Content
71%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 an unusually dense, high-signal decision guide: concrete limits, API names, and verification loops with essentially zero filler. Its main weaknesses are structural — no progressive disclosure despite reference-grade inline material — and the absence of one complete worked widget_code example.
Suggestions
Move reference-grade inventories (allowed HTTPS origins list, the openclaw.data/action/cron/state API reference, and the full CSS custom-property list) into a references/ file, keeping SKILL.md as an overview with clearly signaled one-level-deep links.
Add one minimal complete widget_code example (a small HTML/SVG fragment with a scoped root ID and a textContent-based label) so the most common single case is copy-paste ready.
Tighten the verify section by consolidating its distributed checkpoints into a short ordered checklist (fix syntax errors → re-call → verify assets/controls/width → follow result.presentation).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every sentence carries non-obvious platform facts Claude cannot know (262,144-char widget limit, allowed CDN origin list, host CSS tokens, 4,000-char prompt cap) with no explanation of general concepts. It falls short of 5 because a handful of sentences over-qualify or restate and could be trimmed (e.g., the browser-profile sign-in nuance and repeated surface-availability caveats). | 4 / 5 |
Actionability | Concrete, executable guidance throughout: exact API signatures ("openclaw.data.read(bindingId, params?)"), hard limits, a copy-paste D3 script tag, named CSS classes and custom properties, and precise capability declarations. Not 5 because there is no complete minimal widget_code example tying the pieces together, leaving a small gap for the most common case. | 4 / 5 |
Workflow Clarity | A clear sequenced flow (choose surface → author content → libraries/theme → dashboard actions → verify and deliver) with explicit validation and feedback steps ("Fix reported inline-script syntax errors and call the tool again", "Check capabilityState and report pending or rejected access", "exercise the data read in the actual pinned frame"). Not 5 because checkpoints are distributed across prose rather than consolidated into an explicit ordered checklist or recovery loop. | 4 / 5 |
Progressive Disclosure | Sections are well organized, but at ~158 lines everything lives inline in SKILL.md with no reference files at all — reference-grade material such as the full CDN origin list, the host API reference, and the complete theme token inventory is inlined rather than split out. This matches anchor 3 (content that should be separate is inline); not 4 because there is no disclosure structure beyond the single file. | 3 / 5 |
Total | 15 / 20 Passed |