Web application testing

Use this execution guidance for the browser scenario selected by verify-change or the active frontend workflow. Read the project's verification guide and frontend Playwright config for commands, installation, gate selection and service setup; do not maintain a second runner or acceptance checklist here.

Drive the browser with the playwright-cli skill or the Playwright MCP; this skill owns the policy below. Use the existing TypeScript Playwright suite; do not add a parallel Python product suite. For standalone, explicitly requested automation outside that suite, scripts/with_server.py starts a server around a script; run it with --help before adapting it. It is not the project's E2E gate.

Inspect rendered state before acting: navigate, wait for a meaningful locator or response, inspect DOM/screenshot, then use role/label selectors. Assert user-visible result and recovery; inspect relevant console/network errors. Use keyboard, focus and a narrow viewport for affected UI. Do not wait for networkidle universally (polling/SSE can keep a connection open) or use arbitrary sleeps instead of locator assertions.

Use the isolated service lifecycle and E2E_PORT/E2E_RUN_ID from the verification guide. Never reuse an unidentified running server. Declare mocked network explicitly; mocks and screenshots do not prove real API compatibility. Return observations and limits to verify-change and the existing acceptance record.