Refactor: split prompt definition from execution settings and migrate run contracts to prompt_* + execution_target

This commit is contained in:
2026-05-05 10:09:31 -05:00
parent fdfd8641f5
commit a633c67538
28 changed files with 712 additions and 1021 deletions

View File

@@ -2,11 +2,11 @@
## 1. Purpose and Non-Goals
Scriptorium is a prompt-profile execution engine.
Scriptorium is a prompt-definition execution engine.
It takes named input artifacts, renders prompt templates, calls an LLM, validates output, optionally performs bounded structured-output repair, and returns an artifact with metadata.
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 is not an orchestrator. It should not own transcription, transcript merging, transcript polishing, notification, or cross-step workflow control.
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:
@@ -14,111 +14,116 @@ For the motivating D&D workflow:
- WhisperX transcribes.
- Seriatim merges transcripts.
- Audita polishes transcripts.
- Scriptorium generates output artifacts from prepared inputs.
- Scriptorium generates final artifacts from prepared inputs.
Core Go code must remain domain-generic.
Core Go code remains generic.
## 2. Current Architecture
Current high-level structure:
Current implementation structure:
- `cmd/scriptorium`: binary entrypoint.
- `internal/domain`: core domain types.
- `internal/usecase`: `Runner` use case and repair loop orchestration.
- `internal/profile`: filesystem prompt profile repository and profile validation.
- `internal/artifact`: artifact reference readers (`inline`, `file`) and routing.
- `internal/prompt`: Go template-based prompt renderer.
- `internal/llm`: LLM client interface + OpenAI-compatible HTTP adapter.
- `internal/validate`: output validator implementation (`none/basic/json/json_schema`).
- `internal/domain`: core domain contracts.
- `internal/usecase`: `Runner` run flow, validation integration, bounded repair coordination.
- `internal/profile`: transitional filesystem prompt-definition repository (package rename deferred).
- `internal/artifact`: input artifact resolution (`inline`, `file`).
- `internal/prompt`: template rendering.
- `internal/llm`: provider-neutral client interface + OpenAI-compatible HTTP adapter.
- `internal/validate`: validation implementation (`none/basic/json/json_schema`).
- `internal/adapter/cli`: CLI adapter.
- `internal/adapter/http`: HTTP adapter (`POST /v1/runs`).
This is a practical ports-and-adapters implementation.
## 3. Run Data Flow
`Runner.Run(ctx, RunRequest)` currently executes:
1. Validate minimum request requirements (`profile_id`).
2. Load prompt profile by ID/version.
3. Merge effective model target (profile defaults + request override).
4. Resolve effective output contract (profile + optional request override).
5. Resolve named artifact refs to loaded artifacts.
6. Render prompt messages from templates.
7. Hash rendered prompt for auditability.
8. Call LLM client with provider-neutral `GenerateRequest`.
9. Build output artifact from model content.
1. Validate request (`prompt_id` required).
2. Load `PromptDefinition` by ID/version.
3. Determine selected profile ID (`request.profile_id` or prompt `default_profile`).
4. Resolve effective execution target from request override (execution-profile loading is deferred in this phase).
5. Resolve named input artifact refs.
6. Render prompt messages.
7. Hash prompt definition and rendered prompt.
8. Call LLM client with `GenerateRequest`.
9. Build output artifact.
10. Validate output.
11. If structured validation failed and repair is enabled/bounded, run repair attempts and re-validate.
12. Return `RunResult` with artifact, validation, raw output, and metadata.
11. If structured validation failed and repair is enabled, run bounded repair attempts and re-validate.
12. Return `RunResult` with artifact, raw output, validation result, metadata.
Validation content failure remains a successful run result with `validation.status=failed`.
Validation content failures are returned as successful runs with `validation.status=failed`.
## 4. Package Responsibilities
- `domain`
- Owns core nouns and contracts.
- Must not import adapters/provider SDK types.
- Owns core nouns/contracts.
- Must not depend on adapters/provider SDK types.
- `usecase`
- Owns execution sequence and cross-port orchestration for a single run.
- May coordinate validation and bounded repair.
- Must not contain HTTP/CLI/wire concerns.
- Owns single-run orchestration across ports.
- Owns bounded repair control flow.
- Must not own transport/wire concerns.
- `profile`
- Owns prompt profile loading/parsing/validation.
- Handles YAML strict decoding and profile-level constraints.
- `profile` (transitional)
- Currently loads prompt definitions from YAML.
- Package naming split (`prompt definition repo` vs `execution profile repo`) is deferred follow-up.
- `artifact`
- Owns artifact ref resolution and content loading.
- Produces normalized `Artifact` values with size/hash/content type.
- Loads artifacts from refs and normalizes payload metadata.
