Choose the browser identity, inspect state, and make handoffs explicit

BrowserAct Skills

Browser tasks with a traceable session

Connect local browser work to a coding assistant: read and act on page state, distinguish login identities from task sessions, and hand personal verification to a human. The repository also contains Skill Forge and a catalog of site-specific solutions.

By BrowserAct Team

Setup & first task View on GitHub

Build a scraper from one prompt

View example source

Working environment

An assistant that reads Skills and runs shell commands
Local Chrome / Chromium
A separate BrowserAct account when required

GitHub stars: 6,300
Version: 2.0.2
Checked: 2026-10-30
Categories: Engineering Marketing & Growth Other

Start with one read-only Chrome session

Confirm Skill discovery, a real browser ID, page reading, and closing your own session. Add login reuse, pagination, or human assistance only after that path works, checking the actual result at each step.

Pick a task and get started

Send the setup request below to your chosen assistant to prepare both the entry and CLI. Then run the first read-only exercise, using the complete get-skills response for the actual installed commands.

Choose first-page reading, bounded collection, or a signed-in page check.

Read one public page to check the setupTurn two directory pages into a traceable CSVCheck a signed-in page with a human handoff

Use example.com from the author’s quick start to discover the environment, read a page, and close the session.

Have the assistant install the entry and separate CLI
Send this request to the assistant; it is not a shell command. Prepare both the entry directory and separate CLI, with uv and Python 3.12 available. The pinned link preserves reviewed Skill 2.0.2; the uv command retrieves the CLI available at installation time. Check compatibility through get-skills afterward.
```
Add the browser-act entry Skill from https://github.com/browser-act/skills/tree/11c057b03f92101642cadc9f840564574120d184/browser-act to the Skill directory used by this client. Read the author’s installation guide. If the separate CLI is missing, install it with uv tool install browser-act-cli --python 3.12. Preserve an existing working setup. Verify browser-act --version, activate the installed Skill, and read browser-act get-skills core --skill-version 2.0.2 in full. Report the Skill path, CLI version, and missing requirements; browser creation belongs to the first task.
```
Read the author’s setup guide
Read the runtime guide before choosing a browser
Use browser-act --version during installation verification, then activate the entry and read the command below in full. Check browser type, purpose, actual ID, and current sessions. Load get-skills advanced when Profile, proxy, or privacy settings are relevant. The ordinary first exercise needs no account or imported login.
```
browser-act get-skills core --skill-version 2.0.2
```
Read the two-layer Skill and version workflow
Give your AI this first task
These tasks and the text illustration are editorial examples based on author documentation, not observed runs. Replace bracketed or angle-bracket inputs and keep observed results separate from checks still required.
Use the installed browser-act Skill. Read the complete output of browser-act get-skills core --skill-version 2.0.2, including CLI version, entry version, browsers, and active sessions. If the versions are incompatible or requirements are missing, report what needs attention; a command written in a report is not evidence that it ran. Read only https://example.com. Select an ordinary chrome browser intended for public-page work and verify its actual ID and purpose. If none is suitable, follow the current get-skills instructions and confirmation process to create a clean chrome environment. Do not import a personal Profile, attach through chrome-direct, configure a proxy, or sign in to a service. Create a unique session name for this task. Open the page with browser-act --session <task-session> browser open <actual-browser-id> https://example.com, then run state, get title, and get markdown in that session. Replace the angle-bracket values rather than copying example IDs. Confirm the returned URL, title, and body belong to the requested page, and record visible outbound links. No clicks or form submissions are needed. Write browseract-first-run.md with actual versions, browser type, URL, title, a short content summary, observed links, and errors. Close your session with browser-act session close <task-session> and check session list. Preserve browser configuration and other tasks’ sessions. Mark any unfinished check as unverified.
Check what you receive
- An actual page-reading record with environment details and a source URL.
- Observed results for reading and closing, rather than expected page text presented as a visit.
- Only the task’s named session is used and closed; existing browsers and other work remain intact.
Read the author’s notes for this task

