What problem does it solve?
Provides a safe, auditable way to inspect and continue paired-agent debates stored in a project's .ratchet directory, addressing the need to read full transcripts, recover corrupted state, and resume or conclude escalated debates without manual file surgery.
Core Features & Use Cases
- List debates and statuses: Enumerate debates by reading .ratchet/debates/*/meta.json and present concise choices to the user.
- Full transcript viewing: Aggregate meta.json and per-round generative/adversarial round files to present a readable transcript with verdict and metadata.
- Recovery and resilience: Attempt to regenerate malformed meta.json from existing round files, surface recovery notes, and guide user decisions.
- Controlled continuation: Resume debates only when allowed (initiated/escalated), enforce agent tool boundaries, increment max_rounds or reset rounds, and update meta.json for abandon or verdict actions.
- Use case: A maintainer investigating why an automated debate stalled can inspect transcripts, regenerate corrupted metadata from round files, and safely run another round or render a human verdict.
Quick Start
Run /ratchet:debate to list debates, select an ID to view its full transcript, or add --continue to resume an eligible debate.