agent-browser

What This Skill Does

The Agentic Browser skill provides a robust framework for autonomous browser automation. Leveraging Playwright for browser control and an intuitive '@e' reference system, it allows AI agents to observe, interpret, and manipulate web interfaces. By transforming complex DOM structures into a simplified set of actionable references, the agent can programmatically navigate pages, fill forms, click buttons, and extract data, effectively turning any website into an API-like interface for your AI.

Installation

To integrate this skill into your environment, use the OpenClaw hub CLI. Run the following command in your terminal:

clawhub install openclaw/skills/skills/okaris/agentic-browser

Ensure you have the necessary environment dependencies installed as specified in the repository documentation to support headless browser operations.

Use Cases

This skill is highly versatile for automation-heavy workflows, including:

• Automated Web Research: Agents can navigate news sites or databases, extract relevant textual information, and summarize findings in real-time.
• Complex Form Automation: Simplify multi-step data entry processes by instructing an agent to fill out lengthy registration or survey forms.
• Regression Testing: Automate repetitive UI testing flows to ensure consistent application behavior across deployments.
• Dynamic Web Scraping: Bypass static scraping limitations by interacting with sites that require authentication, JavaScript execution, or multi-step navigation.
• Media Archival: Automatically record sessions or take specific screenshots for visual verification of web states.

Example Prompts

1. "Navigate to the latest news page on techcrunch.com, identify the top three headlines using the @e refs, and summarize the content of the first article."
2. "Go to the login page of our internal dashboard, fill in the credentials using the provided environment variables, click the 'Sign In' button, and confirm if the redirect was successful by taking a screenshot."
3. "Open the demo website, drag the slider element to the right until the value reaches 80, and then extract the final confirmation message text."

Tips & Limitations

• Session Management: Always ensure you call the close function to free up resources and retrieve recording data. Leaving sessions open consumes unnecessary memory.
• Snapshot Strategy: The '@e' refs are dynamic. Whenever you perform an action that triggers a page navigation or dynamic content load, invoke the snapshot function to refresh the element mapping.
• Complexity: While the skill is powerful, it is optimized for standard web interactions. Extremely complex shadow DOMs or legacy frameworks might occasionally require custom JavaScript execution via the execute function rather than standard interact calls.
• Safety: Always run automation in restricted or non-production environments if you are testing scripts that perform 'write' actions, such as form submissions or data deletions.