agent-browser

Operate an agent-browser CLI for auditable navigation, extraction, screenshots, and UI checks.

SKILL.md

GitHub

Agent Browser

Purpose

Operate a headless browser CLI (agent-browser) for AI agents to interact with websites. The skill enables navigation, clicking, typing, screenshot capture, and structured data extraction with full auditability.

When to Use

  • The user wants to interact with a website programmatically
  • Screenshot or visual verification is needed
  • Form submission or data extraction is required
  • Accessibility tree inspection is needed

When NOT to Use

  • The website blocks automated access (check robots.txt first)
  • The action requires human interaction for CAPTCHAs
  • The task is read-only but the website requires login

Capabilities

  • Navigate to URLs with configurable viewport and headers
  • Click, type, scroll, and hover interactions
  • Screenshot capture (full page or viewport)
  • Accessibility tree extraction for LLM-readable content
  • DOM extraction with CSS selectors
  • Wait for elements, network idle, or custom conditions
  • Proxy and authentication support

Inputs

  • url (required) — target URL
  • action (required) — one of: navigate, click, type, screenshot, extract, wait
  • selector (optional) — CSS selector for element targeting
  • value (optional) — text to type or other action parameter
  • timeout (optional, default 30s)

Workflow

  1. Validate — URL format, selector validity, required tools available
  2. Execute — Run the browser action via agent-browser CLI
  3. Capture — Store output (screenshot path, extracted text, error)
  4. Report — Present results to user

Tools

  • agent-browser CLI
  • Optional: screenshot storage, file system for outputs

Examples

User: "Take a screenshot of example.com" Response: Navigating to example.com... Screenshot saved to screenshots/example-2026-02-09.png

User: "Click the login button and extract the form fields" Response: Clicked #login-btn. Extracted form fields: username, password, remember_me

Safety

  • Never store credentials in logs
  • Respect robots.txt and terms of service
  • Confirm before submitting forms or making changes

Source

Auto-generated from openclawskills.net description.

Notes

Maintained as part of canonical web category.