Aller au contenu principal

Skill: qa-chrome

Fork

Visual tests and browser debugging via Chrome. Use to test web pages, verify visual rendering, debug with the console, or automate browser actions. Trigger when the user mentions "visual test", "Chrome", "browser", "browser console", "DOM", "screenshot", "GIF".

Configuration​

PropertyValue
Contextfork
Allowed toolsRead, Bash, Grep, Glob
Keywordschrome, claude in chrome

Detailed description​

Visual Tests and Chrome Debugging (pointer)

DevTools features (network inspection, profiling, accessibility tree, console replay) are canonical at the Chrome team's MCP server:

  • ChromeDevTools/chrome-devtools-mcp — github.com/ChromeDevTools/chrome-devtools-mcp (Apache-2.0, maintained by the Google Chrome DevTools team). Configure in .mcp.json to give Claude Code direct programmatic access during a session.

This skill scopes to the manual-review checklist when using Claude Code's --chrome flag — what to look for, not how the API works.

Prerequisites​

  • Claude Code launched with --chrome
  • "Claude in Chrome" extension v1.0.36+
  • Google Chrome open (not Brave/Arc/Firefox; no headless mode; WSL unsupported)

Manual-review checklist​

When asked to visually validate a page or flow:

  • Page loads with no console error
  • Layout correct (no overflow, no overlap, no z-index bug)
  • Text readable (contrast meets WCAG AA — see wcag-audit)
  • Images loaded (no broken sources, lazy-load completes)
  • Links functional (no 404 in network panel)
  • Forms submittable (validation triggers on bad input)
  • Responsive OK across mobile/tablet/desktop breakpoints — see qa-design (responsive category)
  • No network error (no 4xx/5xx unless intentional)

Expected output​

Structured report:

  • Screenshots/GIFs of issues
  • List of console + network errors with source location
  • Recommendations sorted by severity
  • Overall verdict: OK / Warnings / Errors

See also​

  • chrome-devtools-mcp (vendor MCP server, programmatic layer) — docs/recipes/recommended-vendor-skills.md
  • qa-design skill — UI/UX heuristics + responsive/breakpoint testing complement
  • wcag-audit — accessibility gate when contrast/readability issues surface
  • Audit pilot trace: specs/marketplace-audit/qa-skills-pilot-2026-05-06.md

Automatic triggering​

This skill is automatically activated when:

  • The matching keywords are detected in the conversation
  • The task context matches the skill's domain

Triggering examples​

  • "I want to chrome..."
  • "I want to claude in chrome..."

Context fork​

Fork means the skill runs in an isolated context:

  • Does not pollute the main conversation
  • Results are returned cleanly
  • Ideal for autonomous tasks

See also​