scriptorium/README.md

# scriptorium

Scriptorium is a generic prompt-definition execution engine written in Go.

Given named input artifacts and a prompt definition, Scriptorium:

1. Loads the prompt definition.
2. Resolves input artifact references.
3. Renders prompt messages from templates.
4. Calls an OpenAI-compatible LLM endpoint.
5. Validates output if configured.
6. Optionally performs bounded structured-output repair.
7. Returns a generated artifact plus run metadata.

## Where Scriptorium Fits

Scriptorium is not an orchestrator.

In the D&D workflow:

- Narratio orchestrates the full pipeline.
- WhisperX transcribes audio.
- Seriatim merges transcripts.
- Audita polishes transcripts.
- Scriptorium generates final artifacts from prepared inputs.

D&D-specific behavior belongs in profiles, schemas, fixtures, and caller inputs, not in core Go logic.

## Core Concepts

- Prompt definition: YAML config for templates, inputs, output format, and validation behavior.
- Execution profile: conceptual runtime config (endpoint/model/timeouts/auth source). In this transition, execution settings are supplied as run-time overrides.
- Named inputs: logical names (for example `transcript`, `glossary`) mapped to artifact references.
- Artifact refs: currently `file` and `inline` are supported.
- Template variables: key/value vars passed at run time and referenced as `{{.var_name}}`.
- Execution target: endpoint/model plus generation/runtime parameters (`temperature`, `max_tokens`, `top_p`, `timeout_seconds`, `reasoning_effort`, `api_key_env`).
- Output format: `text`, `markdown`, or `json`.
- Validation mode: `none`, `basic`, `json`, `json_schema`.
- Repair attempts: bounded retries for structured modes (`json`, `json_schema`) when output validation fails.

## Build and Test

```bash
go build -o scriptorium ./cmd/scriptorium
go test ./...
```

## CLI Usage

### `scriptorium run`

Required flags:

- `--prompt-dir`
- `--profile-dir`
- `--prompt`
- `--input` (repeatable `name=path`)

Common optional flags:

- `--profile` (execution profile selector; falls back to prompt `default_profile`)
- `--var` (repeatable `name=value`)
- `--out`
- `--llm-base-url`
- `--model`
- `--api-key-env`
- `--temperature`
- `--max-tokens`
- `--schema-dir`
- `--timeout`

Current transitional behavior: execution-profile loading is not implemented yet, so run-time execution settings must be supplied via overrides. In practice, provide at least endpoint and model (`--llm-base-url` and `--model`).

Example:

```bash
export SCRIPTORIUM_API_KEY="your-key"

go run ./cmd/scriptorium run \
  --prompt-dir ./prompts \
  --profile-dir ./profiles \
  --prompt generic.markdown_summary \
  --profile local-fast \
  --input transcript=./examples/fixtures/transcript.md \
  --input glossary=./examples/fixtures/glossary.yml \
  --llm-base-url http://localhost:8000/v1 \
  --model gpt-4o-mini \
  --api-key-env SCRIPTORIUM_API_KEY \
  --out ./out.md
```

Output behavior:

- Artifact content goes to stdout unless `--out` is set.
- Summaries and errors are written to stderr.
- Exit code `2` means the run succeeded but validation status is `failed`.

### `scriptorium serve`

Starts HTTP API.

Required flags:

- `--prompt-dir`
- `--profile-dir`

Common optional flags:

- `--addr` (default `:8080`)
- `--schema-dir` (default `.`)
- `--model`
- `--timeout` (default `10m`)

## HTTP API

Endpoint:

- `POST /v1/runs`

No built-in authentication is provided by the server itself. Deploy behind a trusted boundary or gateway.

