A Claude Code skill that audits and improves other Claude Code skills.
Anthropic's official skill-creator handles the mechanics of building a skill: folder scaffolding, frontmatter validation, plugin registration. skill-optimizer runs alongside it and owns the quality layer: description triggering, naming, conciseness, anti-patterns, and gotcha rigor.
Five common failure modes that kill skill quality, with a concrete fix for each:
- Description doesn't trigger — vague, narrow, or in the wrong voice. Fix: specific, third-person, loaded with
Use when...trigger phrases. - Over-defining the process — railroading instead of guiding. Fix: match freedom to task type.
- Stating the obvious — wasting tokens on what Claude already knows. Fix: challenge every paragraph for net new signal.
- Missing gotcha section — failures undocumented. Fix: capture every failure pattern. This is where the skill earns its keep.
- Monolithic blob — everything in one file. Fix: SKILL.md under 500 lines; move references to separate files.
Drop the SKILL.md into your Claude Code skills directory (project-scoped or user-scoped). When you ask Claude to create, edit, or improve a skill, this skill loads alongside skill-creator and adds the quality criteria that skill-creator deliberately leaves open.
Built after the principles in The AI Daily Brief's April 2 Skills Master Class (Nufar Gaspar and Nathaniel Whittemore) made clear that most skill failures aren't mechanical — they're rooted in description craft, freedom-matching, and gotcha discipline. skill-creator covers the mechanics; this fills the rest.
MIT.