Tencent has launched BrowserSkill, an open-source tool that bridges the gap between AI coding agents and real web browsing sessions. The tool topped GitHub Trending on September 17 with 1,350 new stars in a single day, drawing attention from the AI developer community for its pragmatic approach to browser automation.
BrowserSkill lets agents like Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, and others work with sites a user is already signed into -- no separate test accounts or authentication setup required. The tool opens a dedicated Agent Window where tasks run visibly, leaving the user's own browser tabs completely undisturbed.
How it works: The architecture sits between the agent harness and the browser. The agent issues commands through a bsk CLI, which communicates with a local daemon. The daemon routes requests to a Chrome or Edge extension that handles browser automation in the background. When an agent needs to interact with a tab, it explicitly borrows it -- and returns it when the task is complete.
Key features include: Reuse of real login state (agents work with active sessions), human-in-the-loop built in (agents can pause at CAPTCHAs and ask the user to intervene), sandbox-aware setup for environments where background processes are reaped after each command, and cross-platform support for macOS, Linux, and Windows.
The project crossed 1,350 stars within its first day on GitHub Trending. The MIT-licensed repository includes the full source across Cargo and pnpm workspaces, with a built-in evaluation suite for testing browser capabilities deterministically. BrowserSkill also ships a native plugin for DeepSeek Harness, providing browser tools directly in that framework.