- `prompt`
- Owns template rendering and required-input enforcement.
- Renders templates and enforces required inputs.
- `llm`
- Owns generation port and provider adapters.
- Current adapter: OpenAI-compatible chat completions over `net/http`.
- Defines generation client contract and protocol adapters.
- `validate`
- Owns output validation semantics and JSON Schema integration.
- Owns output validation semantics and schema validation.
- `adapter/http`, `adapter/cli`
- Owns transport/wire/flag concerns only.
- Should stay thin and delegate business flow to `usecase.Runner`.
- Own request/response/flag mapping only.
- Delegate business flow to `usecase.Runner`.
## 5. Domain Model (Current)
Key types in `internal/domain`:
Key types:
- `RunRequest`: profile selector, named input refs, vars, optional model override, optional validation override.
- `RunResult`: output artifact, validation result, raw output, profile/model metadata, hashes, usage, timestamps, duration.
- `ArtifactRef`: `{type, uri, body}` reference contract.
- `Artifact`: loaded payload (`name`, `content_type`, `body`, `uri`, `size`, `hash`).
- `PromptProfile`: YAML-backed profile definition.
- `RenderedPrompt` / `RenderedMessage`: provider-neutral prompt structure.
- `GenerateRequest` / `GenerateResponse`: provider-neutral model I/O.
- `ValidationResult`: passed/failed/skipped + mode/errors/schema/repair attempts.
- `PromptDefinition`
- `id`, `version`, `default_profile`, `inputs`, `templates`, `output_format`, `validation`.
- `ExecutionProfile`
- Execution/runtime settings shape (`endpoint`, `model`, timeouts, `api_key_env`, etc.).
- Loading/persistence is deferred in this pass.
- `ExecutionTarget`
- Effective execution settings for a run.
- `RunRequest`
- `prompt_id`, `prompt_version`, optional `profile_id`, `inputs`, `vars`, optional `execution` override, optional validation override.
- `RunResult`
- Output artifact, validation, raw output, prompt/profile/model metadata, hashes, timing, usage.
- `ArtifactRef` / `Artifact`
- Input reference and loaded content contracts.
- `RenderedPrompt` / `RenderedMessage`
- Provider-neutral rendered prompt.
- `GenerateRequest` / `GenerateResponse`
- Provider-neutral model I/O.
## 6. Interfaces and Adapters
Primary ports:
- `profile.Repository`
- `profile.Repository` (transitional prompt-definition lookup)
- `artifact.Reader`
- `prompt.Renderer`
- `llm.Client`
- `validate.Validator`
- `usecase.OutputRepairer` (usecase-local abstraction)
- `usecase.OutputRepairer` (usecase-local)
Current adapters:
- Profile repository: filesystem YAML loader.
- Artifact reader: composite reader for `inline` and `file`.
- Prompt renderer: Go templates with input helper + vars.
- LLM adapter: OpenAI-compatible `/chat/completions`.
- Validator: standard validator with `none/basic/json/json_schema`.
- CLI/HTTP adapters: thin request mapping and response mapping.
- Prompt definition repository: filesystem YAML loader.
- Artifact readers: `file`, `inline` via composite reader.
- Prompt renderer: Go templates with `input` helper.
- LLM adapter: OpenAI-compatible `/chat/completions` over `net/http`.
- Validator: standard validator (`none/basic/json/json_schema`).
- CLI/HTTP adapters.
## 7. Validation and Repair Model
@@ -131,12 +136,11 @@ Validation modes:
Repair behavior:
- Only applies to structured modes (`json`, `json_schema`).
- Attempted only when validation fails, repairer exists, and `repair_attempts > 0`.
- Bounded strictly by `repair_attempts`.
- Uses a narrow JSON-repair prompt and re-validates each attempt.
- If still invalid, run succeeds with failed validation and preserved final raw output.
- Validator runtime/config errors are run errors.
- Applies only to structured modes (`json`, `json_schema`).
- Triggered only on failed validation and only when `repair_attempts > 0`.
- Strictly bounded by `repair_attempts`.
- Uses a narrow repair prompt asking for corrected JSON only.
- Runtime validator/repair errors are run errors.
## 8. Public Contracts
@@ -147,86 +151,76 @@ Commands:
- `scriptorium run`
- `scriptorium serve`
`run`:
`run` flags:
- Required: `--profile-dir`, `--profile-id`, `--input`.
- Optional: model/endpoint overrides (`--model`, `--llm-base-url`), vars, output path, schema dir, timeout.
- Artifact bytes go to stdout (or `--out` file); summaries/errors go to stderr.
- Required: `--profile-dir`, `--prompt-id`, `--input`.
- Optional: `--profile-id`, `--var`, `--out`, `--llm-base-url`, `--model`, `--api-key-env`, `--temperature`, `--max-tokens`, `--schema-dir`, `--timeout`.
