BrowserSkill: Let AI Agents Drive Your Real Browser

TL;DR: Tencent/BrowserSkill is a TypeScript CLI plus browser extension that lets shell-capable AI agents operate your real, logged-in browser in the background. MIT licensed, 4,249 stars, the missing layer between headless Playwright and a human-only session.

What it is

Tencent/BrowserSkill ships a small companion CLI and a browser extension that together expose an authenticated browser to any AI agent that can run shell commands. The GitHub description reads: “Let AI agents use your real, logged-in browser without interrupting your work. CLI + extension for browser automation across any shell-capable AI agent.” The repository is written in TypeScript, has 4,249 stars, and is published under the MIT license. The homepage field is empty in the GitHub API response, so the project’s main surface is the repository itself.

What it does well

Browser automation is a solved problem for clean, headless flows — Playwright, Puppeteer, and the rest handle them fine. The hard problem is everything else: SSO portals, OAuth callbacks, MFA prompts, two-tab flows that share cookies, dashboards that fingerprint headless Chrome. BrowserSkill sits in that gap.

• Auth-aware by construction. Because the agent drives the same Chrome profile the human is logged into, every site that works in your normal browser keeps working. No replayed cookies, no proxy tricks.
• Non-interrupting. The CLI drives the browser in a backgrounded tab, so you can keep using your own tabs while the agent works. The extension is the bridge that routes commands to that tab.
• Agent-agnostic on the wire. Anything that can run a shell command can attach, which means Claude Code, Codex, Hermes, Aider, and ad-hoc scripts all speak to the same browser surface.
• Tencent-built, MIT licensed. The license field in the API is MIT, so commercial self-hosting is straightforward.

How to try it

The repo is the install surface; the README walks through the two pieces you need.

# 1. Install the extension in your Chromium-based browser
#    (load unpacked from the repo's extension/ directory)

# 2. Install the CLI side
npm install -g browserskill

# 3. From any agent that can run shell commands, drive your browser
browserskill open https://example.com
browserskill click "Sign in"

Once the extension is paired with the CLI, the agent treats the browser the same way you would — open, click, type, wait for navigation — except the keystrokes never touch the tab you are reading.

When it fits, and when it doesn’t

BrowserSkill is the right answer when an AI agent needs a real authenticated session and Playwright’s clean-room approach is the wrong shape — for example, scraping a behind-login dashboard, filling a form gated by SSO, or driving a SaaS UI that bans headless clients. It is the wrong answer when you only need deterministic scraping of public pages, where Playwright or a simple HTTP fetch is still cheaper. It also does not replace a real browser automation framework when you need pixel-stable visual diffs across CI runs — that is still Playwright’s job.

The 4,249 star count at 1,302 stars today shows the gap it fills is real, and the MIT license means you can ship the extension internally without a legal review.

Source and Accuracy Notes

• Project page: https://github.com/Tencent/BrowserSkill
• Source repository: Tencent/BrowserSkill (verified via GitHub API html_url)
• License: MIT
• Stars at time of writing: 4,249
• Language: TypeScript
• Topics: agent, browser-use, dsh-plugin
• Homepage: not set in GitHub API response