# Configuration Reference

## Config Discovery And Precedence

Application settings are loaded in this order:

1. Built-in defaults
2. `config.yml` values
3. CLI overrides

When `--config` is not provided, Scriptorium searches for config files in this order:

1. `/usr/local/etc/scriptorium/config.yml`
2. `/etc/scriptorium/config.yml`

If neither file exists, Scriptorium continues with built-in defaults.

When `--config <path>` is provided, that file is required.

## Minimal App Config

```yaml
prompt_dir: ./examples/prompts
```

This is enough to use `run` and `render` when prompts select built-in profiles.

## Production-Oriented App Config

```yaml
prompt_dir: /opt/scriptorium/prompts
profile_dir: /opt/scriptorium/profiles
schema_dir: /opt/scriptorium/schemas

server:
  addr: 127.0.0.1:8080
  artifact_root: /var/lib/scriptorium/artifacts
  max_request_bytes: 16777216
  max_artifact_bytes: 16777216
  max_response_bytes: 16777216

defaults:
  render_format: text
```

## App Config File (`config.yml`)

Top-level fields:

- `prompt_dir` (optional): default prompt definition directory.
- `profile_dir` (optional): default custom profile definition directory.
- `schema_dir` (optional): base directory for schema files used by `json_schema` validation.
- `server.addr` (optional): default listen address for `serve`.
- `server.artifact_root` (optional): base directory for HTTP `file` input references.
- `server.max_request_bytes` (optional): maximum HTTP request body size. `0` disables this limit.
- `server.max_artifact_bytes` (optional): maximum HTTP `file` input artifact size. `0` disables this limit.
- `server.max_response_bytes` (optional): maximum encoded HTTP response body size. `0` disables this limit.
- `defaults.render_format` (optional): default `render` output format (`text` or `json`).

Built-in defaults:

- `schema_dir`: `.`
- `server.addr`: `:8080`
- `server.artifact_root`: unset; HTTP `file` input references are rejected until configured.
- `server.max_request_bytes`: `16777216` (16 MiB)
- `server.max_artifact_bytes`: `16777216` (16 MiB)
- `server.max_response_bytes`: `16777216` (16 MiB)
- `defaults.render_format`: `text`

Validation behavior:

- Config decoding is strict; unknown YAML fields are rejected.
- HTTP size limit values must be greater than or equal to `0`.
- Raw API key fields are not supported in `config.yml`.

HTTP artifact root behavior:

- `server.artifact_root` applies only to `serve`.
- HTTP `inline` input references work without an artifact root.
- HTTP `file` input references are resolved against `server.artifact_root` and must stay inside it.
- Relative traversal and absolute paths outside the root are rejected.
- Symlinks inside the root are followed by the operating system; do not make the artifact root writable by untrusted users.
- CLI `run` and `render` file inputs keep their normal direct filesystem path behavior.

HTTP size-limit behavior:

- The request limit covers the encoded JSON request body, including inline input bodies.
- The artifact limit covers HTTP `file` input artifacts read through `serve`.
- The response limit covers the final encoded JSON response, including generated artifact bodies and `raw_model_output` when requested.
- Limits apply only to HTTP `serve`; CLI `run` and `render` keep direct filesystem behavior.

## Prompt Definition Files

Prompt definitions are YAML files anywhere under `prompt_dir`, including nested subdirectories.

Subdirectories are organizational only. Callers still select prompts by the YAML `id`, not by file path. For example, `prompts/dnd/recap.yaml` may still declare `id: dnd.recap`, and callers use `--prompt dnd.recap`.

Example:

```yaml
id: generic.structured_events
version: "1.0.0"
default_profile: local-quality
description: Produce structured event JSON from a transcript.

inputs:
  - name: transcript
    required: true
    content_type: text/markdown
    description: Source transcript content
  - name: glossary
    required: false
    content_type: text/yaml
    description: Optional glossary context

messages:
  - role: system
    content_file: ./generic.structured_events.system.md
  - role: user
    content_file: ./generic.structured_events.user.md

output:
  format: json
  validation_mode: json_schema
  schema_path: structured_events.schema.json
  repair_attempts: 0
```

Field reference:

- `id` (required): prompt identifier.
- `version` (required): prompt version.
- `default_profile` (optional): profile ID used when request does not provide `profile_id`.
- `description` (optional): prompt description.
- `session_id` (optional): Go-template string for OpenRouter sticky-routing `session_id`; rendered from request vars.
- `inputs` (optional list): expected named inputs.
- `messages` (required list): prompt message templates.
- `output` (required object): output contract.

