Content
42%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is well-organized with a useful output template and severity taxonomy, but it functions mostly as a restatement of common web-quality knowledge Claude already possesses rather than skill-specific procedure. The bundled analysis script is disconnected from the documented workflow, and there is no validation or feedback loop.
Suggestions
Cut the 'Audit categories' section to a short category list plus only thresholds Claude would not reliably recall (e.g., specific LCP/INP/CLS numbers), and move the full checklists into reference files under references/, linked one level deep.
Reference scripts/analyze.sh in the body and show its invocation (e.g., `bash scripts/analyze.sh <file_or_directory>`) plus its JSON output shape, so the audit has an executable first step.
Add a validation checkpoint to the workflow, such as re-running the analysis script or verifying each reported issue against the concrete file/line location before writing the final report.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~120-line 'Audit categories' section restates well-known web fundamentals Claude already knows ("LCP (Largest Contentful Paint) < 2.5s", "Minimum 4.5:1 for normal text, 3:1 for large text (WCAG AA)", "Unique title tags. 50-60 characters"), plus a generic 12-item weekly/monthly checklist — padded sections with no skill-specific knowledge. It is not a 1 because the output-format template and severity table do carry procedural value. | 2 / 5 |
Actionability | The 'Audit output format' template is concrete and copy-pasteable ("- **[Category]** Issue description. File: `path/to/file.js:123`"), but there is no executable guidance for performing the audit itself — no lighthouse CLI command, no tool invocation, and the bundled `scripts/analyze.sh` is never mentioned. This sits between anchor 3's 'missing key details' and anchor 4's 'concrete code or commands with minor gaps'; the missing run mechanism is more than a minor gap. | 3 / 5 |
Workflow Clarity | 'How it works' lists a rough 4-step sequence ("1. Analyze the provided code/project... 2. Categorize findings by severity... 3. Provide specific, actionable recommendations 4. Include code examples for fixes") but has no validation checkpoints and no feedback loop to confirm findings. Auditing is read-only so the destructive-cap does not apply; this matches anchor 3 ('steps listed but validation gaps; checkpoints missing or implicit') and not anchor 4, which requires most checkpoints present. | 3 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled ("[Performance Optimization](../performance/SKILL.md)"), but the body inlines large category checklists that belong in reference files, and the actual bundle file `scripts/analyze.sh` is orphaned — never referenced or explained in SKILL.md. Against the real bundle structure this fits anchor 3 ('content that should be separate is inline', references present but the bundle is not wired up), not anchor 4. | 3 / 5 |
Total | 11 / 20 Passed |