QA — Smart Browser Testing

Automates what a developer does manually: navigate, interact, trigger API calls, verify results. Uses Playwright MCP for interactions and agent-browser for video recording and visual diffs.

Input

Accept either:

• Natural language flow: "Create a recipe with 3 ingredients, verify nutrition calculates"
• URL + instructions: http://localhost:3000/orders + "Click New Order, fill supplier, submit"

If no URL provided, default to http://localhost:3000.

Step 1: Prepare

mkdir -p .qa
grep -qxF '.qa/' .gitignore 2>/dev/null || echo '.qa/' >> .gitignore

Check if dev server is running: curl -s -o /dev/null -w "%{http_code}" <url>. If not reachable, ask user to start it.

Step 2: Auth Check

1. Navigate to URL via Playwright MCP browser_navigate
2. browser_snapshot the page
3. Look for login form indicators (input[type=password], /login in URL, auth-related text)
4. If not logged in: Tell user to log in manually in their browser, then say "done". Re-check after.
5. If logged in: Proceed to test execution.

Step 3: Execute Test Flow

Parse the user's flow into numbered steps. Launch a single subagent that uses smart tool selection:

Tool Selection Rules

Task	Tool	Command
Start video	agent-browser	agent-browser open <url> then agent-browser record start .qa/recording.webm
Navigate	Playwright MCP	browser_navigate
Get element refs	Playwright MCP	browser_snapshot (re-snapshot before EVERY interaction)
Click / Fill / Select	Playwright MCP	browser_click, browser_fill_form, browser_select_option
Wait for response	Playwright MCP	browser_wait_for
Check API calls	Playwright MCP	browser_network_requests
Check console errors	Playwright MCP	browser_console_messages with level "error"
Take screenshot	Playwright MCP	browser_take_screenshot with filename .qa/<NN>-<step>.png
Visual before/after diff	agent-browser	agent-browser diff screenshot --baseline .qa/<before>.png
Stop video	agent-browser	agent-browser record stop

Subagent Prompt Template

Execute the following QA test flow at {URL}. Use Playwright MCP tools for all interactions and agent-browser for video recording.

Setup:

1. agent-browser open {URL} — connect to the page
2. agent-browser record start .qa/recording.webm — start video

Test steps: {numbered steps from user's flow description}

For EACH step:

• browser_snapshot first to get fresh element refs
• Execute the interaction via Playwright MCP
• browser_take_screenshot with filename .qa/<NN>-<step-name>.png
• browser_network_requests after any action that triggers an API call
• browser_console_messages to check for new errors
• Report: PASS or FAIL with what was observed

Teardown:

1. agent-browser record stop — save video
2. browser_close — close Playwright browser
3. agent-browser close — close agent-browser

Report format per step: Step N: Result: PASS/FAIL Observed: API: <method endpoint -> status> (if applicable) Screenshot: .qa/-.png Errors: <console errors or "none">

Step 4: Report

After the subagent completes, output the combined report:

======================================
QA: <flow description>
======================================

Step 1: <description>
  PASS — <observation>

Step 2: <description>
  PASS — <observation>
  API: POST /api/endpoint -> 201

Step 3: <description>
  FAIL — Expected X, got Y
  Screenshot: .qa/03-step-name.png

Console errors: <list or "none">
Video: .qa/recording.webm
Screenshots: .qa/ (<N> files)

======================================
VERDICT: <ALL PASS | PARTIAL | FAIL> (N/M steps)
======================================

On failure: include exact expected vs actual, reference the screenshot, and note the video has the full recording for debugging.

Auto-trigger Usage

After completing code changes that affect UI, ask the user:

"What flow should I QA? (e.g., 'test the order creation at localhost:3000/orders')"

Then run the full QA flow above.