	} else {
 		client = &http.Client{Timeout: timeout}
 	}
@@ -65,7 +66,6 @@ func NewOpenAICompatibleClient(cfg OpenAICompatibleConfig) (*OpenAICompatibleCli
 	return &OpenAICompatibleClient{
 		baseURL:      strings.TrimRight(baseURL, "/"),
 		defaultModel: cfg.Model,
-		timeout:      timeout,
 		httpClient:   client,
 	}, nil
 }
@@ -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
@@ -92,46 +84,39 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 	}
 	endpoint = strings.TrimRight(endpoint, "/") + defaults.OpenAIChatCompletionsPath
 
-	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 req.StructuredOutput != nil {
-		responseFormat, err := toOpenAIResponseFormat(req.StructuredOutput)
+	wireReq, err := openAIChatRequestFromGenerateRequest(req, c.defaultModel)
 	if err != nil {
 		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
 	}
-		wireReq.ResponseFormat = responseFormat
+
+	wirePayload, err := openAIChatRequestPayload(wireReq)
+	if err != nil {
+		return nil, fmt.Errorf("%w: %v", ErrInvalidRequest, err)
 	}
 
-	payload, err := json.Marshal(wireReq)
+	payload, err := json.Marshal(wirePayload)
 	if err != nil {
 		return nil, fmt.Errorf("%w: failed to encode request: %v", ErrRequestFailed, err)
 	}
 
-	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(payload))
+	requestContext := ctx
+	if req.Target.TimeoutSeconds > 0 {
+		var cancel context.CancelFunc
+		requestContext, cancel = context.WithTimeout(
+			ctx,
+			time.Duration(req.Target.TimeoutSeconds)*time.Second,
+		)
+		defer cancel()
+	}
+
+	httpReq, err := http.NewRequestWithContext(requestContext, http.MethodPost, endpoint, bytes.NewReader(payload))
 	if err != nil {
 		return nil, fmt.Errorf("%w: failed to create request: %v", ErrRequestFailed, err)
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
-	if envName := strings.TrimSpace(req.Target.APIKeyEnv); envName != "" {
+	if apiKey := strings.TrimSpace(req.Target.APIKey); apiKey != "" {
+		httpReq.Header.Set("Authorization", "Bearer "+apiKey)
+	} else if envName := strings.TrimSpace(req.Target.APIKeyEnv); envName != "" {
 		apiKey := strings.TrimSpace(os.Getenv(envName))
 		if apiKey == "" {
 			return nil, fmt.Errorf("%w: api key environment variable %q is not set", ErrInvalidRequest, envName)
@@ -139,18 +124,9 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 		httpReq.Header.Set("Authorization", "Bearer "+apiKey)
 	}
 
-	effectiveTimeout := c.timeout
-	if req.Target.TimeoutSeconds > 0 {
-		effectiveTimeout = time.Duration(req.Target.TimeoutSeconds) * time.Second
-	}
-
 	httpClient := c.httpClient
 	if httpClient == nil {
-		httpClient = &http.Client{Timeout: effectiveTimeout}
-	} else if httpClient.Timeout != effectiveTimeout {
-		cloned := *httpClient
-		cloned.Timeout = effectiveTimeout
-		httpClient = &cloned
+		httpClient = &http.Client{Timeout: defaults.LLMRequestTimeoutDefault}
 	}
 
 	httpResp, err := httpClient.Do(httpReq)
@@ -160,8 +136,8 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 	defer httpResp.Body.Close()
 
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		body, _ := io.ReadAll(io.LimitReader(httpResp.Body, 4096))
-		return nil, fmt.Errorf("%w: status=%d body=%q", ErrUnexpectedStatus, httpResp.StatusCode, strings.TrimSpace(string(body)))
+		_, _ = io.Copy(io.Discard, io.LimitReader(httpResp.Body, 4096))
+		return nil, fmt.Errorf("%w: status=%d", ErrUnexpectedStatus, httpResp.StatusCode)
 	}
 
 	var wireResp openAIChatResponse
@@ -183,32 +159,166 @@ func (c *OpenAICompatibleClient) Generate(ctx context.Context, req domain.Genera
 			PromptTokens:     wireResp.Usage.PromptTokens,
 			CompletionTokens: wireResp.Usage.CompletionTokens,
 			TotalTokens:      wireResp.Usage.TotalTokens,
+			CachedTokens:     wireResp.Usage.PromptTokensDetails.CachedTokens,
+			CacheWriteTokens: wireResp.Usage.CacheWriteTokens,
 		},
 	}, 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,
+	}
+	if sessionID := strings.TrimSpace(req.Prompt.SessionID); sessionID != "" {
+		if n := utf8.RuneCountInString(sessionID); n > domain.SessionIDMaxLength {
+			return openAIChatRequest{}, fmt.Errorf("session_id length %d exceeds maximum %d", n, domain.SessionIDMaxLength)
+		}
+		wireReq.SessionID = sessionID
+	}
+
+	wireReq.Messages = make([]openAIChatRequestMessage, 0, len(req.Prompt.Messages))
+	for _, msg := range req.Prompt.Messages {
+		wireReq.Messages = append(wireReq.Messages, openAIChatRequestMessageFromRenderedMessage(msg))
+	}
+
+	if req.Target.Temperature != 0 || req.TargetPresence.Temperature {
+		wireReq.Temperature = &req.Target.Temperature
+	}
+	if req.Target.MaxTokens != 0 || req.TargetPresence.MaxTokens {
+		wireReq.MaxTokens = &req.Target.MaxTokens
+	}
+	if req.Target.TopP != 0 || req.TargetPresence.TopP {
+		wireReq.TopP = &req.Target.TopP
+	}
+	if strings.TrimSpace(req.Target.ServiceTier) != "" {
+		wireReq.ServiceTier = req.Target.ServiceTier
+	}
+	if strings.TrimSpace(req.Target.ReasoningEffort) != "" {
+		wireReq.ReasoningEffort = req.Target.ReasoningEffort
+	}
+	if len(req.Target.ExtraParams) > 0 {
+		wireReq.ExtraParams = req.Target.ExtraParams
+	}
+	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"`
+	SessionID       string                     `json:"session_id,omitempty"`
+	Messages        []openAIChatRequestMessage `json:"messages"`
 	Temperature     *float64                   `json:"temperature,omitempty"`
 	MaxTokens       *int                       `json:"max_tokens,omitempty"`
 	TopP            *float64                   `json:"top_p,omitempty"`
+	ServiceTier     string                     `json:"service_tier,omitempty"`
+	ReasoningEffort string                     `json:"reasoning_effort,omitempty"`
 	ResponseFormat  *openAIResponseFormat      `json:"response_format,omitempty"`
+	ExtraParams     map[string]any             `json:"-"`
 }
 
