A specialized headless browser CLI that uses accessibility trees and reference-based selection for reliable AI-driven web automation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-clawdbot
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-clawdbot 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 high-performance headless browser automation tool specifically engineered for AI agents to navigate the web with precision. Unlike standard automation tools, it leverages accessibility tree snapshots to provide deterministic element selection, making it an essential addition to your Openclaw Skills library. By focusing on semantic structures rather than volatile CSS selectors, it ensures that your AI agents remain robust even when page layouts change.
This skill empowers developers to build complex multi-step workflows, handle single-page applications (SPAs), and manage isolated browser sessions with ease. As a core component of the Openclaw Skills ecosystem, agent-browser provides a structured JSON interface that simplifies how AI models interact with real-world web interfaces.
agent-browser Use Cases
- Automating multi-step web workflows such as form submissions and data extraction.
- Performing deterministic element selection in complex React or Vue-based SPAs.
- Managing multiple isolated browser sessions for simultaneous testing or multi-user flows.
- Bypassing repetitive login sequences by saving and loading authentication states.
- Implementing automated network routing to block ads or mock API responses.
How agent-browser Works
- Initialize and Navigate: Start the browser and navigate to a specific URL using the open command.
- Snapshot Generation: Capture the page's accessibility tree using snapshot -i --json to receive a structured map of interactive elements and their references (refs).
- Element Interaction: Use the generated refs (e.g., @e1, @e2) to perform actions like click, fill, or hover with high reliability.
- State Management: Optionally save cookies and local storage to persist authentication across different runs.
- Validation and Extraction: Query the state of elements or extract specific HTML attributes and text to verify the success of the automation.
agent-browser Setup
To integrate this into your Openclaw Skills environment, install the package via npm and download the necessary browser binaries:
npm install -g agent-browser
agent-browser install
For Linux environments requiring system dependencies, use:
agent-browser install --with-deps
agent-browser Data Schema & Taxonomy
The skill outputs data in a structured JSON format to ensure compatibility with AI parsing requirements. The primary data structure involves snapshots and reference maps:
| Property | Type | Description |
|---|---|---|
| success | Boolean | Indicates if the operation was successful. |
| data.snapshot | String | A text representation of the accessibility tree. |
| data.refs | Object | A map of element IDs (e.g., @e1) to their roles and names. |
Example snapshot metadata:
- Role: The semantic role of the element (button, heading, etc.).
- Name: The accessible name or label of the element.
agent-browser Advanced Features
- Session Isolation: Run multiple independent browser contexts simultaneously using the session flag.
- Network Interception: Route, block, or mock network requests to control the testing environment.
- State Persistence: Save and load full browser states including cookies and storage to skip authentication flows.
- Headed Debugging: Run the browser in visible mode with --headed to troubleshoot complex interaction logic.
- Interactive Snapshots: Filter snapshots to only include interactive elements, reducing token usage for AI models within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Bug Report Generator
A professional utility that transforms vague bug descriptions into structured, actionable reports with reproduction steps and severity levels.
Agent Emergence Test
A comprehensive testing framework designed to probe AI agents for emergent properties, self-awareness, and cognitive capabilities beyond their explicit training.
Pixiv
A comprehensive CLI-based skill for searching, downloading, and publishing illustrations and manga on the Pixiv platform.
hex-vetter
A physical-layer hex auditing tool that detects hidden binary data, control characters, and encoding-based attacks in file packages.
Agent Registry
A mandatory discovery system for Claude Code that enables token-efficient lazy loading of specialized agents.
Audio Reply Skill
A sophisticated text-to-speech skill that enables AI agents to read web articles aloud or respond using conversational audio synthesis.
Featured*
