Pulls MUST/SHOULD/MAY statements out of your agent instruction files and dumps them to JSONL. Does a fast regex pass first, then sends each file to a subagent to catch the stuff that's not explicitly marked with RFC 2119 keywords. Things like implied obligations in prose or conditional rules buried in context. Defaults to scanning your ai-workspace/rules directory and AGENTS.md. The output feeds into /policy-algebra and /distill for conflict detection and cleanup. Honestly most useful when you've accumulated a bunch of instruction files and need to see what you've actually committed to across all of them. Reports zero imperatives without erroring, which is the right call.
npx skills add https://github.com/camacho/ai-skills --skill imperatives