-type openAIChatMessage struct {
+func openAIChatRequestPayload(req openAIChatRequest) (map[string]any, error) {
+	out := map[string]any{
+		"model":    req.Model,
+		"messages": req.Messages,
+	}
+	if req.SessionID != "" {
+		out["session_id"] = req.SessionID
+	}
+	if req.Temperature != nil {
+		out["temperature"] = *req.Temperature
+	}
+	if req.MaxTokens != nil {
+		out["max_tokens"] = *req.MaxTokens
+	}
+	if req.TopP != nil {
+		out["top_p"] = *req.TopP
+	}
+	if req.ServiceTier != "" {
+		out["service_tier"] = req.ServiceTier
+	}
+	if req.ReasoningEffort != "" {
+		out["reasoning_effort"] = req.ReasoningEffort
+	}
+	if req.ResponseFormat != nil {
+		out["response_format"] = req.ResponseFormat
+	}
+
+	for key, value := range req.ExtraParams {
+		if key == "" {
+			return nil, errors.New("extra_params key must not be empty")
+		}
+		if _, reserved := reservedOpenAIChatRequestFields[key]; reserved {
+			return nil, fmt.Errorf("extra_params key %q collides with reserved request field", key)
+		}
+		if _, err := json.Marshal(value); err != nil {
+			return nil, fmt.Errorf("extra_params.%s must be JSON-serializable: %w", key, err)
+		}
+		out[key] = value
+	}
+
+	return out, nil
+}
+
+var reservedOpenAIChatRequestFields = map[string]struct{}{
+	"model":            {},
+	"session_id":       {},
+	"messages":         {},
+	"temperature":      {},
+	"max_tokens":       {},
+	"top_p":            {},
+	"service_tier":     {},
+	"reasoning_effort": {},
+	"response_format":  {},
+}
+
+type openAIChatRequestMessage struct {
+	Role    string `json:"role"`
+	Content any    `json:"content"`
+}
+
+type openAIChatTextContentBlock struct {
+	Type         string              `json:"type"`
+	Text         string              `json:"text"`
+	CacheControl *openAICacheControl `json:"cache_control,omitempty"`
+}
+
+type openAICacheControl struct {
+	Type string `json:"type"`
+	TTL  string `json:"ttl,omitempty"`
+}
+
+type openAIChatResponseMessage struct {
 	Role    string `json:"role"`
 	Content string `json:"content"`
 }
 
 type openAIChatResponse struct {
 	Choices []struct {
-		Message openAIChatMessage `json:"message"`
+		Message openAIChatResponseMessage `json:"message"`
 	} `json:"choices"`
 	Usage struct {
 		PromptTokens        int `json:"prompt_tokens"`
 		CompletionTokens    int `json:"completion_tokens"`
 		TotalTokens         int `json:"total_tokens"`
+		PromptTokensDetails struct {
+			CachedTokens int `json:"cached_tokens"`
+		} `json:"prompt_tokens_details"`
+		CacheWriteTokens int `json:"cache_write_tokens"`
 	} `json:"usage"`
 }
 
@@ -223,6 +333,28 @@ type openAIJSONSchemaEnvelope struct {
 	Schema any    `json:"schema"`
 }
 
+func openAIChatRequestMessageFromRenderedMessage(msg domain.RenderedMessage) openAIChatRequestMessage {
+	wireMsg := openAIChatRequestMessage{
+		Role:    msg.Role,
+		Content: msg.Content,
+	}
+	if msg.CacheControl == nil {
+		return wireMsg
+	}
+
+	wireMsg.Content = []openAIChatTextContentBlock{
+		{
+			Type: "text",
+			Text: msg.Content,
+			CacheControl: &openAICacheControl{
+				Type: string(msg.CacheControl.Type),
+				TTL:  msg.CacheControl.TTL,
+			},
+		},
+	}
+	return wireMsg
+}
+
 func toOpenAIResponseFormat(spec *domain.StructuredOutputSpec) (*openAIResponseFormat, error) {
 	if spec == nil {
 		return nil, nil

internal/llm/openai_compatible_client_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"math"
 	"net/http"
 	"net/http/httptest"
 	"strings"
@@ -13,6 +14,91 @@ import (
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
 )
 
