chrome-cdp-live-browser

What This Skill Does

The chrome-cdp-live-browser skill allows your OpenClaw AI agent to interface directly with your live Google Chrome session via the Chrome DevTools Protocol (CDP). Unlike standard automation tools that launch isolated, empty browser instances, this skill attaches to your existing open tabs. This means the AI retains access to your authenticated sessions, existing cookies, and your current page state. By utilizing the CDP, the agent can perform high-level interactions, extract structured data via accessibility trees, capture screenshots, and execute arbitrary JavaScript directly within the browser context. It is designed for seamless integration with your daily workflows, supporting multiple browser types including Chrome, Brave, Edge, and Vivaldi.

Installation

To integrate this skill into your environment, use the OpenClaw command-line interface. Run the following command in your terminal:

clawhub install openclaw/skills/skills/adisinghstudent/chrome-cdp-live-browser

Ensure that you have enabled Remote Debugging in your browser settings by navigating to chrome://inspect/#remote-debugging and checking the "Enable remote debugging" option. No browser restart is required for these changes to take effect. If you are using a non-standard installation path, ensure your CDP_PORT_FILE environment variable points to the correct DevToolsActivePort location.

Use Cases

This skill is perfect for scenarios requiring persistence and context. You can use it to manage your email inbox, summarize content from complex web applications where you are already logged in, or automate tasks on internal dashboards. It is particularly effective for performing research tasks that involve navigating multi-step authentication processes or interacting with dynamic, JavaScript-heavy web applications that often break standard headless scrapers. Developers can also use it to audit network performance or verify DOM structure across multiple open tabs.

Example Prompts

1. "OpenClaw, please list my open browser tabs and take a screenshot of the one currently displaying my GitHub pull requests so I can review it."
2. "Extract the main content from the article open in the first tab and summarize the key takeaways in a bulleted list."
3. "Navigate the active tab to the project management dashboard and run a script to find all elements with the class 'task-item' and count them."

Tips & Limitations

• Performance: The agent works best when it has a clear view of the accessibility tree; use the 'snap' command to help the AI understand the UI layout.
• Scope: Because this connects to a live session, the agent's actions are real; exercise caution when triggering buttons that perform destructive actions like 'delete' or 'send'.
• Concurrency: While the skill supports 100+ tabs, avoid triggering multiple high-frequency commands simultaneously to prevent potential protocol timeouts.
• OS Compatibility: The skill automatically detects browsers across macOS, Linux, and Windows, but ensure your system user has appropriate permissions to communicate with the browser's debug port.