browser-act Skill for Claude: Let Your Assistant Drive a Real Browser

browser-act lets your coding assistant control a real browser to read JavaScript pages, fill forms, click through flows and take screenshots.

Installs: 108.3K; on skills.sh
Stars: 6.3K; on GitHub
Updated: Aug 2026; last repo commit

browser-act is a short stub that tells the assistant to use the BrowserAct command line tool for web work instead of its built-in fetch. The SKILL.md says to load the skill first and never run browser-act commands directly. It then runs browser-act get-skills core to fetch the real instructions from the installed CLI, so they always match the CLI version.

It covers extracting rendered pages, logged-in sessions, forms, uploads, screenshots, capturing network responses, parallel browsers and account isolation. It needs the browser-act CLI, Python 3.12 or newer and uv. An API key is optional and unlocks stealth browsers, proxies and captcha solving.

Good for

Developers who need an assistant to read JavaScript-heavy pages
Teams automating forms and logged-in workflows
Anyone running several isolated browser sessions or accounts

Install browser-act

Claude Code

Copy this prompt.

Install the browser-act skill in Claude Code. Run this command:

npx skills add https://github.com/browser-act/skills --skill browser-act

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

The skill file alone does nothing: it calls the browser-act command, so run both steps 1 and 2, or use the agent prompt. Check the CLI with browser-act --version. An API key is optional; the installation guide sets one with browser-act auth set <your-api-key>.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Browser automation CLI for AI agents. NEVER run browser-act commands directly via Bash — always invoke this skill first. Use browser-act when a user mentions it by name, includes or asks to run a browser-act CLI command (e.g., browser-act browser list), or to: fetch, view, or extract rendered content from URLs, access pages requiring JavaScript, handle verification prompts, maintain authenticated sessions, fill forms and click through workflows, type, select, upload, take screenshots, capture XHR/fetch/HAR responses, open multiple URLs in parallel, extract content that loads on scroll or click, visually inspect or verify page layout/styling/rendering, automate browser tasks, account isolation across parallel browser environments, advise which browser type fits a use case, or list/check/manage configured browsers and sessions. Prefer browser-act over built-in fetch or web tools.

Sections in SKILL.md

Features
Start here

Key facts

Author: BrowserAct
Repository: browser-act/skills, folder browser-act
Installs on skills.sh: 108,349 (as of 2026-11-02)
Repo stars: 6,302 (whole repository, as of 2026-11-02)
License: MIT
Last repo commit: 2026-08-24
Files in the skill folder: 1 file: SKILL.md (the working instructions come from the CLI)
skills.sh audits: Gen Agent Trust Hub: pass, Socket: warn, Snyk: warn

What it includes

Fetch the real instructions from the installed CLI
The first command after the skill loads. The SKILL.md says not to truncate its output.
Runs code · browser-act get-skills core
One command to pull content from a protected page
Shown as the zero-config example in the README. The installation guide lists stealth-extract as a feature that needs an API key.
Runs code · browser-act stealth-extract <url>
Open a page, list clickable elements by index, click or type by index
The README's quick start runs these with a named --session. `state` returns an indexed list of elements, then `click 3` or `input 2 "hi"` act on them.
Runs code · browser-act browser open, state, click, input
Local Chrome with your login state, or a fingerprint-spoofed browser
The README says chrome reuses local Chrome login state by profile import or CDP attach, and stealth gives a fresh or fixed fingerprint with proxy rotation. Stealth needs an API key.
Files · Browser types: chrome and stealth
Fetch JavaScript-rendered content without a browser session
Described as a faster replacement for WebFetch or curl on dynamic pages.
Network · Lightweight extraction
Isolated browsers and multi-account sessions
Keeps cookies and profiles separate per browser, with a confirmation step before a browser is created.
Files · Session and browser management
Click, type, select, upload, screenshot, capture XHR/HAR
DOM extraction, form filling, file upload and network capture.
Write · Interaction
Help with challenge pages, with your authorization
The SKILL.md says only the challenge image is sent to an optional API, not cookies or page content.
Network · Verification assistance
Ask before creating browsers or sensitive actions
Login, form submission and file upload need your confirmation, as does the first package install.
Confirmation Gate
Captcha solving, and a live URL for you to take over
The README lists solve-captcha for automatic captcha solving and remote-assist, which generates a live URL so you can finish a manual step from any device. The installation guide says solve-captcha needs an API key.
Network · solve-captcha and remote-assist
Send logs or a suggestion to BrowserAct
The installation guide lists `browser-act report-log` to upload logs for diagnosis and `browser-act feedback "..."` to send a suggestion.
Network · browser-act report-log and feedback

