playwright

name	playwright
description	Use when the user needs browser automation, web verification, scraping, screenshots, or end-to-end testing via Playwright. Triggers on phrases like "open this URL", "screenshot", "fill the form", "scrape", "browser test".
auto	false
tags	["browser","testing","automation"]

Playwright Browser Automation

Browser automation via the Playwright MCP server. Use this skill when the user asks to navigate, interact with, verify, or capture content from a website.

MCP configuration hint

If the Playwright MCP server is not yet configured, add it to ~/.llmcode/mcp.json:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

After saving, restart llmcode so the MCP server is loaded.

Core workflow

1. Navigate to the page (browser_navigate)
2. Take a snapshot to discover interactive elements (browser_snapshot)
3. Interact with elements by ref (browser_click, browser_type, browser_fill_form)
4. Re-snapshot after navigation or significant DOM changes
5. Capture results (browser_take_screenshot, browser_evaluate)

When to use

• Verifying a deployed page renders correctly
• Filling out forms and submitting
• Capturing screenshots for visual confirmation
• Scraping structured data behind JavaScript
• End-to-end testing of user flows

When NOT to use

• Simple HTTP fetches (use web_fetch instead)
• Pure HTML parsing (use a regex/parser, no browser needed)
• Static content extraction without JS rendering

Tips

• Always snapshot before clicking — refs are session-scoped and change after navigation.
• Prefer semantic locators (role + name) over CSS selectors when possible.
• For authentication-gated flows, save state once and reuse: faster + more reliable.
• Headless by default; pass --headed only for debugging.