Before you start

Prepare both the entry and CLI: Install the entry from the browser-act directory and the browser-act-cli package from PyPI. The entry declares 2.0.2; the current CLI release is 1.5.0. These are separate version lines, with compatibility checked through get-skills.
Use the Python 3.12 installation path: Have uv, an available Python environment, and Chrome/Chromium. The author broadly says Python 3.12+, but CLI 1.5.0 currently declares the 3.12 series in its release metadata. Retain the documented --python 3.12 command and check distribution support for your system architecture.
Define the page and stopping point: Supply the URL, fields, page or record limit, and login requirements. The first exercise only reads example.com. Use a configured dedicated browser for account work rather than putting passwords or cookies in chat.
Separate local execution from cloud credits: The website lists no BrowserAct runtime charge for local CLI execution. Optional fingerprint browsers, proxies, and cloud Bot building/running have separate credit rules; model allowances or charges depend on the assistant service you choose. The pinned README lists stealth-extract and remote-assist among sign-in features. Check current account allowances before using them.

Who it’s for

Developers · Professionals · Founders · Creators

Know the boundaries

This page pins entry version 2.0.2. get-skills was not run, so the installed CLI’s returned workflow and compatibility remain unverified.
Skill-level confirmation and session rules do not replace client-side action controls. Login, import, proxies, and remote assistance each have requirements.
Check local execution, cloud Bots, optional services, and model usage separately. Access restrictions may prevent completion.
MIT applies to this repository’s Skills and examples; it does not establish the license of every CLI implementation, dependency, or hosted service.

Install and run in your chosen tools, then check the output against your own materials.

More about this skill

The local browser-act Skill is a discovery entry rather than the complete operating manual. After loading browser-act/SKILL.md, the assistant requests get-skills core for instructions matched to the installed CLI, plus environment information, browsers, and sessions. Copying the entry file alone does not provide a running browser.

The basic loop is to open a named session, inspect state, act on the current element indices, and inspect state again. Indices can expire when the page changes. Extracted text also needs to be organized and checked; a successful command does not establish that every row in a dataset is correct.

A browser holds an identity and login environment; a session represents a task. Multiple sessions on one browser still share cookies and authentication. Profile import is a snapshot, while chrome-direct controls the actual local Chrome. These distinctions matter when keeping an account-specific report consistent.

When personal verification is required, remote-assist can return a handoff link. Inspect the page again after the human finishes before resuming. The separate Skill Forge entry generates SKILL.md and scripts for repeated site work, while solutions supplies site-specific packages. Each needs its own setup, and managed cloud Bots are a separate product path.

Sources and verification

Sources checked 2026-10-30

Full reads cover both pinned entry Skills, README, license, requirements, all docs pages, and the Solutions catalog. Official pricing text and selected PyPI release fields were also checked. The 101 solution bodies were not reviewed, and no CLI installation, runtime Skill discovery, browser launch, or collection was performed.

Author documentation Original skill CLI 1.5.0 release metadata BrowserAct local and cloud pricing Repository MIT license

Move from browser work to evidence review

Agent Browser offers another browser CLI entry. Deep Research Skills can help define fields and organize the evidence collected afterward. Inspect each tool’s own installation and prerequisites.

agent-browserPage snapshots and element references for gathering data, testing forms and resuming signed-in work.Deep Research SkillsTurn a topic into an item list and fields, research each into JSON, then write a Markdown report.Wondelai Skills51 product, design, marketing, management and engineering skills plus 14 guided journeys.NotFair48 marketing skills for page reviews, content calendars, ad analysis and account operations.

Start with one read-only Chrome session

Pick a task and get started

Have the assistant install the entry and separate CLI

Read the runtime guide before choosing a browser

Give your AI this first task

Check what you receive

Sources and verification

Move from browser work to evidence review