A headless browser automation CLI designed for AI agents, providing deterministic element selection via accessibility tree snapshots with refs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install super-agent-browser
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 super-agent-browser 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 powerful headless browser automation CLI tailored explicitly for AI agents, offering an optimized alternative to traditional web interaction tools. By shifting focus from visual layouts to real-time accessibility tree snapshots and reference-based (ref-based) element selection, it ensures precise, deterministic, and highly reliable web interactions.
As one of the most robust Openclaw Skills, this CLI eliminates visual dependencies, enabling seamless navigation, form-filling, data extraction, and multi-session orchestration. It is particularly valuable for autonomous agents that need to operate in complex single-page applications (SPAs) where traditional selector strategies frequently fail.
Agent Browser Use Cases
- Automating multi-step web workflows and forms without relying on fragile CSS/XPath selectors.
- Gathering structural and textual data from complex SPAs using real-time JSON snapshots.
- Running parallel, isolated browsing sessions (e.g., admin and user) to test or execute simultaneous agent tasks.
- Skipping tedious authentication flows by saving and restoring browser states and session cookies.
- Restricting network calls to block advertisements or mock API payloads during automated tasks.
How Agent Browser Works
- Navigate to Target URL: The agent issues an
opencommand to launch and navigate the headless browser to the desired webpage. - Generate Interactive Snapshot: The skill captures an accessibility tree snapshot (using the recommended
-i --jsonoptions) to map out interactive elements with deterministic refs (like@e1,@e2). - Interact via Refs: The agent reads the JSON metadata, identifies the target reference, and executes direct actions such as click, fill, type, or hover using the assigned ref.
- Handle Dynamic Changes: After page updates, the agent takes a new snapshot to refresh the interactive refs and continue the multi-step flow sequentially.
Agent Browser Setup
Install the CLI globally and download the necessary browser binaries using the following commands:
npm install -g agent-browser
agent-browser install # Download Chromium
agent-browser install --with-deps # Linux: + system deps
Agent Browser Data Schema & Taxonomy
Snapshot Output Schema
When capturing page states using agent-browser snapshot -i --json, the CLI generates a structured JSON payload containing interactive references and their roles:
| Property | Type | Description |
|---|---|---|
success |
Boolean | Indicates if the snapshot operation succeeded. |
data.snapshot |
String | The accessibility tree snapshot payload. |
data.refs |
Object | Map of unique identifiers (refs) to their roles and names. |
Example of returned reference schema:
{
"success": true,
"data": {
"snapshot": "...",
"refs": {
"e1": {"role": "heading", "name": "Example Domain"},
"e2": {"role": "button", "name": "Submit"},
"e3": {"role": "textbox", "name": "Email"}
}
}
}
Agent Browser Advanced Features
- Multi-Session Isolation: Isolate environments seamlessly via
--session <session_name>to run simultaneous, independent contexts for parallel agents. - State Persistence: Skip active login flows entirely with state serialization (
state save/state load) to save and load browser storage and cookies. - Network Routing and Interception: Optimize performance or mock test environments by aborting requests (e.g., ad blocking) or mocking API responses directly via
network routeCLI flags. - Custom Tab and Frame Control: Seamlessly manage nested layouts, navigate multiple tabs, and shift focus directly inside iframes using target element refs.
SKILL.md
Loading
Related Openclaw Skills
Self-Improvement Skill for Coding Agents
An autonomous learning skill that allows AI coding agents to log, analyze, and resolve their errors while promoting key insights into long-term project memory.
Polymarket
A python-powered CLI skill providing real-time data tracking, momentum scanning, and local paper trading simulations for Polymarket prediction markets.
Group Cruise Travel Planner
An MCP skill designed to formulate customized, multi-criteria group and corporate cruise travel itineraries automatically.
Luxury Cruise Experience Recommendation (craftwave-skill-7)
An advanced MCP skill designed to recommend premium, luxury cruise experiences to high-end travelers using multi-dimensional product filtering.
Dev Skill
An AI agent skill that automatically transforms Product Requirements Documents (PRDs) into production-ready, fully architected SwiftUI iOS applications.
AI Dev Runtime
A powerful AI-driven development runtime that automates planning, file operations, terminal commands, and bug-fixing with memory-enhanced hybrid search.
Featured*
