Content
31%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 skill has a clear high-level sequence and some concrete thresholds, but it is severely bloated: it re-enumerates standard vulnerability and quality knowledge Claude already has, inlines simulated output transcripts with fabricated metrics, and keeps everything in one 535-line file with no reference structure. The batch-fix workflow also lacks any failure/recovery feedback loop.
Suggestions
Cut the ~170 lines of simulated example transcripts (fabricated issue counts, metrics, and Lighthouse scores) and the lists of well-known vulnerability types; keep only guidance Claude would not generate on its own, such as the project-specific thresholds and prioritization rules.
Move the audit report template, the long checklists, and the detailed fix examples into separate reference files (e.g. references/report-template.md, references/checklists.md) and link them from a lean SKILL.md overview.
Add an explicit validation feedback loop for the batch-fix workflow: after each fix category run the test suite, and if tests fail, revert or repair the change before proceeding — rather than only a single 'run all tests' step at the end.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 535-line body extensively catalogs issues Claude already knows (SQL injection, N+1 queries, XSS, hardcoded secrets, TODO/FIXME comments) and devotes ~170 lines to simulated output transcripts with fabricated metrics ("analyzing 247 files", "Lighthouse 62 → 94", "42% → 87% coverage") that provide no actionable instruction — severely verbose and heavily padded. | 1 / 5 |
Actionability | There are some concrete signals (god classes >500 lines or >20 methods, complexity >10, bcrypt 12 rounds, coverage >80%, before/after code diffs), but the core workflow is described rather than instructed ("Fix everything automatically", "Use `strReplace` to fix issues in files") with no executable commands or tool specifics for the scan/fix loop. | 3 / 5 |
Workflow Clarity | Steps 1–4 provide a clear Discover → Detect → Fix → Verify sequence and Step 4 says "Run all tests to ensure nothing broke", but this is a large batch of code modifications (~89 files in the example) with no feedback loop — no guidance for what to do when tests fail, no per-fix verification, and no rollback strategy, capping workflow clarity at 3 per the batch-operation guideline. | 3 / 5 |
Progressive Disclosure | No bundle files exist and the body contains no file references at all; ~250 lines of example transcripts, a 40-line report template, and full audit checklists are all inlined in a single monolithic SKILL.md where they clearly belong in separate reference files — content that clearly belongs in separate files is inlined. | 2 / 5 |
Total | 9 / 20 Passed |