+func TestNewOpenAICompatibleClientDoesNotMutateSuppliedZeroTimeoutClient(t *testing.T) {
+	transport := http.DefaultTransport
+	supplied := &http.Client{Transport: transport}
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		HTTPClient: supplied,
+	})
+	if err != nil {
+		t.Fatalf("unexpected constructor error: %v", err)
+	}
+
+	if supplied.Timeout != 0 {
+		t.Fatalf("expected supplied client timeout to remain zero, got %v", supplied.Timeout)
+	}
+	if client.httpClient == supplied {
+		t.Fatal("expected constructed client to use a cloned HTTP client")
+	}
+	if client.httpClient.Timeout <= 0 {
+		t.Fatalf("expected constructed client to use a positive default timeout, got %v", client.httpClient.Timeout)
+	}
+	if client.httpClient.Transport != transport {
+		t.Fatal("expected cloned client to preserve the supplied transport")
+	}
+}
+
+func TestNewOpenAICompatibleClientDoesNotMutateSuppliedNonzeroTimeoutClient(t *testing.T) {
+	transport := http.DefaultTransport
+	suppliedTimeout := 37 * time.Second
+	supplied := &http.Client{
+		Timeout:   suppliedTimeout,
+		Transport: transport,
+	}
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		Timeout:    2 * time.Second,
+		HTTPClient: supplied,
+	})
+	if err != nil {
+		t.Fatalf("unexpected constructor error: %v", err)
+	}
+
+	if supplied.Timeout != suppliedTimeout {
+		t.Fatalf("expected supplied client timeout to remain %v, got %v", suppliedTimeout, supplied.Timeout)
+	}
+	if client.httpClient == supplied {
+		t.Fatal("expected constructed client to use a cloned HTTP client")
+	}
+	if client.httpClient.Timeout != suppliedTimeout {
+		t.Fatalf("expected cloned client timeout %v, got %v", suppliedTimeout, client.httpClient.Timeout)
+	}
+	if client.httpClient.Transport != transport {
+		t.Fatal("expected cloned client to preserve the supplied transport")
+	}
+}
+
+func TestNewOpenAICompatibleClientTreatsSuppliedNegativeTimeoutAsUnset(t *testing.T) {
+	transport := http.DefaultTransport
+	supplied := &http.Client{
+		Timeout:   -time.Second,
+		Transport: transport,
+	}
+	configuredTimeout := 23 * time.Second
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		Timeout:    configuredTimeout,
+		HTTPClient: supplied,
+	})
+	if err != nil {
+		t.Fatalf("unexpected constructor error: %v", err)
+	}
+
+	if supplied.Timeout != -time.Second {
+		t.Fatalf("expected supplied client timeout to remain negative, got %v", supplied.Timeout)
+	}
+	if client.httpClient == supplied {
+		t.Fatal("expected constructed client to use a cloned HTTP client")
+	}
+	if client.httpClient.Timeout != configuredTimeout {
+		t.Fatalf("expected cloned client timeout %v, got %v", configuredTimeout, client.httpClient.Timeout)
+	}
+	if client.httpClient.Transport != transport {
+		t.Fatal("expected cloned client to preserve the supplied transport")
+	}
+}
+
 func TestOpenAICompatibleClientGenerateSuccess(t *testing.T) {
 	type observedRequest struct {
 		Authorization string
@@ -61,6 +147,7 @@ func TestOpenAICompatibleClientGenerateSuccess(t *testing.T) {
 			Temperature: 0.4,
 			MaxTokens:   123,
 			TopP:        0.7,
+			ServiceTier: "priority",
 			APIKeyEnv:   "SCRIPTORIUM_TEST_API_KEY",
 		},
 		StructuredOutput: &domain.StructuredOutputSpec{
@@ -88,6 +175,9 @@ func TestOpenAICompatibleClientGenerateSuccess(t *testing.T) {
 	if resp.Usage.PromptTokens != 11 || resp.Usage.CompletionTokens != 22 || resp.Usage.TotalTokens != 33 {
 		t.Fatalf("unexpected usage: %+v", resp.Usage)
 	}
+	if resp.Usage.CachedTokens != 0 || resp.Usage.CacheWriteTokens != 0 {
+		t.Fatalf("expected absent cache usage fields to remain zero, got %+v", resp.Usage)
+	}
 
 	if obs.Authorization != "Bearer secret-key" {
 		t.Fatalf("unexpected Authorization header: %q", obs.Authorization)
@@ -95,6 +185,18 @@ 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"])
+	}
 
 	msgs, ok := obs.Body["messages"].([]any)
 	if !ok || len(msgs) != 2 {
@@ -131,6 +233,277 @@ func TestOpenAICompatibleClientGenerateSuccess(t *testing.T) {
 	}
 }
 
