Guide · Agent skill

How to use the Agent Browser agent skill: setup, commands and examples

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The Agent Browser skill (agent-browser) lets an AI agent control a Chrome or Chromium browser through the agent-browser command-line tool. It is useful for developers testing web app flows, and for anyone who needs repetitive web tasks such as filling forms, pulling text from pages, or saving screenshots and PDFs. This guide explains what the skill does, how to install it, how to ask for it, and what to watch for.

Agent Browsersipeed/picoclaw · MIT Download the skill

Quick start

1Install prerequisites. Make sure Node.js and npm are installed, then install the CLI and Chromium with npm i -g agent-browser && agent-browser install.
2Add the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
3Confirm the CLI is found. Run which agent-browser in your terminal to confirm the binary is on your path.
4Ask for a browser task. Describe the task in plain words, such as opening a page and filling a form, so the agent matches the skill.
5Review snapshots. Check the interactive elements the agent lists before it clicks or fills anything.
6Close sessions. Ask the agent to close sessions when finished, or run agent-browser close yourself.

What is an agent skill, and what does Agent Browser let the agent do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to use a capability. The agent reads the description and loads the instructions when a request matches it.

With this skill, the agent can open web pages, read the clickable elements on a page, fill in forms, press buttons, scroll, extract text, save screenshots or PDFs, and keep logged-in sessions. Instead of guessing at a website, it works through the page step by step and re-checks it after each change.

Navigate to URLs and read the page title and URL
Fill, type into, select, and check form fields
Click buttons and press keys such as Enter
Extract text and run JavaScript on the page
Take screenshots (including full page) and export PDFs

Install the Agent Browser skill

Download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder. Alternatively, in the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.

The skill also needs the agent-browser CLI, which is an npm package. The SKILL.md install line is shown below. The skill checks for the agent-browser binary and does not install it at runtime, so install it yourself beforehand.

Requires Node.js and npm to install the CLI globally
Requires Chromium, installed with the agent-browser install step
Requires a Chrome or Chromium browser available in the environment

npm i -g agent-browser && agent-browser install

Check that the tool is available

The skill begins by verifying the tool. Before you rely on it, you can run the same check yourself. If the command is not found, the agent should tell you that browser automation needs the agent-browser CLI and Chromium, rather than trying to install them.

which agent-browser

Learn the core workflow: open, snapshot, act, re-snapshot

The skill works in four steps. The agent opens a URL, takes a snapshot to list interactive elements with refs such as @e1 and @e2, acts on those refs, and then takes a new snapshot after any navigation or page change, because refs are invalidated by changes to the page.

Commands can be chained with && when you do not need intermediate output.

open <url> navigates
snapshot -i lists interactive elements with refs
click, fill, type, select, check and press act on refs
Re-snapshot after navigation or DOM changes

agent-browser open https://example.com/form
agent-browser snapshot -i
agent-browser fill @e1 "user@example.com"
agent-browser fill @e2 "secret"
agent-browser click @e3
agent-browser wait --load networkidle
agent-browser snapshot -i

Handle logins safely

The skill can reuse login state so you do not log in on every run. There are four options: importing state from your running Chrome, using a persistent profile, using a named session that saves and restores state automatically, or using a state file you save and load.

Read the safety notes before using this on a shared machine. Saved state and the ability to fill in passwords mean credentials can be exposed.

Option 1: --auto-connect state save imports from your running Chrome
Option 2: --profile keeps a persistent profile across runs
Option 3: --session-name auto-saves and restores state
Option 4: state save and state load use a file you choose

agent-browser --session-name myapp open https://app.example.com/login

Use sessions, iframes and cleanup

Iframe content is inlined in snapshots, so you can interact with iframe refs directly without switching frames. You can run parallel sessions with --session and list them with session list. Always close sessions when you finish.

Parallel sessions with --session name
List sessions with session list
Close sessions when done

agent-browser --session s1 close

What people use it for

Testing a web app's sign-up or checkout form flow during development
Filling repetitive web forms with known values
Pulling text or links from a page with eval or get text
Taking full-page screenshots or PDFs of a live page for records

Frequently asked questions

How do I install the Agent Browser skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills. You also need the agent-browser CLI, installed with npm i -g agent-browser && agent-browser install.

What do I need to run it?

You need Node.js and npm, the agent-browser npm package, Chromium installed through agent-browser install, and a Chrome or Chromium browser in the environment. The skill checks for the binary and does not install it at runtime.

Is Agent Browser free?

The skill is listed under the MIT licence. Check the licence file in the source repository for the full terms before using it commercially.

Why does the agent need to re-snapshot the page?

Element refs such as @e1 are invalidated after navigation or DOM changes, so the agent takes a fresh snapshot to get current refs before acting again.

Is it safe to use with my logins?

Use caution. The skill can save login state and fill in passwords, which creates a risk of credential exposure. Avoid it for sensitive logins or payments on shared machines.

What should I use instead if it does not fit?

If you cannot install Node.js, npm and Chromium, this tool cannot run. For sites that forbid automation in their terms of service or robots rules, do not scrape them with this skill. The source does not compare it with specific alternatives.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Agent Browser suits developers testing web app flows and people who need repeatable form, text or screenshot tasks on live pages, provided Node.js, npm and Chromium are available. Skip it for sensitive logins or payments on shared machines, for scraping sites that forbid automation, or for setups that cannot install the required tools.

Download the skill