Content
75%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.
A well-structured advisory skill with an explicit three-phase workflow, a rule-loading decision table, a copy-paste report template, and concrete thresholds that make it highly actionable. Its main costs are token inefficiency — a Key Principles section restating WCAG/HIG numbers Claude already knows and a 'missing skill' instruction repeated three times — and referenced `rules/` files that are not present in the bundle to verify.
Suggestions
State the missing-skill fallback once (e.g., in a short 'Cross-skill integration' note) and reference it from the charting, visual-design, and screen-recorder paragraphs instead of repeating the near-identical sentence three times.
Trim the Key Principles quick reference to only the skill-specific judgments (response-time bands, 8dp/pt inter-target spacing) and push the well-known WCAG/HIG/Material numbers (4.5:1 contrast, 44pt targets, Hick's Law) into the corresponding rules/*.md files.
Either ship the referenced rules/*.md files in the bundle or note their provenance, and replace the deep relative cross-skill path (../../analysis/screen-recorder/rules/integrations.md) with a stable reference so the linkage survives relocation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body re-states standards Claude already knows in the Key Principles section (WCAG 4.5:1/7:1 contrast, 44pt iOS touch targets, Hick's Law 5-7 choices, 45-75 char line length), and the "a missing skill never blocks / log one line" instruction is repeated three times (lines 63, 65, 75), with the charting composition rule stated both in the table row and again in the following paragraph. Mostly efficient and dense with skill-specific judgment (response-time bands, severity criteria), so above 2, but the redundancy and known-concept re-explanation are more than minor, below 4. | 3 / 5 |
Actionability | Fully concrete guidance for an advisory skill: an exact command (`git diff --name-only HEAD~1`), a decision table mapping code contents to rule files, a copy-paste-ready report template with per-finding fields, a severity table with criteria and examples, and a model phrasing example ("Button on line 42 is 30x30px, below the 44pt iOS minimum"). Per the instruction-only scoring note, absence of code is not penalized when guidance is this actionable; not 4 because no key execution detail is missing. | 5 / 5 |
Workflow Clarity | A clear three-phase sequence (Context Discovery → Analysis → Report) with numbered steps and real checkpoints: "If ambiguous, ask the user", "Do not review code you haven't read", and explicit fallbacks for missing skills and missing stable handles. Not 5 because there is no output-validation or error-recovery loop on its own findings (e.g., re-verifying cited line numbers or contrast claims); not 3 because the checkpoints that exist are explicit rather than implicit. | 4 / 5 |
Progressive Disclosure | Good structure: SKILL.md is an overview with a well-signaled, one-level-deep mapping table from code contents to `rules/*.md` files, plus an always-in-context quick reference with a canonical-source pointer. Not 5 because the referenced `rules/` files are absent from the provided bundle (references cannot be verified as real files), the cross-skill link `../../analysis/screen-recorder/rules/integrations.md` is a fragile deep relative path, and the Key Principles section inlines substantial rule-level detail that overlaps the rule files; not 3 because the navigation and signaling are otherwise clear and clean. | 4 / 5 |
Total | 16 / 20 Passed |