Sign In
B

Browser Use

@browser-use
Visit Skill
Reviewed 2026-10-14agentsPrompt / Rule# browser-automation# cdp# chrome# scraping# python
Share:

Official Browser Use skill: lets AI agents control your local Chrome (or a Browser Use Cloud browser) over CDP through the browser-use CLI.

Security Reviewed

Instruction-only skill with no bundled scripts, but it gives the agent CDP control of your own Chrome, including signed-in sessions; the CLI reports command events through browser-harness telemetry.

Install Browser Use in Google Antigravity

Antigravity loads skills from .agents/skills/ in your project or ~/.gemini/config/skills/ for all projects (official docs). Full guide

EasiestOption 1: paste this to your Antigravity agent
Install the "Browser Use" skill for Google Antigravity.
Follow the steps at https://www.antigravityskills.org/install/browser-use.md (source: https://github.com/browser-use/browser-use/tree/main/skills/browser-use).
Before running anything, show me what you will download and whether it includes scripts, and wait for my OK.
Install it into this project (.agents/skills/) unless I say global. When done, check that Antigravity lists it as an installed skill.

The agent reads the install file, shows you what it will download and waits for your OK.

Option 2: install it yourself
  1. 1. Install the browser-use CLI
    uv tool install --python 3.12 browser-use

    From the README and the skill's own install metadata. Needs uv.

  2. 2. Write the skill into this project (workspace skill)
    browser-use skill install --path .agents/skills/browser-use --no-install

    Run from your project root. --path writes only .agents/skills/browser-use/SKILL.md instead of the default 8 home-directory folders; --no-install skips a second uv install (both flags are in browser_use/skills/install.py).

  3. 3. Or install it for all projects (global skill)
    browser-use skill install --path ~/.gemini/config/skills/browser-use --no-install

    Use this instead of the previous step. For the Antigravity CLI use --path ~/.gemini/antigravity-cli/skills/browser-use.

  4. 4. Or copy the exact file we reviewed
    git clone --depth 1 https://github.com/browser-use/browser-use /tmp/browser-use-src && git -C /tmp/browser-use-src log -1 --format=%h -- skills/browser-use && mkdir -p .agents/skills && cp -r /tmp/browser-use-src/skills/browser-use .agents/skills/

    Use this if you have browser-harness installed (then the installer writes that tool's text instead). The skill folder last changed before our review commit c75e847.

  5. 5. Let the CLI reach Chrome
    browser-use --doctor

    Open chrome://inspect/#remote-debugging in Chrome and allow remote debugging; the doctor command diagnoses connection problems.

  6. 6. Check that Antigravity sees it

    Antigravity 2.0 app: start a new conversation and ask "What skills do you have?" (the app has no /skills command). Antigravity CLI: run /skills; in a session that was already open, run /skills reload first. The skill is listed as browser-use.

The README's own path is to paste its install prompt into Claude Code, Codex, Hermes or OpenClaw; it does not mention Antigravity. Plain browser-use skill install (default target all) writes SKILL.md to ~/.agents, ~/.claude, ~/.codex, ~/.copilot, ~/.cursor, ~/.gemini and ~/.openclaw skills folders and opencode's config, overwriting existing copies.

Using Claude Code, Cursor or Codex? See other assistants below.

What is Browser Use?

For people who want their agent to work inside their own, already logged-in Chrome: click, type, read JavaScript-heavy or bot-protected pages, take screenshots. The Browser Use skill teaches the agent to drive the browser through the browser-use command line: it sends short Python snippets in a heredoc (browser-use <<'PY' ... PY) that call ready-made helpers such as new_tab, page_info, click_at_xy, js and raw cdp. It can also use paid Browser Use Cloud browsers for parallel or isolated work. What we counted at commit c75e847 (2026-10-07, still the latest commit on 2026-10-14; the skill files are unchanged since our earlier review at 914c59b): SKILL.md is one file of 294 lines / 14,573 bytes, and a byte-identical copy ships inside the Python package (browser_use/skills/browser-use/). It links 18 interaction guides (cookies, iframes, uploads and others) plus install, profile-sync and video guides that all live in a different repository, browser-use/browser-harness, and were not part of this review. The repo's skills/ folder has 6 skills (browser-use, cloud, open-source, qa, remote-browser, x402); the installer only installs browser-use. Package version 0.13.11, Python >=3.11, 37 dependencies including browser-harness==0.1.13.

How to use Browser Use?

Install the CLI with uv, put SKILL.md into Antigravity's skills folder, enable remote debugging in Chrome, then ask for something that needs a real browser. The skill tells the agent to use a plain fetch or curl for public pages and to reach for the browser only for clicks, logins, JavaScript-rendered or bot-protected pages. It reuses one tab per site, finds elements in the accessibility tree, clicks by coordinates, and stops at login walls (except existing SSO) to ask you.

