What problem does it solve? Agent Skills accumulate redundant instructions, broken routes, hidden dependencies, and portability assumptions over time, and there is no structured way to review them. This Skill provides a disciplined, read-only audit workflow that evaluates a skill or skill library across five dimensions and only mutates files after an explicit housekeeping request. ## Core Features & Use Cases - Dimension-based routing: Load only the reference contracts you need—instruction quality, decomposition, routing metadata, implementation, or housekeeping—instead of reading everything. - Read-only audit by default: Gathers evidence from canonical sources, cites file and location, and never runs mutating commands during an audit. - Gated housekeeping mode: Applies selected findings only after explicit approval, with preimage backups, atomic writes, and partial-failure recovery. - Use Case: Point it at a skill folder that has grown bloated and say "boil this skill" to get a disposition list (KEEP, CUT, REWRITE, MOVE, SPLIT, ENFORCE) with severity and conflict annotations for every material instruction. ## Quick Start Ask the agent to audit the skill in a given directory, for example by saying "audit this skill in skills/my-skill and report instruction quality and routing issues".
