Content
96%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.
An exemplary single-file skill: fully executable JSON examples, feature-conditional decision trees, and explicit screenshot-verification feedback loops on every state-changing action. The only structural note is that everything lives in SKILL.md with no reference files, which is acceptable at this size but leaves navigation flat.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is entirely directive backend-specific constraints (feature flags, dispatch semantics, error codes) with zero explanations of concepts Claude already knows; nothing is trimmable without losing information. | 5 / 5 |
Actionability | Three complete copy-paste-ready JSON tool invocations (`element_action`, `ax_snapshot`, `ax_action`), an enumerated operations list (`press`/`focus`/`set_value`), concrete targeting decision rules, and exact error-code remediation ("run `/computer-use on` in the A0 Launcher chat"). | 5 / 5 |
Workflow Clarity | Sequences are explicit with feedback loops: "If an action reports ambiguity, take a fresh snapshot and narrow the target", "After any hide or minimize shortcut, inspect the fresh screenshot", and "stop immediately and ask the user" on rearm errors — matching the validate-then-retry checkpoint pattern. | 5 / 5 |
Progressive Disclosure | Sections (AX Targeting, Window Actions, Permissions) are clear and everything in the single file belongs there, but the ~100-line body has no reference split, so it matches the good-structure anchor rather than the reference-organized top anchor. | 4 / 5 |
Total | 19 / 20 Passed |