+func TestOpenAICompatibleClientDirectAPIKeyPreferredOverEnv(t *testing.T) {
+	const directKey = "direct-llm-key"
+	t.Setenv("SCRIPTORIUM_TEST_API_KEY", "env-key")
+
+	var gotAuth string
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotAuth = r.Header.Get("Authorization")
+		_, _ = 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",
+			APIKeyEnv: "SCRIPTORIUM_TEST_API_KEY",
+			APIKey:    directKey,
+		},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if gotAuth != "Bearer "+directKey {
+		t.Fatalf("unexpected Authorization header: %q", gotAuth)
+	}
+}
+
+func TestOpenAICompatibleClientSerializesCacheControlledMessageAsContentBlock(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:    "system",
+				Content: "Stable instructions.",
+				CacheControl: &domain.CacheControl{
+					Type: domain.CacheControlEphemeral,
+					TTL:  "1h",
+				},
+			},
+			{Role: "user", Content: "Dynamic request."},
+		}},
+		Target: domain.ExecutionTarget{Model: "model"},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+
+	for _, forbidden := range []string{"cache_control", "extra_params"} {
+		if _, exists := observedBody[forbidden]; exists {
+			t.Fatalf("expected top-level %s to be omitted, got %#v", forbidden, observedBody[forbidden])
+		}
+	}
+
+	msgs, ok := observedBody["messages"].([]any)
+	if !ok || len(msgs) != 2 {
+		t.Fatalf("unexpected messages payload: %#v", observedBody["messages"])
+	}
+	msg0 := msgs[0].(map[string]any)
+	if msg0["role"] != "system" {
+		t.Fatalf("unexpected first message role: %#v", msg0["role"])
+	}
+	contentBlocks, ok := msg0["content"].([]any)
+	if !ok || len(contentBlocks) != 1 {
+		t.Fatalf("expected first message content block array, got %#v", msg0["content"])
+	}
+	block := contentBlocks[0].(map[string]any)
+	if block["type"] != "text" || block["text"] != "Stable instructions." {
+		t.Fatalf("unexpected text content block: %#v", block)
+	}
+	cacheControl, ok := block["cache_control"].(map[string]any)
+	if !ok {
+		t.Fatalf("expected cache_control on content block, got %#v", block)
+	}
+	if cacheControl["type"] != string(domain.CacheControlEphemeral) || cacheControl["ttl"] != "1h" {
+		t.Fatalf("unexpected cache_control payload: %#v", cacheControl)
+	}
+
+	msg1 := msgs[1].(map[string]any)
+	if msg1["role"] != "user" || msg1["content"] != "Dynamic request." {
+		t.Fatalf("expected uncached message to keep string content, got %#v", msg1)
+	}
+}
+
+func TestOpenAICompatibleClientOmitsEmptyCacheControlTTL(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:    "system",
+				Content: "Stable instructions.",
+				CacheControl: &domain.CacheControl{
+					Type: domain.CacheControlEphemeral,
+				},
+			},
+		}},
+		Target: domain.ExecutionTarget{Model: "model"},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+
+	msgs := observedBody["messages"].([]any)
+	msg0 := msgs[0].(map[string]any)
+	contentBlocks := msg0["content"].([]any)
+	block := contentBlocks[0].(map[string]any)
+	cacheControl := block["cache_control"].(map[string]any)
+	if cacheControl["type"] != string(domain.CacheControlEphemeral) {
+		t.Fatalf("unexpected cache_control type: %#v", cacheControl)
+	}
+	if _, exists := cacheControl["ttl"]; exists {
+		t.Fatalf("expected empty ttl to be omitted, got %#v", cacheControl)
+	}
+}
+
+func TestOpenAICompatibleClientSerializesSessionID(t *testing.T) {
+	var observedBody map[string]any
+	var observedSessionHeader string
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		observedSessionHeader = r.Header.Get("x-session-id")
+		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{
+			SessionID: " session-123 ",
+			Messages:  []domain.RenderedMessage{{Role: "user", Content: "hi"}},
+		},
+		Target: domain.ExecutionTarget{Model: "model"},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if observedBody["session_id"] != "session-123" {
+		t.Fatalf("expected top-level session_id, got %#v", observedBody["session_id"])
+	}
+	if observedSessionHeader != "" {
+		t.Fatalf("did not expect x-session-id header, got %q", observedSessionHeader)
+	}
+}
+
+func TestOpenAICompatibleClientOmitsEmptySessionID(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{
+			SessionID: "   ",
+			Messages:  []domain.RenderedMessage{{Role: "user", Content: "hi"}},
+		},
+		Target: domain.ExecutionTarget{Model: "model"},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if _, exists := observedBody["session_id"]; exists {
+		t.Fatalf("expected empty session_id to be omitted, got %#v", observedBody["session_id"])
+	}
+}
+
+func TestOpenAICompatibleClientRejectsTooLongSessionID(t *testing.T) {
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		BaseURL: "http://example.com/v1",
+		Model:   "model",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	_, err = client.Generate(context.Background(), domain.GenerateRequest{
+		Prompt: domain.RenderedPrompt{
+			SessionID: strings.Repeat("x", domain.SessionIDMaxLength+1),
+			Messages:  []domain.RenderedMessage{{Role: "user", Content: "hi"}},
+		},
+	})
+	if err == nil {
+		t.Fatal("expected invalid request error")
+	}
+	if !errors.Is(err, ErrInvalidRequest) {
+		t.Fatalf("expected ErrInvalidRequest, got %v", err)
+	}
+}
+
+func TestOpenAICompatibleClientParsesCacheUsage(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = w.Write([]byte(`{
+  "choices": [{"message": {"role": "assistant", "content": "ok"}}],
+  "usage": {
+    "prompt_tokens": 100,
+    "completion_tokens": 20,
+    "total_tokens": 120,
+    "prompt_tokens_details": {"cached_tokens": 80},
+    "cache_write_tokens": 60
+  }
+}`))
+	}))
+	defer ts.Close()
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{BaseURL: ts.URL + "/v1", Model: "model"})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	resp, err := client.Generate(context.Background(), domain.GenerateRequest{
+		Prompt: domain.RenderedPrompt{Messages: []domain.RenderedMessage{{Role: "user", Content: "hi"}}},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if resp.Usage.PromptTokens != 100 || resp.Usage.CompletionTokens != 20 || resp.Usage.TotalTokens != 120 {
+		t.Fatalf("unexpected base usage fields: %+v", resp.Usage)
+	}
+	if resp.Usage.CachedTokens != 80 || resp.Usage.CacheWriteTokens != 60 {
+		t.Fatalf("unexpected cache usage fields: %+v", resp.Usage)
+	}
+}
+
 func TestOpenAICompatibleClientOmitsResponseFormatWhenNoStructuredOutput(t *testing.T) {
 	var observedBody map[string]any
 	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -157,6 +530,259 @@ func TestOpenAICompatibleClientOmitsResponseFormatWhenNoStructuredOutput(t *test
 	if _, exists := observedBody["response_format"]; exists {
 		t.Fatalf("expected response_format omitted, got %#v", observedBody["response_format"])
 	}
+	if _, exists := observedBody["service_tier"]; exists {
+		t.Fatalf("expected service_tier omitted, got %#v", observedBody["service_tier"])
+	}
+}
+
+func TestOpenAICompatibleClientSerializesReasoningEffortAndExtraParams(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]any{
+				"string_value":  "on",
+				"number_value":  42,
+				"boolean_value": true,
+				"object_value":  map[string]any{"nested": "value", "count": 2},
+				"array_value":   []any{"first", 3, false},
+			},
+		},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if observedBody["reasoning_effort"] != "high" {
+		t.Fatalf("expected reasoning_effort high, got %#v", observedBody["reasoning_effort"])
+	}
+	if observedBody["string_value"] != "on" {
+		t.Fatalf("unexpected string extra param: %#v", observedBody["string_value"])
+	}
+	if observedBody["number_value"] != float64(42) {
+		t.Fatalf("unexpected number extra param: %#v", observedBody["number_value"])
+	}
+	if observedBody["boolean_value"] != true {
+		t.Fatalf("unexpected boolean extra param: %#v", observedBody["boolean_value"])
+	}
+	objectValue, ok := observedBody["object_value"].(map[string]any)
+	if !ok || objectValue["nested"] != "value" || objectValue["count"] != float64(2) {
+		t.Fatalf("unexpected object extra param: %#v", observedBody["object_value"])
+	}
+	if _, exists := observedBody["extra_params"]; exists {
+		t.Fatalf("expected extra_params wrapper omitted, got %#v", observedBody["extra_params"])
+	}
+	arrayValue, ok := observedBody["array_value"].([]any)
+	if !ok || len(arrayValue) != 3 || arrayValue[0] != "first" || arrayValue[1] != float64(3) || arrayValue[2] != false {
+		t.Fatalf("unexpected array extra param: %#v", observedBody["array_value"])
+	}
+}
+
+func TestOpenAICompatibleClientOmitsReasoningEffortWhenUnset(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"},
+	})
+	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 wrapper omitted, got %#v", observedBody["extra_params"])
+	}
+}
+
+func TestOpenAICompatibleClientSerializesExplicitZeroNumericOverrides(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"},
+		TargetPresence: domain.ExecutionTargetPresence{
+			Temperature: true,
+			MaxTokens:   true,
+			TopP:        true,
+		},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	if observedBody["temperature"] != float64(0) {
+		t.Fatalf("expected explicit zero temperature, got %#v", observedBody["temperature"])
+	}
+	if observedBody["max_tokens"] != float64(0) {
+		t.Fatalf("expected explicit zero max_tokens, got %#v", observedBody["max_tokens"])
+	}
+	if observedBody["top_p"] != float64(0) {
+		t.Fatalf("expected explicit zero top_p, got %#v", observedBody["top_p"])
+	}
+}
+
+func TestOpenAICompatibleClientOmitsImplicitZeroNumericFields(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"},
+	})
+	if err != nil {
+		t.Fatalf("expected no error, got %v", err)
+	}
+	for _, field := range []string{"temperature", "max_tokens", "top_p"} {
+		if _, exists := observedBody[field]; exists {
+			t.Fatalf("expected implicit zero field %q to be omitted, got body %#v", field, observedBody)
+		}
+	}
+}
+
+func TestOpenAICompatibleClientOmittedTimeoutUsesClientTimeout(t *testing.T) {
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(20 * time.Millisecond)
+		_, _ = w.Write([]byte(`{"choices":[{"message":{"content":"ok"}}]}`))
+	}))
+	defer ts.Close()
+
+	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
+		BaseURL: ts.URL + "/v1",
+		Timeout: time.Nanosecond,
+	})
+	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", TimeoutSeconds: 0},
+	})
+	if err == nil {
+		t.Fatal("expected omitted timeout to use client timeout")
+	}
+	if !errors.Is(err, ErrRequestFailed) {
+		t.Fatalf("expected ErrRequestFailed, got %v", err)
+	}
+}
+
+func TestOpenAICompatibleClientRejectsInvalidExtraParamsBeforeProviderCall(t *testing.T) {
+	tests := []struct {
+		name        string
+		extraParams map[string]any
+		want        string
+	}{
+		{name: "empty key", extraParams: map[string]any{"": "empty"}, want: "key must not be empty"},
+		{name: "unserializable value", extraParams: map[string]any{"bad": math.Inf(1)}, want: "JSON-serializable"},
+	}
+	for _, key := range []string{
+		"model",
+		"session_id",
+		"messages",
+		"temperature",
+		"max_tokens",
+		"top_p",
+		"service_tier",
+		"reasoning_effort",
+		"response_format",
+	} {
+		tests = append(tests, struct {
+			name        string
+			extraParams map[string]any
+			want        string
+		}{
+			name:        "reserved key " + key,
+			extraParams: map[string]any{key: "collision"},
+			want:        "reserved request field",
+		})
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			called := false
+			ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				called = true
+				_, _ = 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", ExtraParams: tc.extraParams},
+			})
+			if err == nil {
+				t.Fatal("expected invalid request error")
+			}
+			if !errors.Is(err, ErrInvalidRequest) {
+				t.Fatalf("expected ErrInvalidRequest, got %v", err)
+			}
+			if !strings.Contains(err.Error(), tc.want) {
+				t.Fatalf("expected error to contain %q, got %v", tc.want, err)
+			}
+			if called {
+				t.Fatal("provider should not be called for invalid extra_params")
+			}
+		})
+	}
 }
 
