barber

Audits and repackages Claude/Hermes skills — trims bloat, fixes weak triggers, catches duplication, fills real gaps, all with your approval.

Install

mkdir -p ~/.claude/skills/barber
curl -o ~/.claude/skills/barber/SKILL.md https://raw.githubusercontent.com/maxkilla/barber/main/SKILL.md

Hermes: same command, swap ~/.claude/skills/ for ~/.hermes/skills/. Barber detects which runtime it's in automatically.

What it does

An 8-part audit, one finding per line, on any SKILL.md:

1. Description — weak or missing triggers
2. Body — bloat, repetition, over-explaining
3. Structure — misplaced scripts/assets, broken references, generated files
4. Dependencies — unconditional installs, missing version pins
5. Safety — never trims confirmation/authority language
6. Generalization — triggers that are too narrow or too broad
7. Cross-skill dedup — redundant bodies, missing co-load links
8. Gaps (opt-in, --gaps) — what's dangerously missing, not just what's bloated

Every finding is shown before any edit. Nothing changes without approval.

Usage

• "audit this skill" — full pass
• "why isn't my skill triggering" — sections 1 & 6
• "which of my skills overlap" — section 7
• "tune up my skill library" — cheap pass across everything installed
• "enhance this skill" (--enhance) — bounded audit → approve → fix → retest loop, capped at 3 cycles

Flags: --check-only, --max-lines N, --tokens, --gaps, --enhance.

When to skip it

A single small skill that's already lean, or if you want automated CI regression testing — barber is a manual, in-conversation audit, not a test framework.

License

MIT — see LICENSE.