Browser Control

Use this skill when the task requires controlling a real browser session, interacting with page elements, or extracting in-page data from live websites across macOS, Linux, or Windows.

Tools

• •

  browser_open

  • •Open a URL in Chromium/Chrome/Edge/Firefox/WebKit.
  • •Returns page snapshot fields (title, url, readyState, short text preview).
• •

  browser_click

  • •Click an element by CSS selector on the active tab.
  • •Use after opening a page or when navigating interactive UI.
• •

  browser_type

  • •Type into a field selected by CSS selector.
  • •Supports optional clear + optional submit.
• •

  browser_eval

  • •Execute custom JavaScript in page context.
  • •Use mode="expression" for simple reads (for example document.title).
  • •Use mode="function" for statement blocks and return a structured object.
• •

  browser_extract

  • •Extract page data in one of: text, html, links, forms.
  • •Optional selector scopes extraction to a specific section.
• •

  browser_close

  • •Close one session or all sessions created by this skill.

Operating Pattern

1. •Start with browser_open for target URL.
2. •Use browser_extract mode="text" or mode="links" to map the page quickly.
3. •Use browser_click and browser_type for navigation and form flow.
4. •Use browser_eval for custom DOM reads/actions that the standard tools do not cover.
5. •Re-run browser_extract (or browser_eval) to verify resulting page state.

Constraints

• •Requires Playwright runtime (playwright package) and browser engines.
• •Install once per environment:
  1. •npm install playwright
  2. •npx playwright install chromium firefox webkit
• •Default behavior is headed (visible browser window). Toggle headless from the dashboard skill config when needed.
• •For Linux desktopless environments, use headless=true or run with a virtual display.
• •browser_eval runs arbitrary page JavaScript; use only task-relevant scripts and avoid unsafe actions.