 func TestOpenAICompatibleClientNoAuthorizationHeaderWhenNoAPIKey(t *testing.T) {
@@ -279,9 +905,10 @@ func TestOpenAICompatibleClientEndpointOverride(t *testing.T) {
 }
 
 func TestOpenAICompatibleClientNon2xxError(t *testing.T) {
+	const sensitiveBody = `provider-secret-fragment request_payload_details`
 	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.WriteHeader(http.StatusBadRequest)
-		_, _ = w.Write([]byte(`{"error":"bad request payload"}`))
+		_, _ = w.Write([]byte(`{"error":"` + sensitiveBody + `"}`))
 	}))
 	defer ts.Close()
 
@@ -299,8 +926,11 @@ func TestOpenAICompatibleClientNon2xxError(t *testing.T) {
 	if !errors.Is(err, ErrUnexpectedStatus) {
 		t.Fatalf("expected ErrUnexpectedStatus, got %v", err)
 	}
-	if !strings.Contains(err.Error(), "400") || !strings.Contains(err.Error(), "bad request payload") {
-		t.Fatalf("expected status/body details, got %v", err)
+	if !strings.Contains(err.Error(), "status=400") {
+		t.Fatalf("expected status detail, got %v", err)
+	}
+	if strings.Contains(err.Error(), sensitiveBody) {
+		t.Fatalf("expected provider response body to be redacted, got %v", err)
 	}
 }
 
@@ -375,34 +1005,6 @@ func TestOpenAICompatibleClientTimeout(t *testing.T) {
 	}
 }
 
