BrowserSkill: AI Agents Borrow Your Logged-In Browser Without Test Accounts

Tencent's open-source BrowserSkill lets AI agents like Cursor and Claude Code operate within your authenticated Chrome or Edge window, inheriting login state, borrowing tabs via explicit leases, and handing off CAPTCHAs to humans, all through a local daemon and browser extension with no external servers.

Architecture Digest
Architecture Digest
Architecture Digest
BrowserSkill: AI Agents Borrow Your Logged-In Browser Without Test Accounts

Problem: AI Agents Lack Your Login Context

The author describes a common frustration: asking Cursor to extract a report from a company admin portal, only to have the agent stall at the login page because it has no cookies, no session, and cannot bypass CAPTCHAs or reach internal networks.

Existing Approaches and Their Drawbacks

Headless browsers (e.g., Playwright) – start a fresh session with no cookies; you must maintain test accounts and embed credentials in scripts.

Full browser handover – the agent takes over your active window, preventing you from working simultaneously.

BrowserSkill's Third Way

BrowserSkill opens a dedicated Agent Window inside your running Chrome or Edge. That window inherits your existing login state (cookies, localStorage, etc.), while your personal tabs remain untouched. The agent must explicitly "borrow" a tab before acting on it, and returns it when done.

Five Core Capabilities

Real login reuse – any site you've already logged into (internal tools, e-commerce, etc.) works immediately; no test accounts needed.

Tab leasing – agents request access to a specific tab via a protocol-enforced lease; other windows are unaffected.

Universal agent compatibility – any agent that can run shell commands installs the bsk CLI. Official one-click adapters exist for Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, and Hermes Agent; DeepSeek Harness uses a separate plugin.

Human-in-the-loop fallback – when the agent hits a CAPTCHA, 2FA prompt, or confirmation dialog, it pauses and asks you to intervene; you resolve it and the agent continues. The toggle for this behavior moved into the extension settings in v0.3.0.

Full-page screenshots & remote pairing – long screenshots support configurable timeouts (default 2 min, max 5 min) and restore scroll position. A built-in auth pairing lets a server-side agent connect to your local browser.

Execution Flow

The request chain is:

Agent → bsk CLI → local daemon (background process) → browser extension (WebSocket on 127.0.0.1) → Agent Window

All communication stays on the machine; no third-party server ever sees your cookies or traffic. The mangodeveloper review highlights that the browser is "borrowed, not hijacked."

Quick Start (Three Steps)

Install bsk via the provided script (macOS, Linux, Windows).

Add the BrowserSkill extension from the Chrome/Edge store (other Chromium browsers use the Chrome version; Firefox is marked "planned").

Run bsk install-skill to inject usage instructions into your agent, then verify with bsk doctor.

Known Gotchas (v0.3.0)

The --unattended flag and related headless parameters are deprecated; older scripts need updating.

CLI and extension versions must match, especially for the long-screenshot feature – upgrade both together.

The author has not personally run the install; steps and flags come from the official README – verify before use.

Who It's For / Not For

Good fit: Developers using Cursor or Claude Code on tasks that touch authenticated sites; teams needing agents to validate pages, fill forms, or scrape internal data; privacy-conscious users who don't want to export cookies.

Bad fit: Large-scale parallel scraping (design philosophy is "borrow, not raid"); Firefox users; complete beginners uncomfortable with a terminal.

Real-World Example

A Zhanzhang.com report cites a developer using WorkBuddy + BrowserSkill to have an AI search Zhihu for hot tech topics, summarize key viewpoints, and pull high-vote comments from a social platform to build a sentiment profile – all in minutes.

Author's Take

Turning "agent borrows browser" into an explicit lease with human fallback shows more respect for a developer's boundaries than headless approaches. 5.5k stars in three months signals strong community interest. Risk: v0.3.0 is still early; protocol and settings change frequently – pin versions and watch changelogs for production use.

GitHub:

https://github.com/Tencent/BrowserSkill
Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

CLIAI AgentsChrome extensionopen sourceTencentbrowser automationhuman-in-the-loopBrowserSkill
Architecture Digest
Written by

Architecture Digest

Focusing on Java backend development, covering application architecture from top-tier internet companies (high availability, high performance, high stability), big data, machine learning, Java architecture, and other popular fields.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.