Enable AI agents to perform tasks in your logged-in browser
Imagine an AI agent clicking and typing directly within your usual Chrome or Edge window. BrowserSkill is a tool that combines a CLI and a browser extension to allow AI agents to control your actual browser session. Because the agent works in a separate 'agent window,' you can intervene only when needed without disrupting your own workflow.
How the agent handles your browser
BrowserSkill consists of a command-line interface (CLI) called bsk and a browser extension. If an AI agent can execute shell commands, it can use this tool to read browser text, click buttons, and fill out forms. The key point is that the agent uses your existing login state as is. This means tasks like searching internal sites or reading logged-in documents proceed without separate authentication. Work happens in a separate 'agent window,' though you can also lend your existing tabs to the agent. When the work is done, lent tabs return to their original location, and tabs created by the agent are cleaned up.
Debugging and data extraction beyond simple automation
This tool provides website debugging and data extraction features in addition to simple click automation. It can record network requests, console messages, and page changes that occur when the agent performs specific tasks. You can use these records to diagnose issues, such as finding the cause of form submission failures or identifying slow APIs. It also allows you to extract tables or lists displayed on a page into structured data in JSON or CSV format, which improves the accuracy of web scraping tasks. You can view this evidence in the extension or export it as a JSON file for later review.
Installation and initial setup
Installation is supported on macOS, Linux, and Windows, and it works in Chrome and Microsoft Edge (Chromium 125 or higher). First, install the bsk CLI, add the extension to your browser, and then run the bsk install-skill command to register the skill with your AI agent (e.g., Cursor, Claude Code, Codex). If you are using DeepSeek Harness, you only need to install the dedicated plugin. Once installation is complete, check the connection status with the bsk doctor command and test by instructing the agent to perform its first browser task. Remote connections that run the agent on a server while controlling a local computer's browser are also supported via an authenticated WSS connection.
Security and privacy considerations before use
BrowserSkill does not force cloud services or collect telemetry, but you must remember that the agent operates in your logged-in browser. Since the agent holds the same permissions as your logged-in websites, you should only assign tasks to trusted agents and in safe environments. The extension has two automation settings: confirmation before lending tabs and allowing human intervention requests, both of which are enabled by default. Website debugging logs and task videos are stored in the browser profile or daemon host and may contain sensitive information, so you must review the content before sharing. The license is MIT, and it is recommended to update the CLI, extension, and plugins together for version compatibility.
