Content
75%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 well-organized, highly actionable skill: real commands, working-shaped code, flags tables, and a complete CI example in logical order. The gaps are a non-runnable Playwright option (networkidle2), mild redundancy between the Overview/References sections and inline citations, and no explicit verification step after a build run.
Suggestions
Fix the Playwright example's waitUntil value to 'networkidle' (Playwright does not accept the Puppeteer-only 'networkidle2'), so the flagship snippet is copy-paste executable.
Cut the duplicated material: the Overview SDK list repeats the description, and the References section restates links already cited inline — collapse one of them to save tokens.
Add a short verification step after "percy exec" (e.g., check the build URL/status output or run with --dry-run first) so users get a feedback loop when a build fails or PERCY_TOKEN is missing in CI.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with commands, tables, and code and assumes Claude's competence, but the References section restates links already cited inline, the Overview SDK listing duplicates the description, and the AI-Agents-mode paragraph is product-detail padding — minor trim opportunities rather than severe verbosity. | 4 / 5 |
Actionability | Concrete throughout: install commands, a snapshot code example, canonical "percy exec -- npx playwright test" invocations, a full flags table, and a copy-paste GitHub Actions workflow. It falls short of fully executable because the Playwright example passes "waitUntil: 'networkidle2'" — a Puppeteer option that Playwright rejects (it expects 'networkidle') — so the flagship snippet fails as written. | 4 / 5 |
Workflow Clarity | The install → author snapshot → percy exec → review diffs → CI sequence is clearly ordered with commands at each step, and the operation is non-destructive so the validation cap doesn't apply. Missing an explicit verification loop (e.g., checking build status or handling a failed build), which keeps it below the 5 anchor. | 4 / 5 |
Progressive Disclosure | No bundle files exist, so the whole skill lives in a well-sectioned single file with clear headers and a one-level References list. Structure and signaling are clearly good, but the ~200-line body (options and flags tables) is at the edge of what could be split into a reference file, and the under-50-line exception doesn't apply. | 4 / 5 |
Total | 16 / 20 Passed |