Visual Regression Lab

Use this skill to capture comparable browser evidence and inspect visual regressions. Scope routes, viewports and states to the requested comparison; a capture alone does not prove quality.

Required Checks

1. Start or identify the local server.
2. Open the relevant route in a browser or Playwright.
3. Capture the relevant supported viewports and states, using the same conditions for baseline and candidate.
4. Check console errors, failed requests, layout overflow, missing media, and text clipping.
5. Inspect first viewport composition, hierarchy, spacing, typography, image crops, and CTA visibility.
6. Test key interactions: nav, forms, menus, tabs, accordions, hover states, focus states, and modals.
7. For animation-heavy work, test reduced-motion behavior and watch for jank.
8. For canvas/WebGL/video, verify nonblank rendered pixels and stable framing.
9. Report observed differences with evidence. Repair only when authorized, then recapture affected states; no fixed polish-pass or edit quota applies.

Read baseline contract. When a project has Playwright installed, run node "<skill-root>/scripts/capture-settled-baseline.mjs" <url> <output-dir> to capture settled desktop/mobile/reduced-motion states and a machine-readable inspection manifest.

When native full-page capture is blank, sparse, incomplete, or inconsistent with real scrolling, read stitched full-page capture and run node "<skill-root>/scripts/capture-stitched-full-page.mjs" --url <url> --output <image>. Treat the stitched image as evidence only after visual inspection.

Output

Report:

• routes/viewports checked
• screenshots or saved paths when available
• issues found and fixed
• commands run
• anything not tested

Resolve <skill-root> to this skill folder's actual absolute location; preserve project-local module resolution for Node capture tools.