Content
56%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 body delivers a genuinely useful, conservative audit policy with concrete tool routing and example patterns, and its workflow is clearly sequenced. However, it pays a large token cost for that value: pervasive internal repetition of the decorative-image rules and wholesale duplication of the reference file's highlights and safe patterns, much of which re-teaches frontend knowledge Claude already has.
Suggestions
Deduplicate 'Rule-Derived Highlights' and 'Known Safe Patterns' from references/categories.md — keep one canonical copy (in the reference file, linked early) and reduce the SKILL.md body to a short routing table.
Collapse the decorative-image guidance, which currently appears in Audit Stance, Coverage Checklist, Highlights, Safe Patterns, and Conflict Resolution, into a single 'Conservative findings' rule stated once.
Trim rules that restate standard web knowledge (GPU compositing of transform/opacity, prefers-reduced-motion, DOM read/write batching, table scope basics) and rely on MCP retrieval (get_rule/explain_rule) for that depth, per the skill's own 'Use MCP retrieval instead of trying to recall rules from memory' principle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose: the decorative-image guidance is repeated roughly five times ('Do not treat alt="" as an issue', the Coverage Checklist bullet, the Rule-Derived Highlights bullets, the Known Safe Patterns entry, and the Conflict Resolution section), and large portions of 'Rule-Derived Highlights' teach concepts Claude already knows (GPU compositing of transform/opacity, prefers-reduced-motion, table scope, batching DOM reads/writes). This matches 'Noticeably verbose; several unnecessary explanations or padded sections' rather than 3, where padding would be occasional rather than structural. | 2 / 5 |
Actionability | Guidance is mostly executable: the workflow names concrete MCP tools per situation ('Start with review_code', 'use audit_url for a public page', 'use get_rule, fix_rule, or explain_rule'), and 'Known Safe Patterns' includes copy-ready example snippets like `<img src="/divider.svg" alt="" aria-hidden="true">` and `export async function generateMetadata() {...}`. Not a 5 because no example tool invocations are shown, so 'common cases' are covered by policy text and snippets rather than full executable usage. | 4 / 5 |
Workflow Clarity | The Workflow section is a clear sequenced decision path (review_code/audit_url first, then narrow/expand with search_rules/list_categories/get_workflow, then remediate with get_rule/fix_rule/explain_rule), and the Evaluation Standard supplies a validation-style checkpoint ('When no issue is found automatically, use search_rules to broaden review coverage before concluding clean'). Minor validation gaps (no explicit retry/verify loop after remediation) keep it at 4 rather than 5. | 4 / 5 |
Progressive Disclosure | The reference is one level deep and real ('See references/categories.md' at the end), but it is signaled only once at the very bottom, and the two largest sections — 'Rule-Derived Highlights' and 'Known Safe Patterns' — are duplicated nearly verbatim inside references/categories.md, i.e., content that belongs in the separate file is fully inlined. This matches 'Some structure but could be better organized; references present but not clearly signaled; content that should be separate is inline'. | 3 / 5 |
Total | 13 / 20 Passed |