Tencent BrowserSkill: AI Agents Borrow Your Browser Tabs Instead of Launching New Instances

The Scenario: Reviewing a PR While an Agent Asks to "Click That Button for You"

Picture this: you're reviewing a pull request in your browser, already logged into GitHub, Jira, and Slack. Your AI agent (Claude Code, Cursor, or Codex) says: "I can merge this PR for you, but I need access to your browser."

The traditional options are poor:

• Headless browser: it has none of your login state, so it can do nothing.
• API keys: not every service offers an API.
• Playwright with a new browser instance: the agent must log in again—two-factor authentication, CAPTCHAs, device verification, all of it.

Tencent's BrowserSkill offers a fourth answer: let the agent "borrow" tabs you already have open, then return them when done, without affecting your other tabs.

It gained around 1,350 stars in a single day for good reason.

The "Borrow a Tab" Model: From Simulating Humans to Collaborating with Humans

For the past decade, browser automation has focused on "simulating humans." Selenium simulates clicks, Playwright simulates user behavior, Puppeteer simulates keyboard input. The implicit assumption: the browser is a tool, the agent is the operator, and the human is a bystander.

BrowserSkill flips this: the browser is the human's workspace; the agent is a visiting guest.

The mechanics:

1. Agent Window: BrowserSkill opens a separate window inside your browser, distinct from the one you're working in. You keep working; the agent works in its own window. 2. Explicit borrowing: when the agent needs access to one of your logged-in tabs, it must declare "I want to borrow this tab" via the bsk CLI. No silent scraping, no background monitoring—explicit requests only. 3. Return after use: once the task is done, the tab is returned. The agent cannot retain access. 4. Human-in-the-loop: on CAPTCHAs, login confirmations, or 2FA, the agent pauses and asks you to step in, then continues.

Together, these form a permission model: browser access is not all-or-nothing, but "borrow on demand, return after use, human intervention at critical moments."

Why "Borrowing Tabs" Is Harder Than "Launching a New Browser"

Launching a new browser just means starting a Chromium instance. Borrowing tabs requires:

• A browser extension: a Chrome/Edge extension that communicates with your browser, knowing which tabs exist and which services they're logged into.
• A local daemon: the bsk CLI talks to a local daemon, which talks to the extension. This chain ensures the agent never directly touches your cookies or sessions.
• Session isolation: while borrowing a tab, the agent's actions run in an isolated context. Your other tabs can't see the agent's operations, and the agent can't see your other tabs.

The cost: you must install a CLI and a browser extension. The payoff: the agent can access all your logged-in services without any API keys or passwords.

"Any Shell-Capable Agent Can Use It": CLI as a Universal Interface

A key design decision: no binding to a specific agent framework. All functionality is exposed through the bsk CLI, so any agent that can call shell commands can use it.

This means Cursor, Claude Code, Codex, OpenClaw, CodeBuddy, WorkBuddy, Pi, Hermes Agent, DeepSeek Harness—any of them just need one toolbox entry: "invoke bsk commands," and they instantly gain browser capabilities.

It echoes the Unix philosophy: tools should communicate through text streams, not shared memory. BrowserSkill turns "browser operation" into a set of shell commands. Compared with Playwright's Python/Node SDK bindings or browser-use's Python-first design, the CLI-first strategy is a smart choice in today's fragmented agent ecosystem.

Agent Window: From Shared Screen to Parallel Desktops

Traditional tools either run headless (you can't see what's happening) or take over your browser fullscreen (you can't do anything). Agent Window is a third path: your window and the agent's window coexist in parallel without interference. You have visibility without giving up your browser.

The underlying judgment: agent browser operations mostly don't need human supervision, but occasionally need human intervention. Running in parallel by default, pausing only when needed, fits real workflows better than either full automation or full manual control.

Comparison with Other Browser Automation Solutions

| Dimension | Playwright/Puppeteer | browser-use | BrowserSkill | |------|---------------------|-------------|--------------| | Login state | None (new instance) | None (new instance) | Yes (borrows real tabs) | | Framework binding | SDK bindings | Python-first | CLI-first, any shell agent | | Human intervention | None | Screenshot + click | Explicit human-in-the-loop | | User browser interruption | None (separate instance) | None (separate instance) | None (Agent Window) | | Install complexity | pip install | pip install | CLI + browser extension |

BrowserSkill's unique edge is the login-state column. Other tools solve "how to simulate a human operating a browser"; BrowserSkill solves "how to let an agent reuse browser sessions a human has already established."

Use Cases and Limitations

Best suited for:

• Operations requiring login state: reviewing GitHub PRs, updating Jira tickets, filling forms in intranet systems—where API keys are unavailable or insufficiently privileged.
• Agent + human collaboration: the agent handles most steps; the human handles CAPTCHAs and confirmations.
• Parallel multi-agent work: different agents borrow different tabs independently.

Limitations:

• Requires the browser extension: no Firefox support yet (planned), no Safari.
• Local only: the agent and browser must be on the same machine; no remote agent controlling a remote browser.
• Single-user design: no multi-user sharing of one browser instance.

Deeper Insight: From "Agent Simulates Human" to "Agent Collaborates with Human"

The "borrow a tab" model reflects a broader paradigm shift: agents no longer try to fully replace humans operating a browser, but share the browser with humans.

This shift acknowledges that agents still need humans for certain operations (CAPTCHA recognition, 2FA, complex decisions) rather than pretending those don't exist—while letting agents leverage already-established login sessions instead of simulating human logins from scratch.

From a security standpoint, the model is also more controllable: the agent's browser access is borrowed, temporary, and revocable—not owned, permanent, and unauditable. Every tab borrow is an explicit event that can be logged and audited.

---

The repository is at https://github.com/Tencent/BrowserSkill . Gaining roughly 1,350 stars in a day shows the pain point is real. If you use Claude Code, Cursor, or Codex and frequently need agents to operate logged-in services in your browser, BrowserSkill is worth a try.

"Borrowing tabs" may become the standard interaction model between agents and browsers—like borrowing a car being more convenient than renting one, as long as you trust the borrower.