Ratchet

Records progress through the RPI workflow using a Brownian ratchet model: chaos (exploration), filter (validation gates), and permanent locking. Each step (research, plan, implement, vibe, post-mortem) gets recorded to a JSONL chain once completed, and you can't go backwards. It's explicitly not a loop runner, just the gate keeper. You pair it with `/crank` or `/swarm` to actually execute work. The key insight is that progress is permanent, which prevents backsliding but means you need to validate before ratcheting. Useful when you want enforced workflow discipline and a clear audit trail of what's been completed.

Install to Claude Code

npx -y skills add boshu2/agentops --skill ratchet --agent claude-code