Content
78%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 deliberately minimal router skill that delegates volatile details to the CLI's own help output, which keeps it lean and current. It provides concrete commands and a clear sequenced loop with setup validation, with only minor gaps in mid-loop checkpoints and slight phrasing redundancy.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence, but the 'stop and tell the user to ... or approve an exact-version npm command' instruction repeats nearly verbatim, and the time-sensitive version pin ('agent-device >= 0.14.0') sits in the main flow rather than a deprecated/old-patterns section. This is efficient with minor trims possible, not the every-token-earns-its-place top anchor. | 4 / 5 |
Actionability | Provides copy-paste-ready commands ('agent-device --version', 'agent-device help dogfood') and a concrete loop with subcommands ('snapshot -i + screenshot', 'close --shutdown'). It is not 5 because evidence-capture specifics and report shape are deferred entirely to external help output, leaving minor gaps. | 4 / 5 |
Workflow Clarity | The sequence (version gate -> read help -> open/snapshot/explore/capture/close) is clear, with an explicit setup validation checkpoint and failure feedback ('If that fails, stop and tell the user...') plus conditional shutdown handling. It falls short of 5 because the exploration loop itself lacks mid-loop validation checkpoints or error-recovery guidance. | 4 / 5 |
Progressive Disclosure | The skill is under 50 lines, needs no external reference files (none exist in the bundle), and is organized into clear topical paragraphs. Its single delegation to 'agent-device help dogfood' is a clearly signaled, one-level-deep reference, which satisfies the simple-skill exception for a top score. | 5 / 5 |
Total | 17 / 20 Passed |