Content
85%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 dense, highly actionable skill body: executable code, exact CLI forms, and a well-sequenced verification loop with explicit validation gates. The main weakness is structure: everything lives inline in one long SKILL.md with no reference files, and a few guardrail rules are repeated.
Suggestions
Move the deep-dive material (material-array emissive cloning, the XR poke/select verification command catalog, the Windows PowerShell contact-sheet script) into references/ files and link them one level deep from SKILL.md.
State the image-open budget once as a single policy in the verification section instead of repeating it across 'Fast path' and 'Verify with a bounded loop'.
Consolidate the repeated anti-discovery warnings ('do not call --help', 'do not inspect node_modules') into one place near the verification commands.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every line is non-obvious, SDK-specific hard-won guidance (icon bundling limits, poke contactMs < 800 threshold, camera-facing rules) with no padding of concepts Claude already knows. Not 5 because the image-open budget is stated three times ("do not open any PNG, JPEG, or WebP", "artifacts/ui-runtime-review.jpg is the only image file to open", "do not open those files in model context") and several "do not call --help / do not inspect node_modules" warnings repeat. | 4 / 5 |
Actionability | Fully executable throughout: copy-paste TypeScript snippets (requireElementById, createSystem, material cloning), exact npx @iwsdk/cli commands with JSON arguments, and a complete PowerShell contact-sheet command covering the common verification cases. | 5 / 5 |
Workflow Clarity | Clear sequence (classify surface -> fast-path inspection -> author first pass -> connect runtime -> bounded verify loop) with numbered verification steps, an explicit final-gate checklist ("The final visual check must confirm..."), and concrete feedback loops (rerun the press when contactMs >= 800, use CLI error output instead of --help). | 5 / 5 |
Progressive Disclosure | Well-sectioned with headers, but the skill is a single ~325-line file with no references/, scripts/, or assets/ bundle; deep-dive content that belongs in separate one-level-deep files (material-array emissive cloning, the XR/browser verification command catalog, the PowerShell contact-sheet script) is inlined. Better organized than anchor 2, but inline bulk keeps it below 4. | 3 / 5 |
Total | 17 / 20 Passed |