BrowserSkill

// Overview

BrowserSkill is a local bridge between an AI coding agent and the browser you actually use. Most agent browser tooling hands the model a blank, freshly launched browser with no cookies and no session, which is why so many real tasks stop at a sign-in wall. BrowserSkill takes the other route: it opens a separate Agent Window inside your running Chrome or Microsoft Edge, so the agent inherits the login state that is already there, and your own tabs keep working while it does its job.

Three components make that work. The `bsk` command line tool is what the agent calls from its shell. A local daemon sits behind it handling inter-process messaging. A browser extension is the piece that actually automates the Agent Window. The agent never talks to the browser directly — a request travels from the CLI to the daemon and then to the extension over a WebSocket bound to localhost, which keeps the whole automation surface local and inspectable rather than routed through a remote service.

Because the interface is a plain command line tool, BrowserSkill is agent-agnostic: anything that can run a shell command can use it. The project names Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, Hermes Agent and DeepSeek Harness as supported agents, and `bsk install-skill` wires the skill into a supported agent so it knows which browser commands exist. Human-in-the-loop handling is built in for the moments automation cannot cover on its own, such as CAPTCHAs and confirmation dialogs.

BrowserSkill is MIT-licensed and written in TypeScript. It builds for macOS on Apple Silicon and Intel, Linux on x64 and ARM64, and Windows x64. Chrome and Microsoft Edge are the officially supported browsers, other Chromium-based browsers are expected to work, and Firefox support is planned rather than shipped. The CLI, the browser extension and the DSH plugin are versioned together — all three moved to 0.3.0 in September 2026.

// What it does

• Opens a separate Agent Window inside your running Chrome or Edge, reusing the session you are already signed in to
• Three-part architecture — `bsk` CLI, local daemon, browser extension — with the agent never touching the browser directly
• Local-only transport: CLI to daemon to extension over a WebSocket on localhost
• Works with any shell-capable agent, including Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, Hermes Agent and DeepSeek Harness
• Human-in-the-loop handoff for CAPTCHAs and confirmation dialogs
• Canvas support — discovers canvas candidates and returns typed visual references with visible-region screenshots
• Full-page screenshots from both the CLI and the extension
• Interaction primitives: explicit focus and blur, scroll-to-element, and a real mouse-wheel event
• Authenticated remote gateways and background task tabs
• Local task-execution history for auditing what the agent did
• Cross-platform builds for macOS, Linux and Windows x64, with Korean alongside the existing interface languages

// Getting started

Install the `bsk` binary, install the skill into your agent, then check the setup with the built-in doctor command. The browser extension pairs with the local daemon once it is running.

curl -fsSL https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.sh | sh

export PATH="${BSK_INSTALL_DIR:-$HOME/.local/bin}:$PATH"

irm https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.ps1 | iex

bsk install-skill

bsk --version

bsk doctor

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Letting a coding agent work inside a web app you are already authenticated to, without handing it your password
• Automating internal dashboards and admin panels that have no API and sit behind SSO
• Driving canvas-based tools — design apps, maps, charting surfaces — that a DOM-only agent cannot see
• Research and data-gathering tasks across sites where an anonymous session hits a paywall or rate limit
• Keeping a human in the loop for CAPTCHAs and confirmation steps while the agent handles the rest of a long browser workflow

// How BrowserSkill compares

BrowserSkill alongside other open-source computer & browser use tools AI/TLDR tracks, ranked by GitHub stars.