What problem does it solve? Over time an AI agent's skill collection inflates with near-duplicate skills and dormant entries, making discovery harder and triggers ambiguous. This Skill performs a periodic curation pass over the whole collection, refreshing usage data from transcripts and identifying clusters that should be consolidated — without ever deleting anything or acting without review. ## Core Features & Use Cases - Usage-driven state machine: Refreshes usage counters from transcripts and reports each skill as active, stale (30 days), or archived (90 days), with a grace floor for never-used skills. - Cluster detection and consolidation: Groups skills by prefix or domain keyword and proposes merging near-siblings into one umbrella skill with labelled subsections, or demoting narrow skills to reference/template/script files. - Safe staged proposals: Writes every merge, new-skill, or archive proposal to a pending directory for human review via /skills-review; only agent-proposed skills are ever moved, and dependency checks (launchd jobs, other skills, app config) prevent dangling references. - Use Case: After months of agent sessions, you have five pr-* skills that overlap. Run the curation pass to get a staged proposal merging them into one umbrella skill, with archive proposals for the absorbed siblings — nothing changes until you approve it. ## Quick Start Ask the agent to curate the skills collection and stage consolidation proposals for review.
