Tencent/BrowserSkill

What it solves

BrowserSkill allows AI agents to interact with a user's web browser using their existing logged-in accounts and sessions. It removes the need for the agent to handle complex authentication or manage separate browser profiles, allowing it to perform tasks like reading documents, filling forms, and navigating internal sites while the user remains active in the browser.

How it works

The system consists of three main components: a CLI/daemon that manages the connection, a browser extension (for Chrome or Edge) that provides the automation interface, and a "skill" (a set of instructions) that teaches an AI agent how to use the CLI tools. The agent sends commands via the CLI to the daemon, which communicates with the extension to control the browser. Tasks run in a dedicated "Agent Window" to keep them visible and separate from the user's main work, though the agent can explicitly "borrow" existing tabs if needed.

Who it’s for

• AI Agent Users: People using shell-capable agents (like Cursor, Claude Code, or DeepSeek Harness) who want their agents to perform real-world web tasks using their own accounts.
• Web Developers: Developers who need an AI agent to help debug websites by capturing network requests, console logs, and performance metrics.
• Security/QA Testers: Those who need to automate browser-based workflows and capture evidence of bugs.

Highlights

• Session Persistence: Uses the browser's current login state, eliminating the need for manual credential management.
• Website Debugging: Captures detailed evidence including HTTP requests, response bodies, and console messages for AI-driven debugging.
• Flexible Deployment: Supports local automation, remote server-side agents connecting to a local browser via WSS, and sandboxed environments.
• Visual Transparency: Tasks are performed in a visible Agent Window, and users can confirm when an agent borrows an existing tab.
• Comprehensive Capture: Supports viewport and full-page screenshots with automatic scrolling.
• Multi-Profile Support: Allows binding tasks to specific browser profiles or named instances.