`inputs[]` fields:

- `name` (required)
- `required` (optional, boolean)
- `content_type` (optional metadata)
- `description` (optional)

`messages[]` fields:

- `role` (required)
- `content` or `content_file` (exactly one is required)
- `cache_control` (optional object): provider prompt-cache metadata for this message

Message rules:

- Repeated roles are allowed.
- `content_file` is resolved relative to the prompt YAML file location.
- Nested prompt files keep the same relative `content_file` behavior; `./recap.user.md` next to `dnd/recap.yaml` resolves from `dnd/`.
- Prompt decoding is strict; unknown YAML fields are rejected.
- Duplicate prompt IDs are invalid. If multiple files declare the requested prompt ID, Scriptorium fails instead of choosing one.

`messages[].cache_control` fields:

- `type` (required when `cache_control` is present): currently only `ephemeral`.
- `ttl` (optional): currently only `1h`; omitted from outbound requests when unset.

Example cache-controlled message:

```yaml
messages:
  - role: system
    content_file: ./stable_context.md
    cache_control:
      type: ephemeral
      ttl: 1h
  - role: user
    content: |
      {{input "transcript"}}
```

Use cache control on stable reusable prompt content. Dynamic per-run inputs before the cache-controlled message change the provider cache key.

Example prompt-level session ID:

```yaml
session_id: "{{ .session_id }}"
```

When configured, `session_id` is rendered with the same variable context as messages. The rendered value is trimmed, omitted when empty, and rejected if longer than 256 characters. CLI callers pass the value through `--var session_id=<value>`; HTTP callers pass it through `"vars": {"session_id": "<value>"}`.

`output` fields:

- `format` (required): `text`, `markdown`, or `json`.
- `validation_mode` (required): `none`, `basic`, `json`, or `json_schema`.
- `schema_path` (required when `validation_mode: json_schema`).
- `repair_attempts` (required): integer `>= 0`.

Repair behavior boundary:

- `repair_attempts` is part of the prompt contract.
- CLI and HTTP currently construct the runner without a repairer, so normal `run`/`serve` execution does not perform output repair attempts.

## Profile Definition Files

Scriptorium includes built-in execution profiles. Custom execution profiles are YAML files anywhere under `profile_dir`, including nested subdirectories.

Subdirectories are organizational only. Callers still select profiles by the YAML `id`, not by file path. For example, `profiles/local/local-quality.yaml` may still declare `id: local-quality`, and callers use `--profile local-quality`.

Example:

```yaml
id: local-fast
endpoint: http://localhost:8000/v1
model: gpt-4o-mini
temperature: 0.2
max_tokens: 500
top_p: 1.0
timeout_seconds: 90
api_key_env: SCRIPTORIUM_API_KEY
service_tier: priority
reasoning_effort: medium
extra_params:
  provider_route: primary
  provider_options:
    retry_budget: 2
```

Field reference:

- `id` (required)
- `endpoint` (required)
- `model` (required)
- `temperature` (optional): range `0..2`
- `max_tokens` (optional): `>= 0`
- `top_p` (optional): range `0..1`
- `timeout_seconds` (optional): `>= 0`
- `service_tier` (optional): provider-specific request tier such as OpenRouter `flex` or `priority`
- `reasoning_effort` (optional): serialized as top-level `reasoning_effort` in outbound chat-completions requests
- `api_key_env` (optional)
- `extra_params` (optional map): JSON-compatible provider-specific parameters. Values may be strings, numbers, booleans, objects, or arrays.

Profile rules:

- `profile_dir` is optional. If omitted, only built-in profiles are available.
- If `profile_dir` is set, custom profiles from that directory override built-in profiles with the same `id`.
- Duplicate IDs within the custom profile directory are invalid. Matching IDs across custom and built-in profiles are valid override behavior.
- Profile decoding is strict; unknown YAML fields are rejected.
- Raw `api_key` is rejected; use `api_key_env`.
- If `api_key_env` is set, that environment variable must be set when preparing/running.
- Duplicate profile IDs are invalid. If multiple files declare the requested profile ID, Scriptorium fails instead of choosing one.
- `extra_params` keys must not be empty and must not collide with reserved outbound request fields: `model`, `session_id`, `messages`, `temperature`, `max_tokens`, `top_p`, `service_tier`, `reasoning_effort`, or `response_format`.

