mcarcaso/playwright

Playwright Browser Skill

Primary browser automation now runs through Playwright MCP via the generic MCP client.

Use this skill when you need to:

• Read JavaScript-rendered pages
• Take screenshots
• Interact with pages/forms
• Inspect console/network traffic
• Extract content via snapshots or evaluation

Discover Tools

node src/skills/builtin/mcp-client/mcp-client.mjs list src/skills/builtin/playwright/SKILL.md

Inspect a tool:

node src/skills/builtin/mcp-client/mcp-client.mjs schema src/skills/builtin/playwright/SKILL.md browser_navigate

One-Shot Calls

Navigation returns page metadata and a snapshot:

node src/skills/builtin/mcp-client/mcp-client.mjs call src/skills/builtin/playwright/SKILL.md browser_navigate '{"url":"https://example.com"}'

Multi-Step Browser Sessions

Use batch when page state matters. It keeps one MCP browser connection alive for all steps.

node src/skills/builtin/mcp-client/mcp-client.mjs batch src/skills/builtin/playwright/SKILL.md '[
  {"tool":"browser_navigate","args":{"url":"https://example.com"}},
  {"tool":"browser_snapshot","args":{}},
  {"tool":"browser_take_screenshot","args":{"filename":"example.png","fullPage":true}}
]'

Screenshots/output files are written under user/drive/screenshots/ by the MCP server. Share with MEDIA:/absolute/path when needed.

Safety

• browser_run_code_unsafe is allowlisted because Mike wants full-power browser automation available when needed.
• Treat browser_run_code_unsafe as RCE-equivalent: use it deliberately, keep snippets tight, and do not use it for real-world actions unless Mike explicitly asks.
• Use browser_evaluate for page-context extraction when that is enough.
• For purchases, messages, bookings, form submissions to real people/businesses, or account changes: stop and confirm first.

Legacy Wrapper

The old local wrapper still exists at user/skills/playwright/index.js for fallback/debugging, but prefer MCP for new work.