Guide · Agent skill
How to use the Browser Automation via DOMShell agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains the Browser Automation via DOMShell agent skill (skill name cli-anything-browser), a skill that lets an AI agent drive your Chrome browser through filesystem-style commands. It is aimed at developers testing their own web apps and analysts pulling content from pages without an API. By the end you will know what the skill does, how to install it, how to phrase requests so the agent uses it, and which safety limits apply.
Browser Automation via DOMShellHKUDS/CLI-Anything · Apache-2.0 Download the skill Quick start
- 1Check Node.js. Confirm npx is available by running npx --version. If not, install Node.js from nodejs.org.
- 2Install the DOMShell extension. Add the DOMShell extension to Chrome from the Chrome Web Store and keep Chrome running.
- 3Download and unzip the skill. Download the ZIP from this page and unzip it so SKILL.md sits in ~/.claude/skills/ or a project's .claude/skills/ folder.
- 4Install the CLI. From browser/agent-harness in the CLI-Anything repository, run pip install -e . with Python 3.10 or newer.
- 5Make a request. Ask the agent to open a specific https:// URL and describe the element or action you want.
- 6Review actions. Check what the agent clicked or typed, especially on forms, and stop the daemon with session daemon-stop when finished.
What an agent skill is and what this one does
An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to use a tool. This skill is built on DOMShell's MCP server and maps Chrome's Accessibility Tree, a text map of a page's buttons, inputs, headings and landmarks, onto a virtual filesystem.
Instead of reading screenshots, the agent can open a URL, list elements with ls, move with cd, read content with cat, search with grep, and act with click and type.
- page open, reload, back, forward and info manage navigation
- fs ls, cd, cat, grep and pwd explore the page structure
- act click and act type interact with elements, including filling forms
- session commands manage status and an optional daemon mode
Install the skill from the ZIP
Download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill limited to one project. The package contains a single file, SKILL.md.
Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
Install the prerequisites and the CLI
The skill needs three things before it can work. First, Node.js with npx, which runs the DOMShell MCP server. Second, Google Chrome or Chromium with the DOMShell extension installed, and Chrome must be running before you use the CLI. Third, Python 3.10 or newer.
The source installs the CLI from the repository's agent-harness folder in editable mode. Run these commands from a clone of the CLI-Anything repository, inside browser/agent-harness.
- Check Node.js with npx --version
- Install the DOMShell extension from the Chrome Web Store link in the source
- Keep Chrome open while you run commands
cd browser/agent-harness
pip install -e .Trigger the skill with natural requests
You do not need to type CLI commands yourself. Ask the agent in plain language to open a page and do something on it, and name the site and the goal. Be specific about whether it should only read the page or also click and type.
Good requests name the page, the element and the action, for example: open the example site and list the main headings, or find the Login button on this page and click it. Avoid vague requests that give no URL, since the skill expects explicit http:// or https:// addresses.
- Open https://example.com and list the elements under main
- Search the open page for the word Pricing and show the path to each match
- Type my test address into the signup form input on my local app, then click submit
Learn the path syntax and commands
Paths look like filesystem paths. / is the document root, /main is the main landmark, and /main/div[0]/button[2] is the third button inside the first div. Indices are 0-based, so button[0] is the first button. Use .. to go up one level.
A typical exploration session opens a page, lists the root, enters a landmark and lists again, then returns to the root.
- Use --json for machine-readable output on any command
- Daemon mode (session daemon-start) reuses the connection and avoids the 1-3 second cold start on each command, which helps multi-step agent workflows
- Running the bare command cli-anything-browser starts an interactive REPL
cli-anything-browser page open https://example.com
cli-anything-browser fs ls /
cli-anything-browser fs cd /main
cli-anything-browser fs ls
cli-anything-browser fs cd /
cli-anything-browser fs grep "Login"
cli-anything-browser act click /main/button[0]Understand the safety limits
The skill can act on pages you open, and the browser shares cookies and login state across sessions. Read these limits before you give an agent access to a real browser.
By default, URLs must include http:// or https://; file, javascript, data, chrome and about schemes are blocked. Localhost and private network addresses are reachable unless you enable blocking. Page content, including ARIA labels and aria-hidden elements, can contain text that tries to manipulate the agent, so treat untrusted sites with caution.
- Do not use it to log into banking or email accounts; the agent could act with your real credentials
- Consider a guest or incognito window, separate Chrome profiles, or clearing cookies between sessions
- Set CLI_ANYTHING_BROWSER_BLOCK_PRIVATE=true to block private network addresses (off by default)
- Limit navigation to trusted domains and prefer --json output when handling untrusted pages
export CLI_ANYTHING_BROWSER_BLOCK_PRIVATE=trueWhat people use it for
- Testing a web app you are building by listing its landmarks and clicking through a flow
- Pulling headings, links or table-like content from a page that has no API
- Running repeatable, scripted browser steps inside a larger agent workflow
- Filling in a test form on a local development site
Frequently asked questions
How do I install the cli-anything-browser skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills. The CLI itself is installed separately with pip install -e . from browser/agent-harness.
What do I need to run this skill?
You need Node.js with npx, Google Chrome or Chromium with the DOMShell extension installed and running, and Python 3.10 or newer. The source also specifies a pip install of the cli-anything-browser package from the repository.
Is cli-anything-browser free?
The repository is licensed under Apache-2.0. This page does not state pricing for the skill or for any related service, so check the repository for licence terms.
Can the agent log into my accounts with this skill?
It is not recommended for sensitive accounts such as banking or email. The browser shares cookies and login state across sessions, so the agent could act with your real credentials. Use a guest window or a separate Chrome profile for anything sensitive.
Why does my command fail with npx not found or DOMShell not found?
npx not found means Node.js is not installed; get it from nodejs.org. For DOMShell not found, run npx @apireno/domshell --version. If an MCP call fails, make sure the DOMShell Chrome extension is installed and Chrome is running.
What are alternatives to this skill?
The source does not compare this skill with other tools, so this guide cannot rank them. If you need a browser tool without a Chrome extension or Node.js setup, look for alternatives that match your environment and check their documentation.
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
This skill suits developers testing their own web apps and analysts who need content from pages without an API, provided they can install Node.js, Python 3.10+ and the DOMShell extension. Do not use it for logging into sensitive accounts or for letting an agent act autonomously on untrusted sites. If you need a setup-free option or a tool that avoids shared browser sessions, pick something else.
Download the skill