A high-performance headless browser automation CLI designed for AI agents using deterministic element selection through accessibility trees.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-clawdbot-local
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-local 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 automation tool within the Openclaw Skills ecosystem that enables AI agents to interact with websites with high reliability. Unlike traditional automation tools that rely on fragile CSS selectors or XPaths, this skill utilizes accessibility tree snapshots and unique reference IDs. This approach ensures that agents can identify and interact with interactive elements deterministically, even on complex single-page applications.
By prioritizing the accessibility tree, Agent Browser provides a streamlined view of the DOM that is optimized for LLM processing. This makes it an essential component for developers building robust Openclaw Skills that require navigation, form filling, and data extraction across diverse web environments.
Agent Browser Use Cases
- Automating multi-step workflows across complex web applications.
- Developing agents that require session isolation for multi-user testing.
- Extracting data from dynamic pages where performance and deterministic selection are critical.
- Bypassing repetitive login flows by saving and loading authentication states.
How Agent Browser Works
- The agent opens a target URL using the command-line interface to initiate a browser instance.
- A snapshot of the page is generated, focusing on interactive elements and creating a mapping of unique references (refs).
- The AI agent parses the JSON snapshot to identify the specific ref (e.g., @e1) corresponding to the desired UI element.
- Specific actions like clicking, filling, or hovering are executed using these deterministic references rather than selectors.
- The agent monitors page stability and re-snapshots after navigation or state changes to maintain synchronization with the UI.
Agent Browser Setup
To begin using this tool within your Openclaw Skills projects, install the package and its dependencies via npm:
npm install -g agent-browser
agent-browser install
For Linux environments requiring system-level dependencies, use:
agent-browser install --with-deps
Agent Browser Data Schema & Taxonomy
The skill returns structured JSON data to ensure compatibility with AI workflows. Below is the primary schema for snapshots:
| Field | Type | Description |
|---|---|---|
| success | boolean | Indicates if the operation was successful. |
| data.snapshot | string | A textual representation of the accessibility tree for agent analysis. |
| data.refs | object | A mapping of element IDs to their roles and names (e.g., @e1: {role: button}). |
| state files | JSON | Persistence files containing cookies and local storage for session management. |
Agent Browser Advanced Features
- Multi-session support allows running isolated browser contexts simultaneously for different user roles within the same script.
- Network routing and interception capabilities for blocking unwanted resources or mocking API responses.
- State persistence enables saving and loading authentication cookies to skip tedious login sequences.
- Tab and iframe management allows agents to navigate complex multi-window or nested web environments.
- Headed mode support provides a visual interface for debugging and real-time inspection during skill development.
SKILL.md
Loading
Related Openclaw Skills
AI Bug Fixer
An intelligent AI assistant that automatically diagnoses, analyzes, and repairs code bugs and runtime exceptions.
huahua-spirits
A deterministic AI companion system that provides unique spirit partners with evolving personalities and emotional resonance.
Multi Search Engine
A comprehensive search integration providing access to 17 global and domestic search engines with advanced filtering and zero API key requirements.
DingTalk CLI Auto Skill
An automation powerhouse for DingTalk that leverages the official dws CLI to manage enterprise communications, schedules, and tasks through an AI agent.
Skill Combo: Multi-Skill Coordination Engine
A sophisticated orchestration engine that enables AI agents to execute multiple skills simultaneously for synergistic 1+1>2 results.
Acceptance Test
Automate business requirement verification and user scenario testing to ensure high-quality software delivery.
Featured*
