Content
76%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 tight, well-structured instruction skill that encodes genuinely project-specific conventions and gives concrete detection rules with one executable idiom. Its main weaknesses are the total absence of validation after applying fixes and a reference to a rules file that is not present in the bundle.
Suggestions
Add a verification step after fix=true (e.g. run typecheck/tests and confirm no behavior change) so batch edits have a feedback loop.
Inline the essential content of `.claude/rules/sim-hooks.md` (sentinel-on-mount, prev-tracker pattern) into the skill, or move it into a bundle references/ file so the guidance is self-contained.
Add brief before/after code pairs for the highest-frequency patterns (derived state and server-state-in-useState) to make fixes unambiguous.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and every section earns its place: the Context section encodes project-specific stack ownership ("React Query owns server data, nuqs (URL) owns shareable view-state, Zustand owns global client state") that Claude cannot infer generically, and the seven anti-pattern entries are terse and specific. There is no padding or explanation of concepts Claude already knows. | 5 / 5 |
Actionability | Mostly executable guidance: concrete detection rules ("Never `useState` + `useEffect` to sync React Query data"), a copy-paste-ready derive idiom (`items.find(i => i.id === selectedId)`), and specific remediation ("reset with a remount `key`"). Falls short of 5 because fixes for several patterns (derived state, chained Effects) lack before/after examples and there is no instruction to verify fixes (e.g. typecheck/tests). | 4 / 5 |
Workflow Clarity | The three steps (read references, analyze scope, apply or propose fixes) form a clear, unambiguous sequence, but there are no validation checkpoints at all. Since fix defaults to true and the scope can be "whole codebase", batch code edits proceed without any verification step, which caps this dimension at 3. | 3 / 5 |
Progressive Disclosure | At ~30 lines with no bundle files, the body is well-organized into Context, References, Anti-patterns, and Steps, and the three URL references are clearly signaled with their purpose. However, anti-pattern 3 points to `.claude/rules/sim-hooks.md`, which is not part of the skill bundle, making it a dangling reference that a reader cannot resolve from the skill itself. | 4 / 5 |
Total | 16 / 20 Passed |