Dev & Engineering browser-automationcliextensionchromeedgecursorcodex

BrowserSkill

Let AI agents use your real, logged-in browser without interrupting your work.

FollowAgents review · FARS-2.1
Not recommended
49/ 100 5-point scale 2.5 / 5
1 2 3 4 5 6
1Trust8 / 29 · 1.4/5

Evidence shows: README explicitly states that the agent must explicitly borrow tabs and return them after task completion, and has built-in human-in-loop, indicating design intent for least privilege and user confirmation. However, no implementation details such as permission lists or confirmation dialog triggers are provided, and data flow transparency is not explained. Dependency security: Cargo.toml lists dependencies but no vulnerability scanning or audit evidence. External effects: extension automates browser actions but impact scope not detailed. Rollback not mentioned. Source attribution: repository attributed to Tencent but publisher unverified. Thus most criteria score 1, rollback 0.

2Reliability8 / 14 · 2.9/5

Evidence shows: README, Cargo.toml, package.json, and CI config are structurally consistent, indicating good internal consistency. Dependency versions are pinned in Cargo.toml and CI uses --locked for reproducible builds, so dependency availability is good. But failure messages: no specific error handling or user prompts examples, so score 1.

3Adaptability12 / 18 · 3.3/5

Evidence shows: README clearly lists supported agents (Cursor, Claude Code, etc.) and OS/browsers, audience and scenarios clear. Capability boundaries: agent can only operate browser via CLI and must borrow tabs, boundaries clear. Trigger precision: provides specific command example (/browser-skill open example.com) but not detailed trigger conditions. Environment fit: supports multiple OS and browsers, but Firefox only planned. Thus each criterion scores 2.

4Convention9 / 18 · 2.5/5

Evidence shows: README well-structured with quick start, how it works, developer guide, good information architecture. Install notes detailed with automatic and manual steps. Naming consistent (bsk). Examples and FAQ: basic example but no FAQ. Known limitations: only mentions Firefox planned, not other limitations. License MIT clear. Versioning: version in Cargo.toml but no CHANGELOG. Maintenance responsibility: repository attributed to Tencent but maintainers not explicit. Thus most score 2, examples/FAQ, known limitations, versioning, maintenance score 1.

5Effectiveness9 / 13 · 3.5/5

Evidence shows: Output usability: CLI provides commands and skill installation, output should be directly usable. Marginal value: reusing login state and keeping user work uninterrupted are clear advantages. Cost-benefit: installation simple but requires extension and skill, low cost. Thus each scores 2.

6Verifiability3 / 8 · 1.9/5

Evidence shows: README claims (e.g., support for multiple agents) are partially supported by code and config, but no independent verification. Cross-source corroboration: only internal repository evidence, no external validation. Fact-inference separation: README distinguishes facts (support list) and inferences (expected support) but not clearly. Thus each scores 1.

Evidence confidence: Low Reviewed Aug 13, 2026 Reviewed revision 4549ed08923b
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Safety controls not found in source: rollback or recovery path
Before you use it
  • Publisher identity unverified, treat source with caution.
  • No rollback mechanism provided, errors in automation may be hard to recover.
  • No dependency security audit evidence, recommend checking for vulnerabilities.
  • Data flow transparency insufficient, users may not know how agent accesses browser data.
Review evidence [1][2][3][4][5][6]
See the full review method →

What does this agent do, and when should you use it?

BrowserSkill, open-sourced by Tencent, is a local bridge that connects shell-capable AI agents like Cursor, Claude Code, Codex, and OpenClaw to your real browser via a CLI (bsk) and a browser extension. It reuses your authenticated sessions, executes tasks in a separate Agent Window, and avoids interfering with your normal browsing. Core components include the bsk CLI/daemon and a browser extension, supporting macOS, Linux, Windows, and Chrome/Edge. Its unique tab-borrowing mechanism ensures agents only use tabs you explicitly allow and return them after tasks. Licensed under MIT, it provides prebuilt skill installation for several agent frameworks.

BrowserSkill bridges the agent's shell commands (bsk CLI) to the browser extension via local WebSocket on 127.0.0.1. The agent calls bsk commands to request browser tasks; the daemon routes them to the extension, which automates actions in an Agent Window (e.g., opening pages, reading content). It supports borrowing individual tabs for agent use and returning them afterward. Built-in human-in-the-loop pauses for captchas or confirmations. Installation uses bsk install-skill to place skill files (skill/SKILL.md) into specific agent harnesses, teaching them to use bsk.

  1. Developers using Cursor can have the agent interact with their logged-in GitHub or internal systems without separate test accounts.
  2. Claude Code users need the agent to access personal Gmail or cloud drives without exposing credentials, by leveraging existing login state.
  3. During multi-step web operations, when captchas or confirmations arise, the agent can hand over to the user and resume afterward.
  4. Users who want the agent to operate on a specific tab without disturbing other windows, via the tab-borrowing mechanism.
  5. Codex or OpenClaw users on Windows or Linux quickly integrate browser automation through the bsk CLI.

What are this agent's strengths and limitations?

Pros
  • Reuses real login state, eliminating the need for test accounts, ideal for internal systems.
  • Separate Agent Window keeps user browsing undisturbed.
  • Supports multiple agent frameworks (Cursor, Claude Code, Codex, OpenClaw) with no model lock-in.
  • Built-in human-in-the-loop handles captchas and confirmation dialogs.
Limitations
  • Only supports Chrome/Edge and other Chromium browsers; Firefox is planned but not yet.
  • Requires installing a local daemon and extension, adding deployment complexity.
  • Browser automation is limited by extension capabilities, potentially not supporting all advanced operations.
  • Tab-borrowing requires explicit user authorization, which may add interaction latency in automated workflows.

How do you install or deploy this agent?

  1. Install the CLI: On macOS/Linux run curl -fsSL https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.sh | sh; on Windows PowerShell run irm https://raw.githubusercontent.com/Tencent/BrowserSkill/main/install.ps1 | iex.
  2. Verify with bsk --version.
  3. Install the browser extension from the Chrome Web Store: https://chromewebstore.google.com/detail/hhcmgoofomhgciiibhipgmgkgnoenaoi
  4. Run bsk install-skill, use Space to select the target agent harness (e.g., Cursor, Claude Code), then press Enter.

How do you use this agent?

After installation, start a new agent session and prompt, for example: /browser-skill open example.com and summarize what is on the page. The agent will execute browser tasks via the bsk CLI, displaying results in the Agent Window.

FAQ

Is BrowserSkill free?
It's open source under the MIT license, so free to use. Note that the browser extension is installed from the Chrome Web Store, subject to its terms.
Do I need to create an account?
No. BrowserSkill reuses the existing login state in your browser; no separate account is needed.
What happens if the agent hits a captcha?
The agent pauses and asks you to manually complete the captcha or confirmation dialog, then continues after you're done.
Which operating systems are supported?
macOS (Apple Silicon and Intel), Linux (x64 and ARM64), and Windows x64.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents