agent-browser

What This Skill Does

agent-browser is a powerful, headless browser automation CLI specifically engineered for OpenClaw AI agents. It provides a programmatic interface to interact with the modern web, effectively acting as the eyes and hands of an AI. By automating navigation, form interaction, data extraction, and state management, it allows agents to perform complex workflows—such as logging into secure platforms, scraping dynamic content, or testing web applications—without requiring human supervision. It uses a unique reference system (@e1, @e2) to track and interact with specific DOM elements, ensuring that agent actions are precise and reproducible.

Installation

The skill is hosted within the OpenClaw ecosystem. To get started, ensure you have Node.js installed on your host system. Execute the following command in your terminal to integrate it: clawhub install openclaw/skills/skills/bodietron/openclaw-agent-browser After installation, run scripts/setup.sh inside the skill directory to provision the necessary Chromium dependencies required for the browser engine.

Use Cases

This skill is indispensable for scenarios requiring browser-level access. Common use cases include:

• Automated Testing: Programmatically navigating through web app funnels to ensure UI components function correctly.
• Data Scraping: Extracting text, tables, or complex structures from websites that lack public APIs.
• Agentic Workflows: Allowing an AI to complete multi-step signups, respond to dynamic web content, or monitor websites for specific status changes.
• Resource Retrieval: Downloading files, generating PDFs from web pages, or capturing screenshots for documentation or auditing purposes.

Example Prompts

1. "Open the website https://example.com/login, fill in the credentials using my saved environmental variables for username and password, click the login button, and let me know when the dashboard is reached."
2. "Go to the company's product pricing page, scrape the current monthly subscription fees for all tiers, and save the results into a file named prices.txt."
3. "Take a full-page screenshot of the documentation page at https://docs.openclaw.ai so I can review the layout offline."

Tips & Limitations

• Ref Management: Always re-snapshot the page after any navigation or DOM-changing event. Refs become stale quickly if the page content updates dynamically via JavaScript.
• State Persistence: Utilize the state save and state load commands to preserve cookies and session data. This is critical for avoiding repeated MFA or login hurdles.
• Performance: Chaining commands with && is efficient for static tasks, but use individual command execution when you need to verify output at each step.
• Limitations: Since it is a headless environment, it may struggle with certain complex anti-bot measures like sophisticated Captchas that require human-in-the-loop interaction.