buildcrew-qa — Quality Modes
Pick the sub-mode from the user's intent. Full role prompts live in
.claude/agents/*.md; read the role file before executing. With subagent support,
dispatch it as a subagent; without, adopt the role in-context.
Sub-modes
| Intent | Role file | Loop |
|---|---|---|
| "browser test", "UI test", live-site QA | browser-qa.md |
if score < 70 → developer.md fixes → re-run browser-qa (max 2) |
| "qa audit", "audit my code", diff inspection | qa-auditor.md |
1 run (3 parallel lenses where supported) |
| "health check", "quality score" | health-checker.md |
1 run, report only |
Rules
- Read
.claude/harness/project.mdandrules.mdfirst if present. - browser-qa requires Playwright MCP; if unavailable, stop and instruct:
claude mcp add playwright -- npx @anthropic-ai/mcp-server-playwright. - Fixes are made by the
developerrole, never inline by the QA role. - End with a scored report and the list of issues found/fixed/remaining.