A high-performance CLI tool that enables AI agents to navigate, interact with, and extract data from websites using a structured reference system.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-vercel
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-vercel 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 browser automation interface designed to give AI agents a reliable way to interact with the modern web. Unlike traditional automation tools, it provides a unique element-referencing system via accessibility snapshots, allowing agents to identify and manipulate web elements using simple tags like @e1 or @e2. This skill is a core component for developers building Openclaw Skills that require robust web interaction capabilities.
By providing a command-line interface that bridges the gap between LLMs and browser engines like Playwright and Chromium, agent-browser simplifies complex tasks. It handles everything from session persistence and encrypted authentication to visual debugging and mobile emulation. When integrated into Openclaw Skills, it allows agents to perform tasks as a human would, but with the speed and precision of a programmatic environment.
agent-browser Use Cases
- Automating multi-step form submissions and user registrations on dynamic websites.
- Scraping data from interactive web applications that require clicking or scrolling to reveal content.
- Performing end-to-end testing of web features, including visual regression and functional validation.
- Managing authenticated sessions for secure portals using persistent cookies and state storage.
- Capturing annotated screenshots and session recordings for audit logs or visual verification.
How agent-browser Works
- Navigate: Use the open command to direct the browser to a specific URL and establish a connection.
- Snapshot: Generate an interactive snapshot to map the page's accessibility tree to unique reference IDs for the agent to use.
- Interact: Execute actions such as click, fill, or select by targeting the discovered reference IDs.
- Observe: Take follow-up snapshots or use the diff command to verify that the web page state has changed as expected.
- Persist: Save session states or close the browser to ensure data is handled securely and resources are freed.
agent-browser Setup
To begin using this tool within your Openclaw Skills environment, install the package and configure your environment:
# Install the CLI tool
npx agent-browser@latest install
# (Optional) Set up an encryption key for the secure Auth Vault
export AGENT_BROWSER_ENCRYPTION_KEY=$(openssl rand -hex 32)
# Run a simple test command to verify setup
agent-browser open https://example.com && agent-browser snapshot -i
agent-browser Data Schema & Taxonomy
The skill organizes its data and configuration across several layers to ensure isolation and security:
| Data Type | Location / Format | Description |
|---|---|---|
| Sessions | ~/.agent-browser/sessions/ |
JSON files containing cookies, localStorage, and session metadata. |
| Auth Vault | Encrypted SQLite/JSON | Secure storage for credentials associated with specific domains. |
| Snapshots | Text/Markdown | Accessibility tree representations used by agents to identify element refs. |
| Configs | agent-browser.json |
Project-level settings including proxy details and headed mode preferences. |
| Artifacts | User-defined paths | Screenshots (PNG), PDFs, and video recordings (WebM) generated during runs. |
agent-browser Advanced Features
- Auth Vault Integration: Securely store and inject credentials into login flows without exposing passwords to the LLM context.
- Parallel Session Management: Run multiple, isolated browser instances simultaneously using named sessions to scale Openclaw Skills.
- Mobile Emulation: Full support for iOS Simulator and mobile Safari gestures like swiping and tapping.
- Visual Diffing: Compare accessibility trees or pixel-perfect screenshots to detect UI changes or regressions automatically.
- Security Sandboxing: Implement domain allowlists and action policies to restrict the agent's reach to trusted environments.
- Native Mode: An experimental Rust-based daemon for direct CDP communication, offering higher performance and lower overhead.
SKILL.md
Loading
Related Openclaw Skills
Seedream Image Generation Skill
A professional AI image generation skill leveraging the Seedream 4.5 API to create, edit, and fuse high-quality visual content through intelligent prompt optimization.
OpenClaw Toolbox
A comprehensive management suite for environment initialization, system maintenance, and automated multi-mode backups for AI agent workspaces.
Goal Heartbeat Engine
A robust autonomous state machine that ensures AI agents maintain continuous productivity and balanced attention across multiple long-term goals.
Feishu Suite Skill
A comprehensive integration for Feishu that automates documents, spreadsheets, instant messaging, and organizational workflows.
FlowClaw
FlowClaw is a high-availability LLM load balancer and usage monitor designed to keep AI agents running by rotating provider accounts based on real-time credit availability.
AI Smart Customer Service
A sophisticated LLM-powered customer support agent capable of automated FAQ resolution, intent recognition, and multi-turn dialogue management.
Featured*
