Vendored from anthropics/skills under Apache-2.0. Modified: description and when_to_use trigger rewrite, a Delegating section pointing at the paired webapp-tester agent shipped in this repo, a tone pass (emoji and all-caps emphasis removed), a wait-discipline pass replacing the fixed-duration wait_for_timeout advice and its two example uses with condition waits, a not-found-reads-boot bullet, and a Limits note that a screenshot gets read and judged, with a pixel diff as the secondary check.

Web Application Testing

To test local web applications, write native Python Playwright scripts.

Helper Scripts Available:

• scripts/with_server.py - Manages server lifecycle (supports multiple servers)

Always run scripts with --help first to see usage. Don't read the source until the script has been run and shown not to fit the task. These scripts can be large; reading them pollutes your context window. They exist to be called directly as black boxes.

Decision Tree: Choosing Your Approach

UI/frontend only. For pure backend/API assertions without a browser, use Playwright's request context or a plain HTTP client instead (see Limits).

User task → Is it static HTML?
    ├─ Yes → Read HTML file directly to identify selectors
    │         ├─ Success → Write Playwright script using selectors
    │         └─ Fails/Incomplete → Treat as dynamic (below)
    │
    └─ No (dynamic webapp) → Is the server already running?
        ├─ No → Run: python scripts/with_server.py --help
        │        Then use the helper + write simplified Playwright script
        │
        └─ Yes → Reconnaissance-then-action:
            1. Navigate and wait for networkidle
            2. Take screenshot or inspect DOM
            3. Identify selectors from rendered state
            4. Execute actions with discovered selectors

Example: Using with_server.py

To start a server, run --help first, then use the helper:

Single server:

python scripts/with_server.py --server "npm run dev" --port 5173 -- python your_automation.py

Multiple servers (e.g., backend + frontend):

python scripts/with_server.py \
  --server "cd backend && python server.py" --port 3000 \
  --server "cd frontend && npm run dev" --port 5173 \
  -- python your_automation.py

To create an automation script, include only Playwright logic (servers are managed automatically):

from playwright.sync_api import sync_playwright

with sync_playwright() as p:
    browser = p.chromium.launch(headless=True) # Always launch chromium in headless mode
    page = browser.new_page()
    page.goto('http://localhost:5173') # Server already running and ready
    page.wait_for_load_state('networkidle') # wait for JS to execute before inspecting
    # ... your automation logic
    browser.close()

Reconnaissance-Then-Action Pattern

1. Inspect rendered DOM:

   page.screenshot(path='/tmp/inspect.png', full_page=True)
   content = page.content()
   page.locator('button').all()

   Note: screenshot() only saves an image, it does not diff against a baseline (see Limits).
2. Identify selectors from inspection results
3. Execute actions using discovered selectors

Common Pitfall

Inspecting the DOM before networkidle on a dynamic app reads a half-rendered page. Wait for page.wait_for_load_state('networkidle') first.

Best Practices

• Use bundled scripts as black boxes - To accomplish a task, consider whether one of the scripts available in scripts/ can help. These scripts handle common, complex workflows reliably without cluttering the context window. Use --help to see usage, then invoke directly.
• Use sync_playwright() for synchronous scripts
• Always close the browser when done
• Use descriptive selectors: text=, role=, CSS selectors, or IDs
• Wait on the observable the next assertion reads, never on a duration. expect(locator).to_have_text(...) and its siblings retry until the value arrives; page.wait_for_selector(), page.wait_for_function() and page.expect_console_message(predicate) cover what expect does not. Make the condition the one your code is actually about: a bare expect_console_message() returns on the first message and misses everything logged after it, so pass the predicate that matches the message meaning the work finished. A page.wait_for_timeout() is a bet that the page settles inside a number someone picked, and it goes wrong exactly when the machine is loaded, which is when CI runs. Where an action genuinely produces nothing observable, keep the duration and write one line beside it saying which condition could not be expressed.
• Two timeout budgets exist and they are set separately. Measured against Chromium: expect() defaults to 5000 ms, page.wait_for_* defaults to 30000 ms, and expect.set_options() moves only the first. A per-call timeout= beats both. Do not pass timeout=0 to set_options expecting no wait: 0 is falsy, so the default comes back and every check passes for the wrong reason.
• not found read a fixed few seconds after an action on a dev-server SPA measures the app's boot, not the action's outcome -> poll for the app's own ready signal (an element, a postMessage, a network call) and only then record the absence.

Limits

• screenshot() only saves an image. Read it and judge it as a viewer would; that is the check for "looks right". A pixel diff (Pillow, pixelmatch) answers only "did anything move", and needs a baseline someone judged.
• UI/frontend only. For pure backend/API assertions without a browser, use Playwright's request context (playwright.request) or a plain HTTP client instead.

Reference Files

• examples/ - Examples showing common patterns:
  • element_discovery.py - Discovering buttons, links, and inputs on a page
  • static_html_automation.py - Using file:// URLs for local HTML
  • console_logging.py - Capturing console logs during automation

Delegating

To verify an app as a subagent task, spawn the webapp-tester agent (agents/webapp-tester.md in this repo). It boots through scripts/with_server.py, drives the flows, and returns pass/fail with screenshot + console evidence. Use this skill inline only when driving the browser yourself.