A high-performance Rust-based headless browser CLI that empowers AI agents to interact with web pages through structured commands and element references.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-tool
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-tool 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 specialized tool designed for the modern AI ecosystem, providing a bridge between autonomous agents and the web. Built with Rust for speed and featuring a Node.js fallback, it allows AI agents to perform complex web tasks—such as navigating URLs, clicking buttons, and filling forms—using a simplified command-line interface. By utilizing Openclaw Skills like this, developers can give their agents a "browser-as-a-service" capability that is both lightweight and highly reliable.
The core strength of Agent Browser lies in its ability to translate complex DOM structures into manageable snapshots. These snapshots assign short, stable references (like @e1 or @e2) to interactive elements, making it incredibly easy for an LLM or an automated script to target specific UI components without dealing with fragile XPath or CSS selectors manually.
Agent Browser Use Cases
- Automating multi-step web workflows and form submissions programmatically.
- Extracting structured data and accessibility trees from complex Single Page Applications (SPAs).
- Performing end-to-end UI testing and capturing visual regressions via screenshots or PDFs.
- Creating high-fidelity video recordings of web interactions for debugging or documentation.
- Managing persistent sessions and authentication states across multiple browsing instances.
How Agent Browser Works
- Initialization: The agent opens a target URL using the
opencommand, establishing a headless browser session. - State Analysis: A
snapshotis taken to generate an accessibility tree or a list of interactive elements, each assigned a unique reference handle. - Command Execution: The agent sends interaction commands (click, fill, hover) using the specific handles identified in the snapshot.
- Dynamic Waiting: The system uses
waitcommands to ensure elements are visible or network activity has idled before proceeding. - State Management: Authentication cookies or local storage can be saved to a JSON file to persist sessions across different tasks.
Agent Browser Setup
To integrate this into your workflow with Openclaw Skills, install the CLI globally and set up the necessary browser dependencies:
npm install -g agent-browser
agent-browser install
agent-browser install --with-deps
For developers building from source:
git clone https://github.com/vercel-labs/agent-browser
cd agent-browser
pnpm install && pnpm build
agent-browser install
Agent Browser Data Schema & Taxonomy
Agent Browser handles data through structured CLI outputs and session files. When using the --json flag, it provides machine-readable schemas for automation.
| Data Type | Format | Description |
|---|---|---|
| Snapshots | JSON / Text | Lists interactive elements with roles, names, and @ref IDs. |
| Session State | JSON | Stores cookies, local storage, and session metadata for persistence. |
| Media | PNG / PDF / WebM | Visual outputs from screenshots, PDF exports, or video recordings. |
| Trace | ZIP | Detailed Chrome DevTools Protocol traces for deep debugging. |
Agent Browser Advanced Features
- Session Isolation: Run multiple parallel browser instances using the
--sessionflag to prevent data leakage. - Network Interception: Route, block, or mock network requests directly from the CLI to simulate various API responses.
- Device Emulation: Set custom viewports, user agents, and geolocations (e.g., iPhone 14) to test mobile-specific behaviors.
- CDP Integration: Connect to existing browser instances via the Chrome DevTools Protocol for hybrid automation scenarios.
- Semantic Locators: Interact with elements using natural language roles (e.g.,
find role button) instead of strict selectors.
SKILL.md
Loading
Related Openclaw Skills
Chrome CDP Controller
A lightweight Puppeteer-based tool for controlling existing Chrome instances via the Chrome DevTools Protocol.
Gemini Deep Research CLI
Automate comprehensive multi-source inquiries and structured report generation using the Gemini CLI.
Dialogflow CX to CES Migration Tool
A production-grade utility for automating the migration of Dialogflow CX agents to the Google Customer Engagement Suite (CES) framework.
Stock OCR
A specialized tool for capturing K-line screenshots from financial terminals and extracting Moving Average (MA) values through high-precision OCR.
PPTX Notes Editor
A specialized technical utility for AI agents to unpack, modify, and manage PowerPoint speaker notes through direct XML manipulation.
DeepPurpose AI Skill
A comprehensive technical skill for implementing and managing the DeepPurpose molecular modeling library for advanced drug-target interaction and compound prediction.
Featured*
