Content
92%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, highly actionable workflow skill: exact commands with flags and values, an explicit inspect-before-send validation loop, and honest-labeling rules. The single weakness is that the headless capture tool's JS-step interface is mentioned but never specified.
Suggestions
Show one complete capture.mjs invocation with a click/scroll step (e.g. an env var or argument example) so the 'JS steps to click, open or scroll' mechanism is executable rather than implied.
Add a one-line fallback for projects without a review fixture or debug hook for rule 4's state setup, since that rule currently assumes such a mechanism exists.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~48-line body is dense and padding-free: no explanations of concepts Claude already knows, just rules, exact commands ('scripts/capture.mjs <url> <out.png>'), viewport sizes (1600×900, 390×844), and settle timings. Even the rhetorical lines ('A screenshot mentioned but not sent doesn't count') enforce a rule rather than pad. | 5 / 5 |
Actionability | Commands are mostly copy-paste ready ('scripts/compare.py out.jpg before.png "Before" after.png "After"') with concrete flags and values, but 'JS steps to click, open or scroll' never specifies how those steps are passed to capture.mjs — a minor executable gap that fits the 'minor gaps' anchor rather than the fully-copy-paste-ready 5. | 4 / 5 |
Workflow Clarity | The sequence (capture each pass → inspect every image → caption → send → final report) is explicit and includes a real validate→fix loop: 'Look at every image before sending' with a concrete recapture checklist (stray toasts, half-loaded art, wrong scene), plus error surfacing in the capture tool. | 5 / 5 |
Progressive Disclosure | A sub-50-line body with clean sections (Rules, How to capture, In the final report) and two script references (capture.mjs, compare.py) that are real files used as executable tools rather than nested documentation — one level deep, clearly signaled, well organized, which fits the simple-skill top anchor. | 5 / 5 |
Total | 19 / 20 Passed |