Content
73%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.
The body is a well-sequenced, highly actionable instruction skill with explicit validation and a genuinely executable bundled checker whose behavior matches the documentation. The main costs are token weight in the inline evidence-record specification and the missing worked example that would let an agent produce a valid record on the first try.
Suggestions
Move the detailed evidence-record field specification into a bundled reference file (e.g., references/evidence-record.md) next to the checker, keeping only the command and a compact field summary in SKILL.md.
Add one worked example of a passing local-report.json (with two viewports and a not-applicable state) so the record can be written correctly without reverse-engineering the checker.
Trim the viewport-label matching rules ("375 px matches 375px...") that restate checker internals; the checker's rejection messages already teach these on failure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, imperative, and free of padding about concepts Claude already knows, but it can be tightened: the inline evidence-record field spec (contract, targetViewports, viewport-matching rules, contentStates, renderedEvidence) spans ~25 lines and duplicates semantics the bundled checker already enforces. That fits anchor 3 ("could be tightened") better than anchor 4's minor-trim bar. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: an exact checker invocation ("node <installed-skill>/scripts/check-evidence.js <local-report.json>"), exact record property names, a state/feedback table, and specific anti-patterns ("purple gradients", "cards inside cards"). The gap versus anchor 5 is the absence of a copy-paste example JSON record covering the common case, which would make the evidence workflow fully turnkey. | 4 / 5 |
Workflow Clarity | A coherent multi-step pipeline (choose direction → record contract → build → review checklist → rendered finish gate) with an explicit validation step (the evidence checker), a feedback loop ("Fix failures, then repeat affected checks after structural changes"), and a checklist for a complex process. This matches the anchor-5 pattern of clear sequencing with explicit validation and error-recovery loops. | 5 / 5 |
Progressive Disclosure | Well-sectioned with a real, verified one-level-deep bundle reference (scripts/check-evidence.js) invoked with an exact path, plus one external reference URL. The main gap versus anchor 5 is that the detailed JSON record schema is inlined in SKILL.md where a short reference file next to the checker would keep the overview leaner. | 4 / 5 |
Total | 16 / 20 Passed |