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.
A strong, opinionated fix-it skill: the symptom-table entry point, exact code per fix, baseline block, and never-ship checklist make it immediately actionable with a clear workflow. The only costs are modest token overhead from a scripted greeting, repeated hardware-testing emphasis, and a few lines re-explaining CSS basics Claude already knows, plus a single-file layout that inlines some content a reference file could hold.
Suggestions
Trim the 'Initial Response' scripted greeting and the Emil Kowalski attribution, and state the test-on-hardware rule once (section 11) while leaving one-line pointers to it elsewhere — this removes the main redundant tokens hurting conciseness.
Cut sentences that define CSS features Claude already knows (e.g., the meaning of `(hover: hover)` and `(pointer: fine)`) and keep only the non-obvious guidance, such as why both conditions are needed to rule out styluses.
Consider moving the framework-specific variants (Tailwind `future.hoverOnlyWhenSupported`, Next.js `viewport` export, PWA manifest/`apple-mobile-web-app-status-bar-style` details) into a one-level-deep reference file signaled from the relevant sections, slimming the main file while keeping the symptom table as the index.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is largely efficient — each fix is symptom, one-paragraph why, copy-paste code, caveat — but there are minor instances that could be trimmed: the scripted "Initial Response" greeting (with an Emil Kowalski attribution never used again), the "test on real hardware" rule repeated across Operating Posture, Hard Rule 5, the Symptom Table, section 11, and Never Ship, and a few lines explaining basics Claude already knows ("`(hover: hover)` means the primary input can hover"). This sits between the 'mostly efficient' and 'efficient with minor over-explanation' anchors, noticeably above the midpoint, so it scores 4 rather than 3. | 4 / 5 |
Actionability | Every fix ships exact, copy-paste-ready code (`@media (hover: hover) and (pointer: fine)`, `overscroll-behavior: contain` on inner scrollables, the two-scheme `theme-color` meta tags), the Baseline section is a complete starter block, and hardware debugging is given as concrete commands ("chrome://inspect", "Safari → Develop → the device", dev server on `0.0.0.0`). This matches the anchor for fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | The flow is explicitly sequenced — "Start here. Match what the user is seeing, then read the matching section" — and ends with an explicit self-check checklist ("Self-check before you finish" / Never Ship table) plus verification steps (section 11's hardware procedure and the Output section's "What needs a phone" line). The skill involves no destructive or batch operations, so the validation cap does not apply, and its checklist-and-verify loop matches the top anchor for this single-purpose fix-it shape. | 5 / 5 |
Progressive Disclosure | There are no bundle files at all (no references/, scripts/, or assets/), so there is no reference nesting to penalize, and the Symptom Table functions as a clean internal navigation index to the eleven numbered fix sections. However, at roughly 310 inline lines it exceeds the under-50-line simple-skill exception, and material like the per-framework notes (Tailwind v3/v4, Next.js viewport export, PWA manifest) could plausibly live one level deep in reference files, which keeps it at 'good structure, minor organization gaps' rather than the fully split top anchor. | 4 / 5 |
Total | 18 / 20 Passed |