`serve`:
Current transitional runtime behavior:
- Required: `--profile-dir`, `--llm-base-url`.
- Exposes HTTP run endpoint.
- Prompt definitions may provide `default_profile` selection.
- Execution-profile loading is deferred; execution settings must currently be supplied via run-time overrides.
### HTTP
- Endpoint: `POST /v1/runs`.
- Request maps to `RunRequest` (`profile_id`, `inputs`, `vars`, optional `model` override).
- Request maps to `RunRequest` with `prompt_id` (required), `inputs`, optional `profile_id`, `vars`, optional execution override (`model` object).
- Response includes `artifact`, `validation`, `metadata`, `raw_model_output`.
- Validation content failures are represented as `200` with `validation.status=failed`.
- Error responses are `{error:{code,message}}` with stable code mapping.
- Validation content failures return `200` with failed validation status.
- Error response shape: `{ "error": { "code": "...", "message": "..." } }`.
### Prompt Profile YAML
### Prompt Definition YAML
- `id`, `version`, `expected_inputs`, `templates`, `model_defaults`, `output_format`, `validation`.
- Strict YAML decoding (`KnownFields`) rejects unknown fields.
- `validation.schema_path` required when `validation_mode=json_schema`.
- `validation.repair_attempts` must be non-negative.
Current prompt-definition fields:
### Metadata
- `id`, `version`, optional `default_profile`, optional `description`
- `inputs[]` with `name`, `required`, optional `content_type`, optional `description`
- `templates[]` with `role` and either `content` or `content_file`
- `output_format`
- `validation` (`format`, `validation_mode`, `schema_path`, `repair_attempts`)
Current run metadata includes:
Strict YAML decoding (`KnownFields`) is enabled.
- `run_id` (UUID v4)
- `profile_id`, `profile_version`, `profile_hash`
- `model_name`, `endpoint`, effective `model_params`
- `input_hashes`, `prompt_hash`
- token usage
- start/end timestamps
- duration
- validation mode/status
- repair attempts used
### API Key Policy
- Raw API keys are not accepted in YAML, CLI flags, HTTP body, or domain metadata.
- Auth is configured only by env var reference (`api_key_env`), resolved at request time by the LLM adapter.
## 9. Extension Points (Future Work)
Future features should plug into existing boundaries, not bypass them.
Planned next extensions should reuse current boundaries:
Candidate extensions:
- S3 artifact refs via `artifact.Reader` extension.
- Token budgeting in usecase/model-target policy layer.
- Streaming LLM output via additional `llm.Client` methods/adapters.
- Batch execution as a separate use case (not hidden in single-run path).
- Additional LLM providers implementing `llm.Client`.
- Additional validators/modes in `validate`.
- Additional profile repositories (embedded, remote, object storage).
These are future work, not part of current default behavior.
- Execution-profile repository/loader implementation.
- Split transitional `internal/profile` into clearer prompt-definition/profile repositories.
- S3 artifact refs.
- Token budgeting/policy layer.
- Streaming generation.
- Batch run use case.
- Additional provider adapters.
- Additional validation modes.
## 10. Architectural Guardrails
Contributors should preserve these constraints:
- No D&D-specific behavior in core Go packages.
- No D&D-specific logic in core Go packages.
- No orchestration creep into Scriptorium.
- No unbounded repair loops.
- No silent truncation/omission of rendered inputs or outputs.
- Do not log full artifacts/prompts by default.
- No silent content truncation/omission.
- Do not log full prompts/artifacts by default.
- Keep provider-specific wire/SDK details out of domain types.
- Keep adapter boundaries explicit and thin.
- Keep adapters thin.
## 11. Testing Strategy
Protect behavior at boundaries and in usecase flow:
Protect these behaviors with focused tests:
- Profile loading/parsing/validation errors.
- Artifact reading for inline/file + hash/content type behavior.
- Prompt rendering required inputs/template error behavior.
- LLM adapter request/response/error/timeout behavior.
- Runner success path and metadata population.
- Prompt-definition loading/validation errors.
- Artifact loading/hash/content-type behavior.
- Prompt rendering required-input and template error paths.
- LLM adapter request/response/auth/error/timeout behavior.
- Runner success/failure/metadata behavior.
- Validation failure raw-output preservation.
- Successful/failed/bounded repair flows.
- HTTP request mapping, response shape, and error mapping.
- CLI parsing helpers, required flags, and output stream separation.
- Bounded repair behavior.
- HTTP mapping and error mapping.
- CLI parsing and output stream separation.
Prefer focused unit tests and small integration-style tests with fake LLMs.
Prefer small unit tests and minimal integration-style tests with fake LLMs.