Content
71%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 genuinely lean, actionable audit skill: numbered workflow, per-ecosystem grep commands, quantified thresholds, and a scoring rubric. The main defects are a dangling 'Priority: P1' heading, a missing decision threshold for the backend leakage grep, and a bundle organization gap where detection.md is orphaned from SKILL.md while its content is partially re-inlined in the body.
Suggestions
Link references/detection.md from the body (e.g., in section 2, 'See [framework detection](references/detection.md) for the manifest-to-source-directory mapping ($SRC) for each ecosystem') instead of hardcoding per-ecosystem paths like lib/widgets and src/controllers, which both duplicates and fragments the reference's systematic table.
Remove or contextualize the orphaned '## **Priority: P1 (HIGH)**' heading, and give the backend leakage grep an explicit decision threshold like the Web rule (e.g., 'count > 0 per controller = High Debt') so all three ecosystems have actionable rules.
Tighten awkward phrasing that impedes execution: replace 'Resources > 1,000 lines require granulation' with plain language (e.g., 'split resource files over 1,000 lines'), and rewrite the double-negative Anti-Patterns entries as direct imperatives.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean: it never explains concepts Claude already knows (no 'what is SRP' padding), and each section is a compact rule set with thresholds and one-line grep commands. Not a 5 due to minor noise that could be trimmed: the orphaned '## **Priority: P1 (HIGH)**' heading carries no actionable content, and awkward phrasing ('Resources > 1,000 lines require granulation', 'No applying generic patterns over project-specific rules') forces re-reading. | 4 / 5 |
Actionability | Mostly executable guidance: concrete grep pipelines per ecosystem (e.g. 'grep -rE "useEffect|useState|useMemo" components --include="*.tsx" | wc -l') plus decision rules ('components/ hook count > 20x hooks/ folder, architecture monolithic', '> 1,500 lines indicates "God Class"'). Not a 5 because the Web and Mobile rules have thresholds while the Backend grep yields a count with no action threshold ('Controllers must only handle request parsing' is a principle, not a decision rule), and step 1's duplication detection is described rather than given as a runnable command in the body. | 4 / 5 |
Workflow Clarity | A clear numbered sequence (1 duplication → 2 logic leakage → 3 monoliths → 4 resources) terminates in a quantified 'Scoring Impact' rubric (-15 per layer violation, -10 per monolith), which serves as the output checkpoint. Not a 5 because there are no explicit validation or verification steps for the audit findings (e.g., confirm flagged files against remediation patterns before reporting), though the read-only nature of the audit keeps this at minor-gap level rather than triggering the destructive/batch cap. | 4 / 5 |
Progressive Disclosure | Structure exists and the two cited references (references/implementation.md, linked three times with section-specific context; references/PATTERNS.md) are clearly signaled and one level deep. However, the bundle's references/detection.md — a systematic manifest-to-directory-structure table — is never referenced from SKILL.md, while the body hardcodes per-ecosystem paths (lib/widgets, src/controllers, components/) that this file systematically maps. That is inline content that should live behind a clearly signaled reference, matching 'some structure; references present but not clearly signaled; content that should be separate is inline' more closely than the anchor-4 'minor organization gaps'. | 3 / 5 |
Total | 15 / 20 Passed |