Built-in profile IDs:

| Provider | ID | Model | API key env |
| --- | --- | --- | --- |
| aion-labs | `aion-2` | `aion-labs/aion-2.0` | `OPENROUTER_API_KEY` |
| anthropic | `claude-fable-latest` | `~anthropic/claude-fable-latest` | `OPENROUTER_API_KEY` |
| anthropic | `claude-haiku-latest` | `~anthropic/claude-haiku-latest` | `OPENROUTER_API_KEY` |
| anthropic | `claude-opus-latest` | `~anthropic/claude-opus-latest` | `OPENROUTER_API_KEY` |
| anthropic | `claude-sonnet-latest` | `~anthropic/claude-sonnet-latest` | `OPENROUTER_API_KEY` |
| deepseek | `deepseek-3-2` | `deepseek/deepseek-v3.2` | `OPENROUTER_API_KEY` |
| deepseek | `deepseek-4-pro` | `deepseek/deepseek-v4-pro` | `OPENROUTER_API_KEY` |
| google | `gemini-2-flash` | `google/gemini-2.5-flash` | `OPENROUTER_API_KEY` |
| google | `gemini-2-flash-lite` | `google/gemini-2.5-flash-lite` | `OPENROUTER_API_KEY` |
| google | `gemini-2-pro` | `google/gemini-2.5-pro` | `OPENROUTER_API_KEY` |
| google | `gemini-3-flash-lite` | `google/gemini-3.1-flash-lite` | `OPENROUTER_API_KEY` |
| google | `gemini-flash-latest` | `~google/gemini-flash-latest` | `OPENROUTER_API_KEY` |
| google | `gemini-pro-latest` | `~google/gemini-pro-latest` | `OPENROUTER_API_KEY` |
| google | `gemma-4-31b` | `google/gemma-4-31b-it:exacto` | `OPENROUTER_API_KEY` |
| minimax | `minimax-m2` | `minimax/minimax-m2.5` | `OPENROUTER_API_KEY` |
| minimax | `minimax-m3` | `minimax/minimax-m3` | `OPENROUTER_API_KEY` |
| mistral | `mistral-large-2512` | `mistralai/mistral-large-2512` | `OPENROUTER_API_KEY` |
| mistral | `mistral-medium-3-5` | `mistralai/mistral-medium-3-5` | `OPENROUTER_API_KEY` |
| mistral | `mistral-small-3` | `mistralai/mistral-small-3.2-24b-instruct` | `OPENROUTER_API_KEY` |
| mistral | `mistral-small-4` | `mistralai/mistral-small-2603` | `OPENROUTER_API_KEY` |
| nvidia | `nemotron-3-ultra` | `nvidia/nemotron-3-ultra-550b-a55b` | `OPENROUTER_API_KEY` |
| openai | `gpt-5-mini` | `openai/gpt-5.4-mini` | `OPENROUTER_API_KEY` |
| openai | `gpt-5-nano` | `openai/gpt-5.4-nano` | `OPENROUTER_API_KEY` |

Current outbound request behavior:

- The OpenAI-compatible client currently serializes: `model`, optional `session_id`, `messages`, `temperature`, `max_tokens`, `top_p`, `service_tier`, `reasoning_effort`, optional `response_format` for `json_schema` prompts, and `extra_params`.
- `extra_params` are flattened into provider-specific top-level JSON request fields. They are not wrapped in an `extra_params` object on the outbound provider request.
- Messages without `cache_control` serialize with string `content`.
- Messages with `cache_control` serialize as a single text content-block array containing `cache_control`.

## Schema Behavior

Schemas are JSON files, typically in `schema_dir`.

Rules:

- `output.validation_mode: json_schema` requires `output.schema_path`.
- Relative `schema_path` values resolve from `schema_dir`, including explicit nested paths such as `dnd/structured_events.schema.json`.
- Absolute `schema_path` values are used directly.
- Scriptorium does not recursively search schemas by basename; nested schemas must be referenced by their relative path.
- Missing or invalid schema documents cause runtime validation errors.
- Invalid generated JSON causes validation status `failed` (not a runtime error).

Supported artifact reference types for request inputs are `file` and `inline`.
For HTTP `serve`, `file` references require `server.artifact_root` and must stay
inside that root. CLI `run` and `render` file inputs are not restricted by
`server.artifact_root`.

## Secrets Handling

- Keep secret values in environment variables.
- Store only environment-variable names in profile `api_key_env`.
- Do not put raw API keys in config, prompts, profiles, CLI flags, or HTTP request bodies.

## Maintained Examples

- App config: `examples/config.yml`
- Prompt examples: `examples/prompts/`
- Custom profile examples: `examples/profiles/`
- Schema examples: `examples/schemas/`
- Input fixtures: `examples/fixtures/`
- Render example script: `examples/render-markdown-summary.sh`
- HTTP request example: `examples/http-run.json`

Example organizational layout:

```text
examples/prompts/dnd/recap.yaml
examples/profiles/local/local-quality.yaml
examples/schemas/dnd/structured_events.schema.json
```

## Integration References

- [Inbound HTTP contract](integrations/http-api.md)
- [Outbound OpenAI-compatible contract](integrations/openai-compatible-chat.md)