playwright-browser（Playwright 浏览器）

Playwright Browser Automation

Headless browser automation using Playwright CLI and MCP tools for E2E testing, UI verification, screenshots, and web scraping.

When to Apply

Positive Triggers

• E2E test execution (test:e2e, playwright test)
• UI screenshot capture for review or regression
• Web scraping / content extraction
• Form submission testing
• Visual regression testing
• Lighthouse / performance auditing via browser
• Multi-page workflow validation

Negative Triggers (Do NOT Apply)

• Unit tests (use Vitest/Pytest)
• API-only tests (use httpx/fetch)
• Personal browser sessions (use claude-browser skill)
• Static code analysis

Core Principles

1. Headless by default — use --headed only when debugging
2. Screenshot at each step — capture visual evidence for review
3. Parallel sessions — leverage Playwright's built-in parallelism
4. Persistent profiles — reuse auth state across test runs
5. CI-safe — all operations must work in headless CI environments

MCP Playwright Tools

These tools are available via the mcp__playwright__* namespace:

Usage Pattern

1. Load tools: ToolSearch "playwright"
2. Navigate: browser_navigate to target URL
3. Interact: browser_click, browser_fill_form, browser_type
4. Capture: browser_take_screenshot at each significant step
5. Verify: browser_snapshot for accessibility tree, browser_evaluate for assertions
6. Clean up: browser_close when done

CLI Playwright Commands

# Install browsers
npx playwright install chromium

# Run all E2E tests
npx playwright test

# Run specific test file
npx playwright test tests/e2e/login.spec.ts

# Run with headed browser (debugging)
npx playwright test --headed

# Generate test from user interaction
npx playwright codegen http://localhost:3000

# Show HTML report
npx playwright show-report

# Run with specific project/browser
npx playwright test --project=chromium

Screenshot Workflow

Naming Convention

{step-number}-{description}.png

Examples: 01-login-page.png, 02-form-filled.png, 03-dashboard-loaded.png

Storage

ai-review/screenshots/     # Automated review screenshots
test-results/              # Playwright test artifacts
playwright-report/         # HTML report with traces

Capture Strategy

• Before action: Capture initial state
• After action: Capture result state
• On failure: Automatic screenshot (Playwright default)
• Full page: Use fullPage: true for long pages

Session Management

Auth State Reuse

// Save auth state after login
await page.context().storageState({ path: 'auth-state.json' });

// Reuse in subsequent tests
const context = await browser.newContext({
  storageState: 'auth-state.json'
});

Persistent Contexts

For multi-step workflows that span multiple tool calls, maintain browser context between operations. Close explicitly when done.

Parallel Execution

Playwright supports parallel test execution natively:

# Run tests in parallel (default)
npx playwright test --workers=4

# Run serially (debugging)
npx playwright test --workers=1

For MCP tool usage, open multiple browser tabs for parallel story validation.

Project Integration

Anti-Patterns

• Running headed browsers in CI — always use headless
• Hardcoding selectors — prefer data-testid or accessible roles
• Ignoring timeouts — set explicit timeouts per action
• Skipping screenshots — always capture evidence for review
• Using page.waitForTimeout() — prefer page.waitForSelector() or browser_wait_for

Pre-Merge Checklist

• All E2E tests pass in headless mode
• Screenshots captured for new UI flows
• Auth state properly managed (no credential leaks)
• Network requests validated (no unexpected external calls)
• Console errors checked (no uncaught exceptions)
• Viewport tested at standard breakpoints (mobile, tablet, desktop)