Request example:

```json
{
  "prompt_id": "generic.structured_events",
  "prompt_version": "1.0.0",
  "profile_id": "local-default",
  "inputs": {
    "transcript": {"type": "file", "uri": "./examples/fixtures/transcript.md"},
    "glossary": {"type": "file", "uri": "./examples/fixtures/glossary.yml"}
  },
  "vars": {
    "session_date": "2026-05-04"
  },
  "model": {
    "endpoint": "http://localhost:8000/v1",
    "model": "gpt-4o-mini",
    "temperature": 0.0,
    "max_tokens": 600,
    "top_p": 1.0,
    "timeout_seconds": 120,
    "api_key_env": "SCRIPTORIUM_API_KEY"
  }
}
```

Response shape:

```json
{
  "artifact": {
    "name": "output",
    "content_type": "application/json",
    "body": "{...}",
    "uri": "",
    "size": 123,
    "hash": "..."
  },
  "validation": {
    "status": "passed",
    "mode": "json_schema",
    "errors": [],
    "schema_path": "structured_events.schema.json",
    "repair_attempts": 0,
    "is_valid": true
  },
  "metadata": {
    "run_id": "xxxxxxxx-xxxx-4xxx-8xxx-xxxxxxxxxxxx",
    "prompt_id": "generic.structured_events",
    "prompt_version": "1.0.0",
    "prompt_hash": "...",
    "rendered_prompt_hash": "...",
    "selected_profile_id": "local-default",
    "model_name": "gpt-4o-mini",
    "endpoint": "http://localhost:8000/v1",
    "model_params": {
      "endpoint": "http://localhost:8000/v1",
      "model": "gpt-4o-mini",
      "temperature": 0,
      "max_tokens": 600,
      "top_p": 1,
      "timeout_seconds": 120,
      "api_key_env": "SCRIPTORIUM_API_KEY"
    },
    "input_hashes": {"transcript": "...", "glossary": "..."},
    "usage": {"prompt_tokens": 10, "completion_tokens": 20, "total_tokens": 30},
    "start_time": "...",
    "end_time": "...",
    "duration_ms": 1523,
    "validation_mode": "json_schema",
    "validation_status": "passed",
    "repair_attempts_used": 0
  },
  "raw_model_output": "{...}"
}
```

Validation content failures return `200` with `validation.status = "failed"` and preserve `raw_model_output`.

Error response shape:

```json
{
  "error": {
    "code": "artifact_read_failed",
    "message": "failed to read input artifact"
  }
}
```

## Prompt Definition Authoring

### Minimal Markdown prompt definition

```yaml
id: generic.markdown_summary
version: "1.0.0"
default_profile: local-default
inputs:
  - name: transcript
    required: true
templates:
  - role: system
    content: "You are a concise assistant."
  - role: user
    content: |
      Summarize:
      {{input "transcript"}}
output_format: markdown
validation:
  validation_mode: basic
```

### Structured JSON prompt definition with schema validation

```yaml
id: generic.structured_events
version: "1.0.0"
default_profile: local-default
inputs:
  - name: transcript
    required: true
templates:
  - role: system
    content: "Return only JSON."
  - role: user
    content: |
      Extract events from:
      {{input "transcript"}}
output_format: json
validation:
  format: json
  validation_mode: json_schema
  schema_path: structured_events.schema.json
  repair_attempts: 1
```

`repair_attempts` is strictly bounded and only applies to structured validation modes.

## Validation Modes

- `none`: skipped validation result.
- `basic`: fails for empty/whitespace output.
- `json`: output must parse as JSON.
- `json_schema`: output must parse as JSON and satisfy configured schema.

Validation content failures are returned in the structured result; raw model output is preserved.

## Examples

- Prompt definitions: `prompts/`
- Execution profiles: `profiles/`
- Schemas: `schemas/`
- Fixtures: `examples/fixtures/`
- Local experimentation: `local-test/`

## Development Notes

- Core follows ports-and-adapters and remains domain-generic.
- Domain/usecase packages do not depend on HTTP/CLI wire DTOs.
- To add a new LLM adapter: implement `internal/llm.Client`.
- To add a new artifact reader: extend `internal/artifact.Reader` routing.
- To add a new validation mode: extend `internal/validate` and preserve run semantics.