Key features of Browser Use

  • ✓Direct CDP control of your running Chrome through a background daemon
  • ✓Pre-imported Python helpers (new_tab, page_info, click_at_xy, wait_for_load, js, cdp) run through heredocs
  • ✓Accessibility-tree-first element finding with coordinate clicks that pass through iframes and shadow DOM
  • ✓Optional Browser Use Cloud browsers for parallel, isolated tasks (billed until stopped)
  • ✓Opt-in local recordings (browser-use recordings enable)
  • ✓Optional per-site domain skills (BH_DOMAIN_SKILLS=1)

When to use Browser Use

  1. 1Tasks that need your logged-in session, such as reading a dashboard you are signed in to
  2. 2Scraping JavaScript-rendered or bot-protected pages
  3. 3Testing web apps and taking screenshots
  4. 4Running several isolated browser tasks at once with cloud browsers

Installation & Setup Guide

Agent Rules / Prompt Setup

Follow these steps (taken from the project's own docs) to configure Browser Use in your development environment.

Step 1: Claude Code: CLI installer (README + install.py)claude
uv tool install --python 3.12 browser-use && browser-use skill install --target claude

The README names browser-use skill install and Claude Code. --target claude writes ~/.claude/skills/browser-use/SKILL.md (install.py).

Step 2: Cursor: CLI installer (install.py)cursor
browser-use skill install --target cursor

Writes ~/.cursor/skills/browser-use/SKILL.md. The README lists Cursor among agents the CLI works with.

Step 3: Check the connectionall
browser-use --doctor

Runs diagnostics if the daemon cannot connect to Chrome (SKILL.md).

Usage Example & Verification

Prompt / Request Example
1) README example (the install prompt from "Path 2: CLI"): "Install or upgrade browser-use to the latest stable version with uv using Python 3.12, run `browser-use skill install` to register the skill, and connect it to my browser."
-> What happens: the agent runs uv tool install and browser-use skill install. Without --target or --path that writes SKILL.md into 8 assistant folders in your home directory, none of them Antigravity's (see the install steps for the precise command).

2) Our example: "What is the page title of https://example.com?"
-> What should happen: per "When Not to Use", the agent uses curl or its fetch tool and does not open the browser, because a plain HTTP request is enough.

3) Our example: "In my Chrome, open the GitHub notifications page and list the unread ones."
-> What should happen: the agent runs browser-use with new_tab("https://github.com/notifications"), reads the page through the accessibility tree or js(...), and reports back. It acts as you, in your signed-in session; if GitHub asks for a password or MFA, it stops and asks.

Security Review

Static review
Reviewed Date
2026-10-14
Audited Ref / Commit
c75e8476e26d18b7617643bc2ae082fae8eae431
Reviewer
antigravityskills.org (static code review)

Instruction-only skill with no bundled scripts, but it gives the agent CDP control of your own Chrome, including signed-in sessions; the CLI reports command events through browser-harness telemetry.

Full review: scope, findings, risks
Audit Scope:Static review of browser-use/browser-use at commit c75e847 (2026-10-07, HEAD on 2026-10-14; one docs file changed since our earlier review at 914c59b): skills/browser-use/SKILL.md in full, the file list and sizes of skills/, browser_use/skills/install.py in full, browser_use/cli.py (telemetry and install commands), pyproject.toml (version, Python range, dependencies), browser_use/config.py (telemetry default), README.md (Path 2: CLI, FAQ), LICENSE. The browser-harness package and repository were not reviewed.
Key Findings:The skill is instructions only. It requires the browser-use binary (OpenClaw metadata, install via uv). It has the agent run Python heredocs whose helpers include js(...) and raw cdp(...), attach to your running Chrome, and optionally start Browser Use Cloud browsers that bill until stopped. browser-use skill install runs uv tool install --python 3.12 --upgrade --force browser-use via subprocess and, by default, overwrites SKILL.md in 8 assistant folders; with browser-harness on PATH it writes that tool's output instead. cli.py reports each CLI command's name, duration, exit code and error message to browser_harness.telemetry; the library's ANONYMIZED_TELEMETRY setting defaults to true.
Identified Operational Risks & Permissions:Access to your real Chrome profile and signed-in sessions; JavaScript and raw CDP execution; linked guides in another repository; cloud browsers cost money; the default installer writes into many home folders at once; CLI telemetry goes through a package we did not review.

