agent-browser

What This Skill Does

The agent-browser skill is a high-performance, headless browser automation tool designed specifically for AI agents. It prioritizes efficiency and deterministic behavior by utilizing accessibility tree snapshots, which allow the agent to "see" the page structure in a machine-readable format. By assigning unique reference IDs (refs) to interactive elements, the tool eliminates the flakiness often associated with standard CSS or XPath selectors in dynamic single-page applications (SPAs). It supports full session isolation, state persistence, and granular network control, making it an essential utility for complex, multi-step web automation workflows.

Installation

To install this skill, use the ClawHub CLI command within your OpenClaw environment:

clawhub install openclaw/skills/skills/nicoataiza/agent-browser-clawdbot-bak-2026-01-28t18-01-09-10-30

Once installed, you can trigger browser commands directly through your agent interface. Ensure your environment has the necessary dependencies for headless browser rendering.

Use Cases

• Automated Web Testing: Run regression tests for SPAs where UI components change dynamically.
• Data Extraction & Scraping: Efficiently navigate through paginated content using ref-based selectors to extract data points consistently.
• Workflow Automation: Automate complex form filling, dashboard monitoring, or multi-step account management tasks that require session persistence.
• Network Interception: Debugging frontend applications by mocking API responses or blocking unnecessary tracking scripts.

Example Prompts

1. "Open https://portal.example.com, log in using the saved credentials, navigate to the dashboard, and extract the latest status count from the element labeled 'active-tasks'."
2. "Go to the contact page, fill out the form using the provided details for the name and email fields, and wait for the success message to appear before closing the session."
3. "Monitor the pricing page for updates, blocking all advertisement requests to speed up load times, and notify me if the price element value drops below 50."

Tips & Limitations

• Snapshot Best Practices: Always use the -i --json flags with the snapshot command to ensure you receive the most relevant interactive elements and structured data for your agent.
• Session Management: Leverage the --session flag to maintain multiple independent browser states, which is vital for testing multi-user scenarios simultaneously.
• Performance: For large pages, use the -d (depth) flag in snapshots to limit the accessibility tree size and reduce context window token usage.
• Limitations: This skill is strictly for automation and interaction. If you require deep visual analysis or pixel-perfect screenshots for debugging layout issues, fall back to the built-in browser tool instead.