skill-cleanerlisted

# Skill Cleaner ## Goal Keep PM Copilot's skill layer small, discoverable, and safe to load by auditing prompt-budget usage, duplicate skill copies, unused candidates, and frontmatter description bloat. ## Workflow 1. Identify the roots to inspect: - Default roots are this repository's `skills/`, `~/.codex/skills`, `~/.codex/plugins/cache`, and `~/.agents/skills` when present. - Add extra roots with `--root <path>` when auditing archives, forks, or external skill collections. 2. Run the analyzer: ```bash python3 skills/skill-cleaner/scripts/skill_cleaner.py --months 3 ``` Useful variants: ```bash python3 skills/skill-cleaner/scripts/skill_cleaner.py --no-logs python3 skills/skill-cleaner/scripts/skill_cleaner.py --months 6 --max-log-mb 800 python3 skills/skill-cleaner/scripts/skill_cleaner.py --context-tokens 272000 --budget-percent 2 python3 skills/skill-cleaner/scripts/skill_cleaner.py --root /path/to/other/skills --no-logs ``` 3. Read the report in this order: - `Skill Budget`: rendered skill-list token estimate, 2% budget pressure, and minimum no-description cost. - `Description Candidates`: long frontmatter descriptions that may be shortened without losing trigger nouns. - `Duplicates`: same skill name or identical bodies across repo, Codex, plugin cache, and user skill roots. - `Unused Candidates`: skills with no recent `$skill`, `SKILL.md` read, or skill-path evidence in scanned Codex logs. - `Root Summary`: which roots supplied skills and