rules-distill

You can automate governance by using a distillation workflow that scans installed skills and rules, clusters recurring principles by theme, and yields actionable recommendations to append, revise, or create new rule files.

LLM-based judgment analyzes clustered principles gathered by deterministic scripts, evaluating cross-cutting patterns to output structured verdicts like Append, Revise, New Section, or New File into a results JSON for user review.

Yes, you need the jq dependency installed to execute the deterministic data collection scripts, scan-skills.sh and scan-rules.sh, which gather the skill and rule data required for the consolidation phase.

The best way is to cross-read and cluster principles by themes across all installed skills, comparing them against existing rules to generate actionable recommendations that revise outdated content or create new rule files.

No, the workflow provides a structured, auditable process that produces recommendations in a structured results JSON, requiring explicit user review before any changes are applied to existing or new rule files.

Principles may not convert if the deterministic data collection scripts fail to inventory installed skills properly, or if the LLM-based judgment phase lacks sufficient clustered themes to generate structured Append or Revise verdicts.