If you've accumulated dozens of K-skills but only use a handful, this interviews you about which agent you run, which skills are sacred, and which you know you never touch, then scans local transcripts and logs from Claude Code, Codex, OpenCode, and others to count actual trigger events over the past 30, 90, or 180 days. It ranks removal candidates by combining your answers with real usage evidence, produces a JSON report with remove/review/keep buckets, and only deletes directories after you approve the shortlist. The helper script handles agent-specific log formats and warns you when an agent has no readable local transcripts so you can export usage stats manually instead of guessing.
npx skills add https://github.com/nomadamas/k-skill --skill k-skill-cleaner