-func TestOpenAICompatibleClientRequestTimeoutOverride(t *testing.T) {
-	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		time.Sleep(100 * time.Millisecond)
-		_, _ = w.Write([]byte(`{"choices":[{"message":{"content":"ok"}}]}`))
-	}))
-	defer ts.Close()
-
-	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
-		BaseURL: ts.URL + "/v1",
-		Model:   "m",
-		Timeout: 50 * time.Millisecond,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	resp, err := client.Generate(context.Background(), domain.GenerateRequest{
-		Prompt: domain.RenderedPrompt{Messages: []domain.RenderedMessage{{Role: "user", Content: "hi"}}},
-		Target: domain.ExecutionTarget{TimeoutSeconds: 1},
-	})
-	if err != nil {
-		t.Fatalf("expected request-level timeout override to succeed, got %v", err)
-	}
-	if resp.Content != "ok" {
-		t.Fatalf("expected response content ok, got %q", resp.Content)
-	}
-}
-
 func TestOpenAICompatibleClientNegativeTimeoutRejected(t *testing.T) {
 	client, err := NewOpenAICompatibleClient(OpenAICompatibleConfig{
 		BaseURL: "http://example.com/v1",

internal/profile/builtin/assets/aion-labs/aion-2.yml

@@ -0,0 +1,9 @@
+id: aion-2
+endpoint: https://openrouter.ai/api/v1
+model: aion-labs/aion-2.0
+temperature: 0.72
+reasoning_effort: high
+top_p: 0.95
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/anthropic/claude-fable-latest.yml

@@ -0,0 +1,7 @@
+id: claude-fable-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~anthropic/claude-fable-latest"
+reasoning_effort: high
+timeout_seconds: 600
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/anthropic/claude-haiku-latest.yml

@@ -0,0 +1,7 @@
+id: claude-haiku-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~anthropic/claude-haiku-latest"
+reasoning_effort: medium
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/anthropic/claude-opus-latest.yml

@@ -0,0 +1,7 @@
+id: claude-opus-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~anthropic/claude-opus-latest"
+reasoning_effort: high
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/anthropic/claude-sonnet-latest.yml

@@ -0,0 +1,7 @@
+id: claude-sonnet-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~anthropic/claude-sonnet-latest"
+reasoning_effort: high
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/deepseek/deepseek-3-2.yml

@@ -0,0 +1,7 @@
+id: deepseek-3-2
+endpoint: https://openrouter.ai/api/v1
+model: deepseek/deepseek-v3.2
+reasoning_effort: high
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/deepseek/deepseek-4-flash.yml

@@ -0,0 +1,7 @@
+id: deepseek-4-flash
+endpoint: https://openrouter.ai/api/v1
+model: deepseek/deepseek-v4-flash
+#reasoning_effort: medium
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/deepseek/deepseek-4-pro.yml

@@ -0,0 +1,7 @@
+id: deepseek-4-pro
+endpoint: https://openrouter.ai/api/v1
+model: deepseek/deepseek-v4-pro
+reasoning_effort: high
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-2-flash-lite.yml

@@ -0,0 +1,9 @@
+id: gemini-2-flash-lite
+endpoint: https://openrouter.ai/api/v1
+model: "google/gemini-2.5-flash-lite"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-2-flash.yml

@@ -0,0 +1,9 @@
+id: gemini-2-flash
+endpoint: https://openrouter.ai/api/v1
+model: "google/gemini-2.5-flash"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-2-pro.yml

@@ -0,0 +1,9 @@
+id: gemini-2-pro
+endpoint: https://openrouter.ai/api/v1
+model: "google/gemini-2.5-pro"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-3-flash-lite.yml

@@ -0,0 +1,9 @@
+id: gemini-3-flash-lite
+endpoint: https://openrouter.ai/api/v1
+model: "google/gemini-3.1-flash-lite"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-flash-latest.yml

@@ -0,0 +1,9 @@
+id: gemini-flash-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~google/gemini-flash-latest"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/google/gemini-pro-latest.yml

@@ -0,0 +1,9 @@
+id: gemini-pro-latest
+endpoint: https://openrouter.ai/api/v1
+model: "~google/gemini-pro-latest"
+#temperature: 0.15
+reasoning_effort: high
+#top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

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

@@ -0,0 +1,9 @@
+id: gemma-4-31b
+endpoint: https://openrouter.ai/api/v1
+model: google/gemma-4-31b-it:exacto
+temperature: 0.15
+reasoning_effort: high
+top_p: 0.98
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/minimax/minimax-m2.yml

@@ -0,0 +1,9 @@
+id: minimax-m2
+endpoint: https://openrouter.ai/api/v1
+model: minimax/minimax-m2.5
+temperature: 0.5
+reasoning_effort: high
+top_p: 0.95
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/minimax/minimax-m3.yml

@@ -0,0 +1,9 @@
+id: minimax-m3
+endpoint: https://openrouter.ai/api/v1
+model: minimax/minimax-m3
+#temperature: 0.5
+reasoning_effort: high
+#top_p: 0.95
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/mistral/mistral-large-2512.yml

@@ -0,0 +1,7 @@
+id: mistral-large-2512
+endpoint: https://openrouter.ai/api/v1
+model: mistralai/mistral-large-2512
+temperature: 0.15
+top_p: 0.98
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY

internal/profile/builtin/assets/mistral/mistral-medium-3-5.yml

@@ -0,0 +1,8 @@
+id: mistral-medium-3-5
+endpoint: https://openrouter.ai/api/v1
+model: mistralai/mistral-medium-3-5
+temperature: 0.15
+reasoning_effort: high
+top_p: 0.98
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY

internal/profile/builtin/assets/mistral/mistral-small-3.yml

@@ -0,0 +1,7 @@
+id: mistral-small-3
+endpoint: https://openrouter.ai/api/v1
+model: mistralai/mistral-small-3.2-24b-instruct
+temperature: 0.05
+top_p: 1.0
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY

internal/profile/builtin/assets/mistral/mistral-small-4.yml

@@ -0,0 +1,8 @@
+id: mistral-small-4
+endpoint: https://openrouter.ai/api/v1
+model: mistralai/mistral-small-2603
+temperature: 0.1
+reasoning_effort: high
+top_p: 0.98
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY

internal/profile/builtin/assets/nvidia/nemotron-3-ultra.yml

@@ -0,0 +1,7 @@
+id: nemotron-3-ultra
+endpoint: https://openrouter.ai/api/v1
+model: nvidia/nemotron-3-ultra-550b-a55b
+reasoning_effort: high
+timeout_seconds: 180
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/openai/gpt-5-mini.yml