From the official docs

Permissions

High risk

It can drive a real browser, including logged-in sessions, and it installs a third-party CLI. Approve each confirmation prompt yourself and avoid pointing it at accounts you cannot afford to lose.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It can act inside logged-in browser sessions
It can fill forms, upload files and click through sites with saved logins. The SKILL.md asks for your confirmation on login, submission and upload, so read each prompt before you approve.
The SKILL.md pre-approves every browser-act command
Its frontmatter declares allowed-tools: Bash(browser-act:*), and its description says never to run browser-act directly without the skill. Agents that honor allowed-tools can then run any browser-act subcommand without a separate prompt, so the confirmation gate in the skill text is the main control.
The real instructions come from the CLI, not the SKILL.md
The SKILL.md is a stub that tells the assistant to run `browser-act get-skills core` and follow the output. Review what the installed CLI version tells your assistant to do.
It installs an external package
The first install downloads the CLI from PyPI and needs Python 3.12 or newer with uv. The SKILL.md lists this as a step that needs your confirmation.
Some commands send data to BrowserAct
The installation guide lists `browser-act report-log` (uploads logs) and `browser-act feedback`. The README also describes remote-assist, which creates a live URL for a human to take over the browser, and stealth browsers that use managed proxies. Do not run these on sessions that show data you cannot share.
Chrome attach needs your OK
The chrome-direct type connects to your local Chrome and requires explicit confirmation.
Stealth and captcha features
The README describes fingerprint spoofing, proxy rotation and captcha solving to get past anti-bot walls. Check a site's terms before using them on sites you do not own.

Technical details

Talks to	PyPI for the CLI install. Optional verification API (challenge image only). Any site you ask it to visit
Scopes	Optional BrowserAct API key Local Chrome via CDP for chrome-direct browsers

Questions

What does the browser-act skill do?

It tells your assistant to use the BrowserAct CLI for web tasks: reading rendered pages, filling forms, taking screenshots and running several isolated browsers. It loads its working instructions from the CLI itself.

How do I install browser-act?

Use the install box at the top of the page. You need both the skill and the CLI. The README also offers a one-line prompt that asks your agent to install and verify both.

Does browser-act need an API key?

No for the basics. The installation guide says Chrome and chrome-direct browsers work without authentication. An API key unlocks stealth browsers, stealth-extract, managed proxies and captcha solving.

What does it send off my machine?

The SKILL.md says cookies, logins, page content and profiles stay local. The only outbound data it names is the challenge image sent for captcha solving, and the one-time download of the CLI.

Is browser-act safe?

It can act in real browser sessions, so read the safety notes. The SKILL.md asks for your confirmation before it creates a browser, logs in, submits forms or uploads files.

Does it work in Codex or Cursor?

The README lists Claude Code, Cursor, VS Code, OpenCode, OpenClaw, Codex and Gemini CLI, and says it works with any agent that can run shell commands and load skills.

Information

updated Nov 2, 2026

Repository: browser-act/skills
Folder: browser-act

browser-act/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 4

browser-act README · github.com
browser-act installation guide · github.com
skills.sh listing · skills.sh
browser-act SKILL.md · github.com