jwmossmoz/skill-checker

skill-checker

Audit a skill directory with validators and eval-artifact inventory. Eval evidence is advisory; use skill-creator for the actual eval loop.

~/.claude/skills/skill-checker/scripts/check-skill.sh <skill-path> [--json]

USE FOR:

Skill validation, quality audits, readiness checks, and diagnosing checker output.

DO NOT USE FOR:

Running evals, creating or iterating skills, or Markdown linting.

Checks

• skills-ref: hard spec.
• waza: token budget/advisories.
• skill-validator: structure, links, density, contamination.
• skill-check: quality/security.
• eval-evidence: detects evals/evals.json, trigger-loop files, eval.yaml, and sibling benchmark workspaces.

Treat one complaint as a hint. Treat the same issue from two or more validators as a fix candidate.

Examples

• "Check skills/taskcluster": run the script and summarize overlapping issues.
• "Is this production-ready?": require validator health plus eval evidence.

Readiness

For production-ready substantive skills, expect hard validators to pass, accepted warnings explained, waza Medium or better, and evals/evals.json plus trigger-loop results or benchmark evidence.

Troubleshooting / Gotchas

• Validator versions drift; distrust a single-tool regression unless skills-ref also flips.
• skill-validator may warn that evals/ is unknown. This repo intentionally keeps eval files there.
• The checker reports missing or weak eval evidence, but it never spawns runs, grades assertions, aggregates benchmarks, or launches the viewer.

Detailed interpretation notes and validator-finding → spec-rule mapping live in references/usage.md. Read references/waza.md for waza-specific workflows and WAZA-AUDIT.md guidance. Implementation is scripts/check-skill.sh.