Content
100%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 lean skill body: every step is an executable command, validation criteria are stated explicitly, error recovery is provided for the known failure mode, and detail is properly delegated to the agent-browser skill. Nothing taught that Claude already knows, and nothing essential is missing.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~20-line body contains only non-obvious, repo-specific facts ('Chrome for Testing is not an npm artifact', 'bun install in CI does not download Chrome') and judgment guidance ('An unnamed control is a bug'); no concept re-teaching or padding — every token earns its place. | 5 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common cases: 'bunx agent-browser open <url>', 'set viewport <w> <h>', 'snapshot -i', 'screenshot --annotate page.png', 'set media dark', 'diff screenshot --baseline baseline.png', plus concrete setup and server-start commands. | 5 / 5 |
Workflow Clarity | The numbered sequence (start server → open → viewport → snapshot → screenshot → dark mode → diff) includes explicit validation criteria ('An unnamed control is a bug', 'Browser-console warnings in a page or story are bugs') and an error-recovery loop ('--with-deps on Linux if launch fails on missing libraries'). This is a read-only verification skill, so the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | Under 50 lines with no bundle files needed; sections are well-organized (Set up / Steps), the full command guide is delegated one level deep to the clearly-signaled agent-browser skill rather than duplicated, and repo paths (docs/local-https.md, scripts/install-agent-browser-chrome.ts) are each referenced with their purpose. | 5 / 5 |
Total | 20 / 20 Passed |