@@ -0,0 +1,7 @@
+id: gpt-5-mini
+endpoint: https://openrouter.ai/api/v1
+model: "openai/gpt-5.4-mini"
+reasoning_effort: high
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/assets/openai/gpt-5-nano.yml

@@ -0,0 +1,7 @@
+id: gpt-5-nano
+endpoint: https://openrouter.ai/api/v1
+model: "openai/gpt-5.4-nano"
+reasoning_effort: high
+timeout_seconds: 240
+api_key_env: OPENROUTER_API_KEY
+service_tier: flex

internal/profile/builtin/repository.go

@@ -0,0 +1,31 @@
+package builtin
+
+import (
+	"embed"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
+)
+
+const assetRoot = "assets"
+
+//go:embed assets/**/*.yml
+var assets embed.FS
+
+func NewRepository() profile.Repository {
+	return profile.NewFSRepository(assets, assetRoot)
+}
+
+func NewRepositoryWithPrimary(primary profile.Repository) profile.Repository {
+	if primary == nil {
+		return NewRepository()
+	}
+	return profile.NewOverlayRepository(primary, NewRepository())
+}
+
+func NewRepositoryWithDirectory(dir string) profile.Repository {
+	if strings.TrimSpace(dir) == "" {
+		return NewRepository()
+	}
+	return NewRepositoryWithPrimary(profile.NewFilesystemRepository(dir))
+}

internal/profile/builtin/repository_test.go

@@ -0,0 +1,127 @@
+package builtin
+
+import (
+	"context"
+	"errors"
+	"io/fs"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/profile"
+	"gopkg.in/yaml.v3"
+)
+
+func TestBuiltInProfilesValidateThroughRepository(t *testing.T) {
+	repo := NewRepository()
+	ids := loadBuiltInProfileIDs(t)
+	if len(ids) == 0 {
+		t.Fatal("expected built-in profiles")
+	}
+
+	for id := range ids {
+		t.Run(id, func(t *testing.T) {
+			p, err := repo.GetProfile(context.Background(), id)
+			if err != nil {
+				t.Fatalf("expected built-in profile %q to load, got %v", id, err)
+			}
+			if p.ID != id {
+				t.Fatalf("expected profile id %q, got %q", id, p.ID)
+			}
+		})
+	}
+}
+
+func TestBuiltInProfilesDoNotContainDuplicateIDsOrRawAPIKeys(t *testing.T) {
+	loadBuiltInProfileIDs(t)
+}
+
+func loadBuiltInProfileIDs(t *testing.T) map[string]string {
+	t.Helper()
+
+	ids := map[string]string{}
+	err := fs.WalkDir(assets, assetRoot, func(name string, d fs.DirEntry, err error) error {
+		if err != nil {
+			return err
+		}
+		if d.IsDir() || !strings.HasSuffix(name, ".yml") {
+			return nil
+		}
+
+		data, err := assets.ReadFile(name)
+		if err != nil {
+			t.Fatalf("failed to read built-in profile %s: %v", name, err)
+		}
+
+		var raw map[string]any
+		if err := yaml.Unmarshal(data, &raw); err != nil {
+			t.Fatalf("failed to decode built-in profile %s: %v", name, err)
+		}
+		if _, ok := raw["api_key"]; ok {
+			t.Fatalf("built-in profile %s contains raw api_key", name)
+		}
+		id, ok := raw["id"].(string)
+		if !ok || strings.TrimSpace(id) == "" {
+			t.Fatalf("built-in profile %s has missing id", name)
+		}
+		if previous, ok := ids[id]; ok {
+			t.Fatalf("duplicate built-in profile id %q in %s and %s", id, previous, name)
+		}
+		ids[id] = name
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("failed to walk built-in profiles: %v", err)
+	}
+	return ids
+}
+
+func TestRepositoryWithPrimaryUsesPrimaryBeforeBuiltIns(t *testing.T) {
+	repo := NewRepositoryWithPrimary(staticProfileRepo{
+		profiles: map[string]string{"mistral-small-3": "custom-model"},
+	})
+
+	p, err := repo.GetProfile(context.Background(), "mistral-small-3")
+	if err != nil {
+		t.Fatalf("expected profile to load, got %v", err)
+	}
+	if p.Model != "custom-model" {
+		t.Fatalf("expected primary profile to override built-in, got %+v", p)
+	}
+}
+
+func TestRepositoryWithPrimaryFallsBackToBuiltIns(t *testing.T) {
+	repo := NewRepositoryWithPrimary(staticProfileRepo{})
+
+	p, err := repo.GetProfile(context.Background(), "mistral-small-3")
+	if err != nil {
+		t.Fatalf("expected built-in profile to load, got %v", err)
+	}
+	if p.ID != "mistral-small-3" {
+		t.Fatalf("unexpected profile: %+v", p)
+	}
+}
+
+func TestRepositoryWithPrimaryDoesNotFallBackAfterPrimaryError(t *testing.T) {
+	repo := NewRepositoryWithPrimary(staticProfileRepo{err: profile.ErrInvalidProfile})
+
+	_, err := repo.GetProfile(context.Background(), "mistral-small-3")
+	if !errors.Is(err, profile.ErrInvalidProfile) {
+		t.Fatalf("expected primary error, got %v", err)
+	}
+}
+
+type staticProfileRepo struct {
+	profiles map[string]string
+	err      error
+}
+
+func (r staticProfileRepo) GetProfile(_ context.Context, id string) (*domain.ExecutionProfile, error) {
+	if r.err != nil {
+		return nil, r.err
+	}
+	if model, ok := r.profiles[id]; ok {
+		return &domain.ExecutionProfile{ID: id, Endpoint: "http://primary/v1", Model: model}, nil
+	}
+	return nil, profile.ErrProfileNotFound
+}

internal/profile/filesystem_repository.go

@@ -5,11 +5,13 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"io/fs"
 	"os"
-	"path/filepath"
+	"path"
 	"strings"
 
