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.
Excellent instruction-only skill body: lean, highly actionable, with a clear sequenced workflow and validation/feedback loops for fragile GUI operations. The only gap is progressive disclosure — reference-grade material is inlined in a single SKILL.md with no separate reference files.
Suggestions
Move the full Survey return sample and the detailed coordinate-space/permission specs into a separate references/ file (e.g. REFERENCE.md) and link to it one level deep, keeping SKILL.md as a concise overview.
The menu-bar listing and AX-profile examples could live in that reference file with a brief inline summary, improving token efficiency on load.
Add a short 'See REFERENCE.md' navigation pointer where the Survey-return block currently sits so the inline example is clearly optional depth.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and efficient throughout; assumes Claude's competence, never explains what an AX tree or a menu bar is, and every section (tool table, path priorities, scenarios, rules) earns its place with no padding. | 5 / 5 |
Actionability | Fully executable guidance — concrete tool calls with exact params across six scenarios (e.g. `MacosAct { type: "intent", app: "WeChat", intent: "open_moments" }`) cover the common cases copy-paste ready. | 5 / 5 |
Workflow Clarity | The three-phase Survey→Plan→Act+Verify framework is explicitly sequenced with validation checkpoints (Observe after each Act), error-recovery feedback loops ('WINDOW_CHROME_BLOCKED → re-plan the 4 candidate paths'), and explicit stop-and-tell-user rules for fragile/sensitive cases. | 5 / 5 |
Progressive Disclosure | Well-organized sections and no bundle files exist (references/, scripts/, assets/ absent), but all reference-grade material — the full Survey return sample, menu-bar listing, and detailed coordinate-space spec — is inlined into one file rather than split into one-level-deep references. | 3 / 5 |
Total | 18 / 20 Passed |