Skill Review
Audit an Agent Skill directory against the Agent Skills Specification and quality criteria. Produce a structured pass/fail report with specific fix instructions for every failure.
Context Loading
Load only the reference files needed for the current check group. Do NOT load all references upfront.
| Check Group | Load | Do NOT Load |
|---|---|---|
| Spec Compliance | references/checks-spec.md | checks-references, checks-scripts, checks-structure |
| Reference Quality | references/checks-references.md | checks-spec, checks-scripts, checks-structure |
| Script Standards | references/checks-scripts.md | checks-spec, checks-references, checks-structure |
| Structure & README | references/checks-structure.md | checks-spec, checks-references, checks-scripts |
| Report Generation | references/report-template.md | All checks-*.md files |
Skip guidance based on skill contents:
| Condition | Skip | Why |
|---|---|---|
No scripts/ directory | Script Standards check group | Nothing to audit |
| No workflow files | Workflow metadata checks in checks-structure.md | Nothing to audit |
| Fewer than 3 procedural steps | Task tracking check in checks-structure.md | Overhead exceeds value |
Review Process
Use TaskCreate to track progress — this review has 8 check groups plus locate and report steps, and always qualifies for task tracking.
1. Locate the Skill
Resolve the target skill path. Confirm SKILL.md exists at the root. If it does not exist, fail immediately — no skill to review.
2. Spec Compliance
Load references/checks-spec.md. Validate the name field, description field, frontmatter structure, and body content against the Agent Skills Specification.
3. Reference Quality
Load references/checks-references.md. Audit every file in references/ for topic coherence, citation presence, and self-containment. Use WebFetch to verify every URL found in reference files — record each URL's reachability status.
4. URL Reachability
This is part of the Reference Quality check group but reported separately. For every URL found across all skill files, fetch it with WebFetch and record whether it returns content or fails.
5. Progressive Disclosure
Load references/checks-structure.md. Count the lines in SKILL.md (report the actual number). Verify that detailed content lives in reference files, not in the main SKILL.md. Confirm reference loading is conditional (table or if/then pattern), not a blanket "load everything" instruction.
6. Workflow Files
Load references/checks-structure.md (workflow metadata section). If workflow files exist, verify each declares execution:, parallelism:, and needs-user-interaction: metadata. If no workflow files exist, mark as N/A.
7. Task Tracking
Load references/checks-structure.md (task tracking section). If the skill describes 3 or more procedural steps, verify that SKILL.md instructs the consuming agent to use TaskCreate. If fewer than 3 steps, mark as N/A.
8. Script Standards
Load references/checks-scripts.md. If a scripts/ directory exists, audit every script against shell and script standards. If no scripts exist, mark as N/A.
9. README
Load references/checks-structure.md (README section). Verify README.md exists and meets all completeness criteria.
10. Generate Report
Load references/report-template.md. Compile all check results into the report format. List every failure with the specific file path and what to change. Separate required fixes from optional recommendations.
Rules
- Never hallucinate spec requirements — every check must trace to a cited source URL
- Never use curl or wget — use WebFetch for all URL verification
- Never reference other skills in this repository — this skill is functionally independent
- Report facts, not opinions — pass/fail against stated criteria, not subjective quality judgments
- Include actual values — when a check measures something (line count, character count), report the measured value alongside the threshold
- Mark inapplicable checks as N/A — do not fail a skill for lacking scripts when it has no scripts directory