 	"gitea.maximumdirect.net/eric/scriptorium/internal/domain"
+	"gitea.maximumdirect.net/eric/scriptorium/internal/filecatalog"
 	"gopkg.in/yaml.v3"
 )
 
@@ -29,44 +31,89 @@ func NewFilesystemRepository(dir string) Repository {
 }
 
 func (r *filesystemRepository) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	return loadProfile(ctx, os.DirFS(r.dir), ".", id)
+}
+
+type fsRepository struct {
+	fsys fs.FS
+	root string
+}
+
+func NewFSRepository(fsys fs.FS, root string) Repository {
+	return &fsRepository{fsys: fsys, root: root}
+}
+
+func (r *fsRepository) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	return loadProfile(ctx, r.fsys, r.root, id)
+}
+
+type overlayRepository struct {
+	primary  Repository
+	fallback Repository
+}
+
+func NewOverlayRepository(primary, fallback Repository) Repository {
+	return &overlayRepository{primary: primary, fallback: fallback}
+}
+
+func (r *overlayRepository) GetProfile(ctx context.Context, id string) (*domain.ExecutionProfile, error) {
+	if r.primary != nil {
+		prof, err := r.primary.GetProfile(ctx, id)
+		if err == nil {
+			return prof, nil
+		}
+		if !errors.Is(err, ErrProfileNotFound) {
+			return nil, err
+		}
+	}
+	if r.fallback == nil {
+		return nil, ErrProfileNotFound
+	}
+	return r.fallback.GetProfile(ctx, id)
+}
+
+func loadProfile(ctx context.Context, fsys fs.FS, root string, id string) (*domain.ExecutionProfile, error) {
 	if strings.TrimSpace(id) == "" {
 		return nil, fmt.Errorf("%w: profile id is required", ErrInvalidProfile)
 	}
+	if fsys == nil {
+		return nil, fmt.Errorf("failed to read profile directory: filesystem is nil")
+	}
 
-	files, err := os.ReadDir(r.dir)
+	files, err := filecatalog.FindFSYAMLFiles(ctx, fsys, root)
 	if err != nil {
 		return nil, fmt.Errorf("failed to read profile directory: %w", err)
 	}
 
-	for _, file := range files {
+	var matches []profileMatch
+	for _, fullPath := range files {
 		select {
 		case <-ctx.Done():
 			return nil, ctx.Err()
 		default:
 		}
 
-		if file.IsDir() || (!strings.HasSuffix(file.Name(), ".yaml") && !strings.HasSuffix(file.Name(), ".yml")) {
-			continue
-		}
-
-		fullPath := filepath.Join(r.dir, file.Name())
-		data, err := os.ReadFile(fullPath)
+		relPath := filecatalog.DisplayPath(root, fullPath)
+		fileMatch := filecatalog.Stem(path.Base(fullPath)) == id
+		data, err := fs.ReadFile(fsys, fullPath)
 		if err != nil {
-			return nil, fmt.Errorf("failed to read profile file %s: %w", file.Name(), err)
+			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 {
-			if strings.Contains(err.Error(), "field api_key not found") {
-				if strings.TrimSuffix(strings.TrimSuffix(file.Name(), ".yaml"), ".yml") == id {
-					return nil, fmt.Errorf("%w: %s", ErrRawAPIKeyNotAllowed, file.Name())
-				}
-				continue
-			}
-			if strings.TrimSuffix(strings.TrimSuffix(file.Name(), ".yaml"), ".yml") == id {
-				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, file.Name(), err)
+			if idMatch {
+				return nil, fmt.Errorf("%w: %s: %v", ErrInvalidYAML, relPath, err)
 			}
 			continue
 		}
@@ -76,16 +123,68 @@ func (r *filesystemRepository) GetProfile(ctx context.Context, id string) (*doma
 		}
 		if err := validateProfile(&prof); err != nil {
 			if errors.Is(err, ErrRawAPIKeyNotAllowed) {
-				return nil, fmt.Errorf("%w: %s", err, file.Name())
+				return nil, fmt.Errorf("%w: %s", err, relPath)
 			}
-			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidProfile, file.Name(), err)
+			return nil, fmt.Errorf("%w: %s: %v", ErrInvalidProfile, relPath, err)
 		}
-		return &prof, nil
+		matches = append(matches, profileMatch{
+			profile: &prof,
+			path:    relPath,
+		})
+	}
+
+	if len(matches) > 1 {
+		paths := make([]string, 0, len(matches))
+		for _, match := range matches {
+			paths = append(paths, match.path)
+		}
+		return nil, fmt.Errorf("%w: duplicate execution profile id %q found in: %s", ErrInvalidProfile, id, strings.Join(paths, ", "))
+	}
+
+	if len(matches) == 1 {
+		return matches[0].profile, nil
 	}
 
 	return nil, ErrProfileNotFound
 }
 
+type profileMatch struct {
+	profile *domain.ExecutionProfile
+	path    string
+}
+
+type profileFileMetadata struct {
+	id           string
+	hasRawAPIKey bool
+}
+
+func readProfileFileMetadata(data []byte) profileFileMetadata {
+	var node yaml.Node
+	if err := yaml.NewDecoder(bytes.NewReader(data)).Decode(&node); err != nil {
+		return profileFileMetadata{}
+	}
+	if node.Kind != yaml.DocumentNode || len(node.Content) == 0 {
+		return profileFileMetadata{}
+	}
+	mapping := node.Content[0]
+	if mapping.Kind != yaml.MappingNode {
+		return profileFileMetadata{}
+	}
+
+	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 {
 	if strings.TrimSpace(p.ID) == "" {
 		return errors.New("id is required")