Content
81%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 high-quality, action-dense skill body: a numbered workflow with per-step verification, concrete tool calls with arguments, worked examples, and an explicit recovery pattern with feedback loops. Its remaining flaws are minor — repeated permission-modal guidance across three sections and example content that could be split out or tightened.
Suggestions
Consolidate permission-modal guidance (currently in workflow step 2, Recovery Pattern, and Tips) into a single place to remove redundancy and trim tokens.
Move the worked login/scroll examples into an examples reference file or tighten them to one canonical example, freeing SKILL.md to stay a lean overview.
Align the login example with the workflow's own guidance by showing describe/debugger-component-tree coordinate discovery before the tap steps, instead of raw guessed coordinates.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean — dense tables, terse bullets, no explanation of concepts Claude already knows. Not 5 because permission-modal handling is stated three times (workflow step 2, Recovery Pattern, and Tips), which is redundant trimming material. | 4 / 5 |
Actionability | Concrete tool invocations with arguments ('gesture-tap { x: 0.5, y: 0.4 }', 'gesture-swipe { fromY: 0.7, toY: 0.3 }', 'adb -s <serial> reverse tcp:8081 tcp:8081') plus worked login and scroll examples. Not 5 because some snippets are shorthand (e.g., 'paste { text: ... }' is illustrative rather than copy-paste ready) and the login example uses guessed coordinates while the workflow mandates discovery tools first. | 4 / 5 |
Workflow Clarity | A clear 5-step sequence (baseline screenshot, discover, interact, verify, repeat) with explicit verification checkpoints at every step, plus a Recovery Pattern with feedback loops: retry-with-new-coordinates, wait-and-retake on transitional states, and a stop condition after two failed taps. No destructive/batch cap applies. | 5 / 5 |
Progressive Disclosure | Well-organized sections with clearly signaled one-level references to sibling skills (platform setup table, Related Skills table with 'When to use'). Not 5: at ~90 lines, the Examples section and repeated permission-modal guidance could be consolidated or externalized, and the stray '---' divider between numbered sections 3 and 4 slightly disrupts navigation. | 4 / 5 |
Total | 17 / 20 Passed |