qa-agent-browser

name	qa-agent-browser
description	Run deterministic browser QA using agent-browser. Produce QA report + trace/screenshot evidence.
disable-model-invocation	true
argument-hint	[baseUrl] [storiesPath|-] [mode=sweep|smoke|verify|fix]
allowed-tools	Read, Grep, Glob, Bash(agent-browser:*), Bash(mkdir *)

QA with agent-browser

Outputs

• reports/qa/QA_REPORT.md
• qa-artifacts/<date>/screenshots/
• qa-artifacts/<date>/traces/
• qa/bugs/*.md (if failures)
• qa/repros/*.sh (if failures)

Hard rules

1. Use the deterministic ref workflow:
   • agent-browser open <url>
   • agent-browser snapshot -i -C --json
   • Interact ONLY via refs (@e1, @e2, ...) from the latest snapshot.
   • Re-snapshot after any navigation or DOM change (refs become invalid).
2. Prefer --json output for anything you will assert/compare.
3. For every failure: collect screenshot, trace, console + page errors, and write a bug file.
4. Use one stable session: --session-name qa-${CLAUDE_SESSION_ID}

Modes

• sweep: Full test matrix from stories/specs.
• smoke: App loads, no console errors, primary workflow works.
• verify: Only rerun existing repro scripts in qa/repros/.
• fix: Reproduce bugs, identify root cause, implement fix, re-verify.

Steps

1. Confirm app is reachable at baseUrl.
2. Confirm agent-browser is installed.
3. Create artifact folders.
4. Build test matrix from stories (or do smoke if none provided).
5. Execute each scenario with trace capture.
6. For failures: screenshot + trace + bug file + repro script.
7. Write reports/qa/QA_REPORT.md with coverage summary and bug list.