AgentBrowse provides a robust browser automation layer for AI agents to interact with real websites through a specialized CLI.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agentbrowse
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 agentbrowse 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 AgentBrowse?
AgentBrowse serves as the essential browser interaction layer for autonomous agent tasks. As a key component within the Openclaw Skills ecosystem, it enables agents to perform complex web-based workflows by launching instances or attaching to existing ones via CDP. It bridges the gap between raw web content and agent intelligence by providing structured observation and action capabilities.
Built for reliability, this skill allows agents to navigate, click, and type while maintaining a high level of situational awareness. By utilizing the AgentBrowse CLI, developers can integrate seamless web navigation into their AI workflows, ensuring that agents can handle everything from simple data extraction to complex multi-step web interactions with ease.
AgentBrowse Use Cases
- Automating multi-step navigation and interaction on complex web applications.
- Extracting structured data from websites using AI-assisted extraction schemas.
- Capturing page screenshots for session debugging or evidence of task completion.
- Attaching to existing browser sessions via CDP for hybrid manual-automated workflows.
- Observing page states to make informed decisions based on visible UI elements and target references.
How AgentBrowse Works
- Initialize the session by using the Openclaw Skills compatible CLI to launch a new browser or attach to an active CDP endpoint.
- Execute an observation command to allow the AI to perceive the current page state and identify interactable elements.
- Perform actions such as clicking, typing, or selecting by referencing specific target IDs returned during the observation phase.
- Navigate directly to specific URLs when the destination is predetermined to save execution time.
- Extract structured JSON data from the page content when the goal is information gathering rather than UI interaction.
- Close the browser session securely once the task reaches a verified success state based on page evidence.
AgentBrowse Setup
To get started with this skill in your Openclaw Skills environment, ensure the CLI is installed and configured:
# Install the AgentBrowse CLI globally
npm i -g @mercuryo-ai/agentbrowse-cli@latest
# Verify the installation
agentbrowse --version
# Initialize AI features with your API key
agentbrowse init <your_api_key>
# Run a diagnostic check to ensure everything is ready
agentbrowse doctor
AgentBrowse Data Schema & Taxonomy
AgentBrowse organizes its interaction data and page state using a structured reference system to ensure precision during automation.
| Term | Description |
|---|---|
| targetRef | A unique identifier for a specific DOM element that can be acted upon. |
| scopeRef | Defines the boundaries for data extraction or observation within a page. |
| session | The lifecycle of a single browser instance or attachment. |
| screenshots | PNG files generated for debugging and state verification. |
| schema-json | User-defined JSON structures used to guide the AI extraction process. |
AgentBrowse Advanced Features
- Natural Language Observation: Use AI-assisted goals to filter page state and find relevant UI elements.
- Structured Data Extraction: Define custom JSON schemas to transform unstructured HTML into clean data.
- Session Recovery: Use the browser-status and attach commands to recover from stuck states without losing progress.
- Companion Integration: Seamlessly hand off flows to protected-flow tools for handling sensitive login or payment credentials.
- Multi-Agent Support: Compatible with complex agentic protocols that require separate browsing and logic layers.
SKILL.md
Loading
Related Openclaw Skills
Execution Loop
A sophisticated control framework for AI agents that enforces persistent execution, prevents task drift, and verifies completion through multi-layered stop hooks.
Pipeworx Bored
A dynamic activity suggestion skill that provides instant ideas for hobbies, social gatherings, and solo tasks based on category or group size.
UX Architect
A technical architecture skill for building scalable CSS systems and responsive layout frameworks.
Pipeworx Breweries
A dedicated AI skill for accessing the Open Brewery DB to find and retrieve detailed information on craft breweries across the United States.
PhysiXpert: Elite Physics Master AI
A high-performance modular framework that transforms AI agents into elite physics experts for theoretical analysis and experimental design.
Bible Verse & Scripture API
A high-performance skill for fetching Bible verses, passages, and random scriptures from multiple translations via bible-api.com.
Featured*
