Content
88%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.
An exemplary process-oriented skill body: a measured, validated 5-step workflow with concrete code, thresholds, decision tables, and a verification checklist, plus well-signaled one-level-deep references. The only notable gaps are a repeated reference to a performance-checklist file not present in the bundle and a few prose-heavy sections that could be moved or tightened.
Suggestions
Fix or remove the dangling '../../references/performance-checklist.md' path cited three times — either ship the file in the bundle or fold its unique content into the existing references file, since a reader following it hits a dead end.
Move the extended caching and index-tuning rules (key composition, layer choice, stampede guards, statistics staleness) into optimization-patterns.md alongside their worked fixes, keeping only the signature and one-line rule in SKILL.md.
Trim the 'Common Rationalizations' table to the rows that are not already implied by the 'When NOT to use' and Red Flags sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 267-line body is almost entirely operational content Claude does not already know — threshold tables (Core Web Vitals, budgets), a symptom decision tree, decision tables, and concrete rules — and it avoids explaining basic concepts. It is not a 5 because a few sections could be trimmed: the ~180-word 'Missing caching' bullet is dense prose, the 11-row 'Common Rationalizations' table elaborates responses Claude could generate, and Step 4 includes rhetorical padding like 'This is the step teams skip: the change is already written, throwing it away feels wasteful'. | 4 / 5 |
Actionability | Copy-paste-ready code exists for the common cases ('import { onLCP, onINP, onCLS } from \'web-vitals\'', 'console.time/timeEnd', 'npx bundlesize', 'npx lhci autorun') alongside exact numeric targets, per-symptom investigation trees, and a keep-or-revert decision table. Guidance is specific down to rules like 'sized so instances × max stays under the database\'s connection ceiling'. This matches the anchor-5 'fully executable, specific examples cover the common cases' standard. | 5 / 5 |
Workflow Clarity | The 5-step workflow (MEASURE → IDENTIFY → FIX → VERIFY → GUARD) is clearly sequenced with explicit validation: re-measure under identical conditions, compare against run-to-run variance, a strict keep-or-revert table, a feedback loop ('When either guard fires, return to Step 1'), an attempt ledger for error recovery, and a final 14-item verification checklist. This is the anchor-5 example (explicit validation steps, feedback loops, checklists) fully realized. | 5 / 5 |
Progressive Disclosure | Good structure: SKILL.md states each rule and signature with one-level-deep links to worked fixes in references/optimization-patterns.md, whose anchors (#n1-queries-backend, #queries-that-ignore-their-index, etc.) all match real headings in that file. Not 5 because the externally referenced '../../references/performance-checklist.md' (cited three times, including in 'See Also') does not exist in this bundle — a dangling path — and some dense inline material (the caching and index-tuning bullets) reads as reference-level detail that belongs in the patterns file. | 4 / 5 |
Total | 18 / 20 Passed |