webapp-testing

# Chat Buddy Browser Verification Use a browser tool actually available in the current agent, or the project's installed Playwright suite. The former Antigravity-only `browser_subagent` API is not a prerequisite and must not be assumed available. Follow each selected tool's current documentation. Keep the existing helpers and license notices. ## Establish the tested mode Read `package.json`, `playwright.config.js` and the relevant test. Record whether this is the local compatibility app, mocked cloud UI, or a real local/hosted backend. `VITE_USE_CLOUD=false` in the unit-test configuration does not exercise a live server. Cloud-only behavior must be validated through the actual cloud bridge and server consequence, not an optimistic browser render. ## Focused workflow 1. Reuse the task's running server/browser when appropriate; otherwise start the existing dev/test command and verify its actual URL. Do not guess port/state from historical screenshots or kill unrelated servers. 2. Inspect the rendered page and discover current elements before interacting. Wait for the relevant UI condition; chat streaming or long-lived connections may never reach network-idle. 3. Exercise the changed user action and the denied/error/empty state it affects. For persistence/sync, reload or reconnect and verify the actual saved result. 4. Capture visible evidence for a visual claim, plus relevant console/network failures. A screenshot cannot establish server authorization or hidden-data isolation; use the product skill's server tests for those boundaries. From the repository root, `npm run test:e2e -- e2e/cloud-window.smoke.spec.js` selects a relevant existing suite after its prerequisites are satisfied. Inspect `e2e/p0-acceptance.spec.js` and `e2e/app.spec.js` for other supported flows. Unit mocks, browser fixtures, and live accounts have separate evidence levels. Use isolated fixtures for tests; do not send real social messages or spend model quota merely to verify UI or skill changes. If lifecycle automation is useful, inspect the bundled [scripts/with_server.py](scripts/with_server.py) or read its `--help`; resolve it from this skill directory, not a guessed root `scripts/` path. Only start/stop servers owned by the test. Do not install another browser package when the project's existing tooling suffices. Save artifacts to the task/repository's established directory and report tested URL, mode, action, result and remaining verification gap. Do not claim automatic video recording or successful navigation without observing that output.