Playwright (Automation + MCP + Scraper)

• JS-rendered pages — the content you need only exists after JavaScript executes (SPAs, dynamic tables, lazy-loaded data).
• Multi-step browser workflows — forms, login flows, file uploads/downloads, or navigation sequences that depend on real browser events and DOM state.
• Screenshots, PDFs, and visual evidence capture — you need pixel-accurate output from a rendered page, not raw HTML.
• Playwright test authoring and debugging — writing, running, or fixing a @playwright/test suite, tracing CI flakes, or reproducing UI bugs in headed mode.
• MCP-driven browser control — the agent already has browser_* tools in the loop and you want navigate-click-fill-screenshot workflows without writing new automation code.

• Static HTML or simple HTTP responses are sufficient — a plain fetch or HTTP skill is faster, cheaper, and less brittle when the page doesn't require JavaScript rendering.
• Pure API tasks — if the target service has a documented API, calling it directly is more reliable than driving a browser.
• Extraction is the only goal and a non-browser path works — prefer documented APIs or plain HTTP before spinning up a full browser just to scrape.

Run npx @playwright/mcp --headless to expose browser_navigate, browser_click, browser_type, browser_snapshot, and more as first-class tool actions. This path requires no new automation code — the agent drives the browser directly through MCP tool calls.

Write and run direct Playwright scripts (chromium.launch, page.goto, page.screenshot) or full @playwright/test suites with npx playwright test. Supports headed runs, trace capture, retries, and parallel workers out of the box.

Use npx playwright codegen <url> to record interactions and generate stable locators. Eliminates guessing selectors from source and produces role-, label-, and text-based locators that survive minor DOM changes.

Capture full-page screenshots, generate PDFs, record execution traces, and download files — all through either MCP tool actions or direct Playwright API calls. Artifacts stay local in the workspace or system temp directory.

The skill enforces role, label, text, and test-ID selectors over fragile CSS/XPath, and web-first assertions over raw command checks. Built-in guidance on actionability waits prevents flaky sleep-based timing hacks.

No hidden files, persistent sessions, or undeclared network calls. Auth state is temporary unless explicitly reused. High-stakes production flows require explicit user confirmation before proceeding.

Run and extend an existing @playwright/test suite against a local or staging app. Use npx playwright test --trace on to capture artifacts for CI failures, then diagnose flake with trace viewer and headed re-runs.

Let an agent navigate a site, fill out a multi-step form, and capture a screenshot — all via browser_navigate, browser_type, browser_click, and browser_snapshot MCP actions — without writing a single line of Playwright script.

Navigate to a JS-rendered page, wait for the DOM to settle, then use browser_evaluate or a direct Playwright script to extract structured data. Ideal when the target has no API and the content only appears after JavaScript runs.

Reproduce a reported UI regression in headed mode with npx playwright test --headed, capture a trace with --trace on, and inspect network, console, and DOM snapshots to isolate the root cause before writing a fix.