rules-distill

To extract cross-cutting principles into reusable rules, you scan installed skills and existing rules inventories deterministically, then apply LLM-based cross-reading to produce candidate principles with evidence and violation risk. The process outputs a results.json file routing each candidate to Append, Revise, New Section, New File, or Skip actions.

Rules distillation is the process of finding overlapping best practices trapped inside individual skills and promoting them to consistent, reusable governance rules. You need it during periodic rule maintenance, after skill stocktakes, or whenever rule coverage feels incomplete relative to your current skill set.

No, the rules-distill skill does not auto-modify rules. It produces candidate principles with evidence and saves a results.json file proposing Append, Revise, New Section, New File, or Skip actions. Explicit user approval, modification, or skipping is required before any rules are changed.

To maintain quality control over overlapping best practices, you run a deterministic inventory scan of all installed skills and existing rules, followed by LLM-based cross-reading verdicts. This generates actionable do X / don't Y behavioral guidance routed to specific rule update actions for user approval.

Yes, you can use LLM judgment to draft governance rules from automation skills. The skill uses scripts to exhaustively inventory skills and existing rules, then leverages an LLM to cross-read the full context and make verdicts, producing actionable rule drafts without auto-modifying your files.

Deterministic inventory scanning provides exhaustive coverage but relies on LLM cross-reading for verdicts, meaning output quality depends on context provided. It only generates candidate principles in results.json without auto-modifying rules, requiring manual user review and approval for every proposed Append, Revise, or New File action.