Automate Browser Interactions with Playwright

Skill covering reliable browser automation with Playwright (the recommended default: 96% success rate, roughly 4.5s average execution, cross-browser, Microsoft-backed) and Puppeteer (Chrome-only, 75% success rate at scale, but the stronger stealth-plugin ecosystem), plus a nod to Selenium for legacy systems needing specific language bindings. It draws a hard line between testing automation (predictable apps you control) and scraping or agent automation (unpredictable sites that actively fight back) as fundamentally different problems needing different tactics. Core patterns: user-facing locators in strict priority order (getByRole, then getByText, getByLabel, getByTestId, and CSS/XPath only as a last resort) instead of brittle CSS classes or XPath; Playwright's built-in auto-wait, never adding waitForTimeout or manual sleeps and instead waiting on specific conditions like waitForResponse or waitForURL; full test isolation via fresh browser contexts, with a shared-authentication pattern that saves login state once via storageState and reuses it across tests; error recovery via automatic screenshot-on-failure, retained traces and video on failure, and exponential-backoff retries; parallel execution across isolated browser contexts with concurrency limits via p-limit; and network interception to block images, fonts, and analytics for faster scraping or to mock API responses for deterministic tests. For anti-bot-protected sites it documents stealth techniques - puppeteer-extra-plugin-stealth, playwright-extra, undetected-chromedriver, hiding navigator.webdriver, randomized delays and human-like mouse movement, proxy and user-agent rotation - and managed cloud-browser services (Browserbase, BrowserStack) as an alternative to self-maintained stealth.

The skill also catalogs recurring failure modes: waitForTimeout causing flaky, slow tests; CSS selectors breaking on redesigns; navigator.webdriver triggering instant blocks; shared test state causing order-dependent failures; missing CI traces making failures undebuggable; headless-versus-headed rendering differences; rate-limiting after bursts of requests; unhandled popups and new windows; and iframe content requiring an explicit frameLocator rather than the page's default locator scope.

Use for building or hardening E2E tests, web scrapers, or browser-driving AI agents where reliability, not just "it worked once," matters. Choose Playwright by default in 2025 unless Puppeteer's stealth ecosystem or a Chrome-only constraint applies. This skill is not for API testing, load testing, accessibility testing, or visual regression testing - it explicitly delegates those to backend, performance-thinker, accessibility-specialist, and ui-design skills respectively, and hands off desktop automation beyond the browser to computer-use-agents.

Input is a target URL plus a set of user-facing selectors (role, text, label) or, for scraping, an unpredictable third-party page. Output is either a pass/fail test result with attached screenshot, trace, and video artifacts on failure, or extracted structured data (with a success flag and captured debug info on error) for scraping tasks.

// By role - THE BEST CHOICE
await page.getByRole('button', { name: 'Submit' }).click();
await page.getByRole('link', { name: 'Sign up' }).click();
await page.getByRole('heading', { name: 'Dashboard' }).isVisible();
await page.getByRole('textbox', { name: 'Search' }).fill('query');

Playwright and Puppeteer as the core frameworks, with puppeteer-extra-plugin-stealth, playwright-extra, and undetected-chromedriver for anti-detection; p-limit for concurrency-limited parallel scraping; and Browserbase or BrowserStack as managed cloud-browser alternatives to running stealth infrastructure yourself.

QA engineers writing E2E test suites, data engineers building scrapers against sites with anti-bot protection, and developers giving an AI agent browser-control tools - anyone who needs the difference between a flaky script and a reliable automation system spelled out as concrete patterns and fixes.