BrowserSkill is an open-source browser automation and agent skill harness developed by Tencent. It provides a structured framework for recording, orchestrating, and executing browser-based actions as modular AI skills. The project bridges web interaction capabilities with agent workflows by capturing detailed browser telemetry, DOM operations, and user actions across complex multi-tab environments. The framework operates across a monorepo setup consisting of browser extension components, native Rust crates for performant trace management, and published npm packages. In daily workflows, developers and agent harnesses interact with BrowserSkill through browser extension overlays, sidebar interfaces, and recorded skill definitions (such as Trace v3 formats), allowing agents to navigate, manipulate, and observe real browser sessions with support for out-of-process iframes (OOPIF) and tab switching. What distinguishes BrowserSkill from traditional headless browser scripts or standard CDP runners is its focus on reusable skill recording and interactive extension-based observation. Rather than writing brittle end-to-end selector scripts from scratch, agents leverage recorded interaction traces, session-scoped observation sidebars, and Picture-in-Picture (PiP) status windows to execute complex web navigation tasks. BrowserSkill is distributed as an open-source project on GitHub under Tencent's repositories. It is actively maintained with frequent updates to its extension packages and Rust-based trace recording engine.
Tags: agent harness, ai agent, cli, developer tools, OpenSource