How is BrowserSkill different from a headless browser tool?

Most agent browser tools open a blank, fresh browser with no session. BrowserSkill instead opens a separate Agent Window inside the Chrome or Edge you are already signed in to, so logins, cookies and extensions are simply there. Your own tabs keep working while the agent drives its window, and a human can step in for CAPTCHAs and confirmations.

Which AI agents can use BrowserSkill?

BrowserSkill works with any shell-capable agent, because the agent just calls the `bsk` command line tool. The README names Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, Hermes Agent and DeepSeek Harness. Running `bsk install-skill` wires the skill into a supported agent so it knows the available browser commands.

What is new in BrowserSkill 0.3.0?

Version 0.3.0 teaches BrowserSkill to find canvas elements and hand back typed visual references with visible-region screenshots, which used to be a blind spot. It also adds full-page screenshots to both the CLI and extension, explicit focus and blur plus scroll-to-element and a real mouse-wheel primitive, authenticated remote gateways with background task tabs, local task history and Korean localization.

Does the agent get direct access to my browser?

No. BrowserSkill splits into three parts: the `bsk` CLI the agent calls, a local daemon that handles inter-process messaging, and the browser extension that does the automation. The agent never speaks to the browser itself — requests go CLI to daemon to extension over a WebSocket on localhost, which keeps the automation surface local and inspectable.

Which browsers does BrowserSkill support?

BrowserSkill officially supports Chrome and Microsoft Edge, and other Chromium-based browsers are expected to work. Firefox support is planned but not shipped. On the operating-system side the project builds for macOS on Apple Silicon and Intel, Linux on x64 and ARM64, and Windows x64.

Tencent · 2026-09-17 · major

BrowserSkill 0.3.0 — Tencent's agent bridge gets canvas and remote gateways

BrowserSkill lets an AI agent drive a separate Agent Window inside your already logged-in Chrome or Edge. Version 0.3.0 adds canvas visual references, full-page screenshots, authenticated remote gateways and Korean localization.

GitHub repository card for Tencent BrowserSkill, a bridge that lets AI agents use your logged-in browser

Your agent borrows a window in the browser you are already logged into, then hands it back.

Quick facts

Maker	Tencent
Version	0.3.0 — CLI, extension and DSH plugin
License	MIT
Language	TypeScript
Stars	3,772 at release
Runs on	macOS, Linux, Windows x64; Chrome and Edge
What's new	Canvas refs, full-page screenshots, remote gateways, Korean UI

What is it?

BrowserSkill 0.3.0 adds canvas handling to Tencent's bridge between coding agents and a real browser. The tool opens a separate Agent Window inside the Chrome or Edge you already use, so the agent inherits your login state instead of starting from a blank profile, and your own tabs stay untouched while it works. The 0.3.0 tags line the CLI, the browser extension and the DSH plugin up on one version.

How does it work?

Three pieces do the work: the `bsk` command line tool that an agent calls from its shell, a local daemon that routes messages, and the browser extension that actually clicks. Requests travel CLI to daemon to extension over a WebSocket bound to localhost, so the agent never touches the browser directly. New in this release, the extension can discover canvas candidates and return typed visual references with visible-region screenshots, plus full-page screenshots, focus and blur, scroll-to-element and a real mouse-wheel primitive.

Why does it matter?

Logged-in automation is where most agent browser tools fall over — the agent gets a fresh session, hits a sign-in wall, and stops. Reusing your live session skips that, with a human still available for CAPTCHAs. Canvas support matters more than it sounds: design tools, maps and charting apps render into a canvas, which used to be invisible to an agent working from the DOM. BrowserSkill is MIT-licensed and gained more than 1,300 stars in a day.

Who is it for?

developers wiring agents into real web workflows

Frequently asked questions

How is BrowserSkill different from a headless browser tool?: Most agent browser tools open a blank, fresh browser with no session. BrowserSkill instead opens a separate Agent Window inside the Chrome or Edge you are already signed in to, so logins, cookies and extensions are simply there. Your own tabs keep working while the agent drives its window, and a human can step in for CAPTCHAs and confirmations.
Which AI agents can use BrowserSkill?: BrowserSkill works with any shell-capable agent, because the agent just calls the `bsk` command line tool. The README names Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, Hermes Agent and DeepSeek Harness. Running `bsk install-skill` wires the skill into a supported agent so it knows the available browser commands.
What is new in BrowserSkill 0.3.0?: Version 0.3.0 teaches BrowserSkill to find canvas elements and hand back typed visual references with visible-region screenshots, which used to be a blind spot. It also adds full-page screenshots to both the CLI and extension, explicit focus and blur plus scroll-to-element and a real mouse-wheel primitive, authenticated remote gateways with background task tabs, local task history and Korean localization.
Does the agent get direct access to my browser?: No. BrowserSkill splits into three parts: the `bsk` CLI the agent calls, a local daemon that handles inter-process messaging, and the browser extension that does the automation. The agent never speaks to the browser itself — requests go CLI to daemon to extension over a WebSocket on localhost, which keeps the automation surface local and inspectable.
Which browsers does BrowserSkill support?: BrowserSkill officially supports Chrome and Microsoft Edge, and other Chromium-based browsers are expected to work. Firefox support is planned but not shipped. On the operating-system side the project builds for macOS on Apple Silicon and Intel, Linux on x64 and ARM64, and Windows x64.

Try it

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