A comprehensive command-line toolset that enables AI agents to interact with web browsers through programmatic navigation, element interaction, and visual snapshots.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-8
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-browser-8 using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is agent-browser?
agent-browser is a purpose-built CLI designed to give AI agents full control over a web browser using the Chrome DevTools Protocol (CDP). It allows agents to perform actions like navigating to URLs, filling out forms, clicking buttons, and extracting data from the accessibility tree. As a core part of the Openclaw Skills ecosystem, it provides a structured way for agents to handle the complexity of the modern web, including dynamic content and authentication.
The tool is unique because it translates the complex DOM into a simplified accessibility tree with reference IDs (like @e1), making it easy for LLMs to identify and interact with specific page elements. With built-in support for session persistence and security boundaries, it ensures that web automation is both reliable and safe for autonomous agents.
agent-browser Use Cases
- Automating complex multi-step login flows and form submissions.
- Scraping data from dynamic, JavaScript-heavy websites that require user interaction.
- Performing visual regression testing and UI verification via snapshot diffing.
- Capturing annotated screenshots to help vision-capable agents reason about layout.
- Managing authenticated sessions across multiple browser restarts using state files.
How agent-browser Works
- Initialization: The agent starts a browser session or connects to an existing one using a named session.
- Navigation: The agent uses the open command to load a target website and waits for the network to settle.
- Discovery: A snapshot command is executed to generate a list of interactive elements with unique @ref tags.
- Interaction: The agent issues commands like click or fill using the discovered references to manipulate the page.
- Re-discovery: After any navigation or DOM change, the agent re-snapshots to get updated references.
- Extraction: The agent retrieves text, screenshots, or PDF exports to satisfy the user request.
agent-browser Setup
Install the CLI globally using npm, Homebrew, or Cargo:
npm i -g agent-browser
# or
brew install agent-browser
Download the required browser binaries:
agent-browser install
To leverage these Openclaw Skills in your agent workflows, verify the setup by opening a page:
agent-browser open https://example.com && agent-browser snapshot -i
agent-browser Data Schema & Taxonomy
The skill manages browser state and extracted information using the following structures:
| Data Type | Description | Format |
|---|---|---|
| Snapshots | Accessibility tree mappings with element references (@e1, @e2) | Markdown / JSON |
| Session States | Cookies, localStorage, and session metadata | Encrypted JSON |
| Auth Vault | Securely stored credentials and login metadata | Internal Storage |
| Screenshots | Page captures, including full-page and annotated versions | PNG / JPEG |
| Network Logs | HAR files for traffic analysis and debugging | JSON / HAR |
agent-browser Advanced Features
- Annotated Screenshots: Automatically overlays reference numbers on visual captures for vision-based reasoning.
- Content Boundaries: Wraps page output in nonces to protect the agent from prompt injection attacks.
- Auth Vault: Securely stores credentials and automates login sequences without exposing passwords to the LLM.
- Cross-Platform Emulation: Built-in support for emulating mobile devices and running Safari via iOS Simulator.
- Visual Diffing: Compares snapshots or screenshots to identify what changed on a page after an action.
- Multi-Engine Support: Capability to switch between Chrome and high-performance engines like Lightpanda.
SKILL.md
Loading
Related Openclaw Skills
Web Content Fetcher
A high-performance web scraping skill that converts complex URLs into clean, structured Markdown for AI processing.
Products Page Generator
A comprehensive tool for designing and optimizing high-converting e-commerce product listing and category pages.
Cyber Kill Chain Security Assessment
An enterprise-grade framework assessment tool that evaluates organizational security controls across the MITRE ATT&CK kill chain stages.
Network Engineer Roadmap
A professional career development platform that generates personalized learning paths for networking professionals based on skills assessment and career goals.
CCPA Privacy Checker Tool
An automated compliance assessment platform that audits 31 dimensions of data handling practices to ensure CCPA adherence.
Gmail Lead Monitor
A lightweight Python-based monitoring tool that filters Gmail inboxes for specific keywords and sends instant alerts to Telegram.
Featured*
