BrowserSkill is a browser automation tool that lets AI agents control your real, logged-in browser through a CLI and browser extension without interrupting your work.
The tool solves the problem of AI agents needing to interact with web applications while preserving your existing login sessions and browser state. It works by running agent tasks in a separate, visible Agent Window that the agent controls through the bsk CLI command, which communicates with a browser extension. When an agent needs to use a tab you already have open, it must explicitly borrow that tab and return it when finished, leaving the rest of your browser untouched. The extension and CLI daemon form the runtime that enables this separation.
BrowserSkill suits developers and users who want AI agents to automate web tasks on sites where they maintain active login sessions, eliminating the need for separate test accounts. It works with any shell-capable AI agent including Cursor, Claude Code, Codex, and others, with no lock-in to a specific model or framework. The tool includes built-in human-in-the-loop capabilities: when a task encounters a captcha, login requirement, confirmation dialog, or other human-only step, the agent can pause and ask you to take over before continuing. It supports macOS, Linux, and Windows, with Chrome and Microsoft Edge as primary targets, and Firefox planned for future support.
The project shows active development with regular updates to both the CLI and extension components. Work spans multiple runtime environments and operating systems, indicating sustained effort to maintain cross-platform compatibility. The team has invested in documentation for specialized setups, including guidance for sandboxed agent environments that require persistent daemon management. Feature additions continue to expand agent capabilities, such as full-page screenshot support with cancellation and export options.
