Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, actionable conventions skill with a clear decision procedure and clean one-level-deep deferral to the canonical reference. Its single weakness is conciseness: the motivational prose carries rhetorical padding that could be trimmed without losing the rule.
Suggestions
Tighten the motivation in 'Hooks live at the edge': collapse the rhetorical negation-lists ('not X, not Y, not Z') into a single sentence stating the prevention-over-discipline principle.
Consider moving the extended rationale ('Features you tuned will get silently dropped... no bisect helps, because nothing was ever measurable') into docs/contributing/react.md, keeping only the load-bearing rule and the two-situation test inline.
The 'short version' section usefully duplicates the rules for scanning; ensure it stays a true summary and does not re-expand into prose as the skill grows.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The guidance is accurate and domain-specific (not basic concepts Claude already knows), but the motivational prose is padded with rhetorical negation-lists ('not test your hooks better, not be more careful with deps, not add a renderHook suite'; 'discipline fails at the third layer of composition; prevention does not') that could be tightened. It is above level 1 because it does not explain generic concepts, but below level 3 because not every token earns its place. | 2 / 3 |
Actionability | Concrete, specific rules with named APIs (useSyncExternalStore, useContext), a two-situation decision test, fine/never example pairs, and copy-paste-ready data-testid values (sidebar-right-inspect-node-properties). Per the code-vs-instruction scoring note, absence of code blocks is not penalized when the guidance is this actionable, placing it above the pseudocode/incomplete level 2. | 3 / 3 |
Workflow Clarity | The decision procedure is explicit and sequenced: 'A hook is acceptable in exactly two situations: 1... 2... If a hook fails both — extract the logic out... and reduce the hook to a three-line call site.' No destructive/batch/XML operations are involved, so the missing-validation cap does not apply, keeping it above level 2. | 3 / 3 |
Progressive Disclosure | Sections are well-organized, the load-bearing rule is inline, and the full reference is deferred exactly one level to docs/contributing/react.md with explicit signaling ('The rule above is the load-bearing part'), plus a sibling-skill cross-link. References are clearly signaled and one level deep, so it is above the level 2 'references present but not clearly signaled' anchor. | 3 / 3 |
Total | 11 / 12 Passed |