rules-distill

What problem does it solve? As a skill library grows, the same principles get repeated across many skills while the central rules files drift out of date. This Skill scans all installed skills, finds principles that appear in two or more of them, and proposes concrete additions or revisions to your rules files — with mandatory user approval before anything changes. ## Core Features & Use Cases - Deterministic Inventory: Bash scripts scan skill frontmatter and rule file headings, producing structured JSON inventories of both. - LLM Cross-Read Analysis: Skills are grouped into thematic clusters and analyzed in batches against the full rules text, producing verdicts such as Append, Revise, New Section, New File, Already Covered, or Too Specific. - User-Gated Execution: Every candidate is presented with evidence, violation risk, and draft text; rules are never modified without explicit approval, and results are saved to results.json. - Use Case: After installing a batch of new skills, run a monthly maintenance pass to detect that three skills independently describe LLM output validation, then promote that principle into coding-style.md as a new section. ## Quick Start Ask the assistant to run rules-distill to scan all installed skills and propose updates to the rules files.