Content
93%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 tight, well-sequenced instruction sheet that gives executable commands for every phase of the workflow and explicitly defers detail to the tool's own interactive hints and help system. The only notable gap is that emulator booting and mid-drive failure recovery are implied rather than written. Overall a strong, token-efficient skill body.
Suggestions
State explicitly how the emulator is started (e.g., whether 'agent-device open' boots an attached virtual device automatically or a separate boot step is needed), since the shutdown line 'if your session booted the emulator' presupposes a boot mechanism that is never described.
Add one line on recovering from a failed 'open' (e.g., package not installed or no device attached), mirroring the existing explicit handling of the unavailable-CLI case, so the mid-drive failure path is as well covered as the startup path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~35-line body is lean with no explanations of concepts Claude already knows; every line carries behavioral instruction ('Do not probe first with --help...', 'Treat installation and upgrades as user-owned setup steps'). The '@latest' tag is not a time-sensitive version pin, so no conciseness penalty applies. | 5 / 5 |
Actionability | Commands are fully executable and copy-paste ready: 'npm install -g agent-device@latest', 'agent-device open <app-or-package-id> --platform android --foreground', 'agent-device close --shutdown', plus five concrete help-topic routes. Delegating mid-drive commands to the tool is explicitly written ('Follow the initial interactive snapshot and corrective error hints'), a justified flexibility rather than a gap. | 5 / 5 |
Workflow Clarity | The lifecycle is clearly sequenced (install precondition → open with platform flag → follow snapshot and corrective hints → conditional 'agent-device close --shutdown') with an explicit error-recovery loop for the unavailable-CLI case. It falls short of 5 because how the emulator gets booted and how to recover from a failed 'open' are left implicit rather than stated as checkpoints. | 4 / 5 |
Progressive Disclosure | At under 50 lines with no bundle files, the body keeps the main path lean and defers specialized detail to one-level-deep, version-matched help topics gated behind an explicit condition ('Only when the task is specialized or a command shape is unclear'), which is progressive disclosure done correctly. No references are buried and no bulk content is inlined. | 5 / 5 |
Total | 19 / 20 Passed |