Context
Long WhisperForge runs can involve transcription, several LLM stages, optional images, Notion save, markdown export, and history writes. Today too much of that is session-state-only, which makes interruption/retry behavior fragile.
Scope
- Add a stable local run ID.
- Checkpoint intermediate stage outputs under
.cache/runs/<run_id>/ or an equivalent ignored artifact path.
- Allow retrying Notion save, markdown export, and image generation without rerunning the entire pipeline.
- Attach cost/token/cache/model/settings/source metadata before save.
Acceptance criteria
- A partially completed run can be inspected from local artifacts.
- Save/export retry does not duplicate or corrupt history unexpectedly.
- Run history can link to local artifacts as well as Notion URLs.
- Tests cover the run-record/checkpoint behavior.
Verification
venv/bin/python -m pytest tests/ -q
- Manual or automated run-interruption/retry smoke documented in
STATUS.md
Refs ROADMAP.md Phase 3.
Context
Long WhisperForge runs can involve transcription, several LLM stages, optional images, Notion save, markdown export, and history writes. Today too much of that is session-state-only, which makes interruption/retry behavior fragile.
Scope
.cache/runs/<run_id>/or an equivalent ignored artifact path.Acceptance criteria
Verification
venv/bin/python -m pytest tests/ -qSTATUS.mdRefs
ROADMAP.mdPhase 3.