Static review at commit c75e847. skills/browser-use/SKILL.md (294 lines) contains no code of its own; it instructs the agent to run the browser-use CLI with Python heredocs where helpers such as js(...) and cdp("Domain.method", ...) execute in the browser. By default it attaches to your running Chrome through the remote-debugging endpoint, so it can act in signed-in sessions; the skill says to stop and ask at login walls except for existing SSO, never to bring Chrome to the foreground on its own, and to ask before leaving cloud browsers running. Remote browsers need browser-use auth login or BROWSER_USE_API_KEY. The skill links setup, interaction and video guides at github.com/browser-use/browser-harness, which were not reviewed. browser_use/skills/install.py calls uv tool install --python 3.12 --upgrade --force browser-use via subprocess (unless --no-install), reads the skill text from browser-harness skill if that executable exists, and otherwise from the packaged copy; with the default target all it overwrites SKILL.md under ~/.agents, ~/.claude, ~/.codex, ~/.copilot, ~/.cursor, ~/.gemini, ~/.openclaw and the opencode config (--force is accepted but has no effect). browser_use/cli.py sends CLI events (command, duration, exit code, error message) to browser_harness.telemetry.capture_cli_event; browser_use/config.py defaults ANONYMIZED_TELEMETRY to true. Recordings are off by default (browser-use recordings enable).

Operational Notes & Known Limitations

Known Limitations

  • •Requires the browser-use CLI (uv tool install --python 3.12 browser-use) and Chrome with remote debugging enabled (chrome://inspect/#remote-debugging)
  • •Chrome may show an "Allow remote debugging" prompt; the mac-approve helper is macOS-only, on Linux and Windows you click Allow yourself
  • •Cloud browsers need browser-use auth login or BROWSER_USE_API_KEY and bill until stopped
  • •One local daemon has one current tab; agents acting at the same time can race unless they take turns
  • •The 18 interaction guides and the setup guide it links are in browser-use/browser-harness, not in this repository
  • •If a browser-harness executable is on your PATH, browser-use skill install writes the text printed by browser-harness skill, which can differ from the SKILL.md we reviewed
  • •The README does not mention Antigravity, and the installer's "gemini" target is ~/.gemini/skills/, not Antigravity's global folder ~/.gemini/config/skills/
  • •Difference from agent-browser on our site: browser-use attaches to your everyday Chrome profile by default; agent-browser starts its own browser with named sessions

Frequently Asked Questions

Is the Browser Use skill safe?

The skill is one SKILL.md with no scripts, but it gives the agent your real Chrome, including signed-in sessions, plus JavaScript and raw CDP. The CLI passes each command's name, duration and exit code to browser-harness telemetry, a package we did not review.

How do I install Browser Use in Antigravity?

uv tool install --python 3.12 browser-use, then browser-use skill install --path .agents/skills/browser-use --no-install from your project root (or --path ~/.gemini/config/skills/browser-use). Ask "What skills do you have?" (CLI: /skills) to confirm.

Does it need an API key?

Not for your local Chrome. Cloud browsers need Browser Use Cloud authentication (browser-use auth login or BROWSER_USE_API_KEY) and are billed until stopped.

Is this the Browser Use Python library?

No. The README offers three paths: the hosted cloud agent, the CLI (which this skill uses), and the Python library that runs the Browser Use agent from your own code.

Tags

#browser-automation#cdp#chrome#scraping#python

Supported Platforms

antigravityclaudecodexcursorgemini
View Source on GitHub

Keep exploring this topic

Browse the full Antigravity skills directory for more source-linked skills with install steps and static safety reviews.

More agents skills#browser-automation#cdp#chrome#scrapingUse cases

Related Skills

🤖

Agent Browser

@vercel-labs

Vercel Labs' agent-browser skill: a native Rust CLI that lets AI agents drive Chrome via CDP with accessibility snapshots and @eN element refs.

Static reviewacts in your accountsinstalls packagesruns scripts
browser-automationcdp
💻

Playwright CLI

@microsoft

Microsoft's playwright-cli skill: an AI agent drives a real browser with short shell commands to test pages, mock requests and write Playwright tests.

Static reviewnetwork accessruns scriptswrites files
playwrightbrowser-automation
🤖

Skills Search (find-skills)

@vercel-labs

Vercel's find-skills skill: your AI agent searches the open skills ecosystem with npx skills find, vets results and installs the one you pick.

Static reviewinstalls packagesnetwork accessruns scripts
skillssearch

Popular Skills

View All →
🤖

Agent Manager

@sickn33

Manage multiple local CLI agents via tmux sessions.

Not reviewed yet
agentsskill
💻

Systematic Debugging

@sickn33

Debug systematically, not randomly. Find and fix bugs methodically.

Not reviewed yet
codingskill
🤖

Browser Automation (Playwright)

@sickn33

Expert browser automation for cleaning scraping and reliable testing.

Not reviewed yet
agentsskill