A high-performance browser automation tool leveraging accessibility tree snapshots and deterministic element references for AI-driven workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install projectamazonph-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 projectamazonph-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 Skill?
The Agent Browser Skill is an advanced browser automation integration designed to supercharge AI coding agents with high-speed, reliable, and deterministic web interaction capabilities. Unlike traditional selector-heavy automation frameworks, this skill utilizes accessibility tree snapshots with unique reference tags (refs) to interact with complex web elements directly. This deterministic selection model ensures your automation scripts and AI agents remain resilient even when underlying page layout structures change.
As one of the core Openclaw Skills, it acts as a superior alternative to heavy visual browser engines. It offers massive performance gains for multi-step workflows, single-page applications (SPAs), and headless execution pipelines where visual inspection isn't required.
Agent Browser Skill Use Cases
- Automating multi-step web workflows with high reliability and performance.
- Interacting with dynamic Single Page Applications (SPAs) where traditional DOM selectors frequently change.
- Managing isolated multi-user browser testing sessions simultaneously on a single system.
- Skipping repetitive login and verification flows by preserving and restoring session states.
- Mocking API network requests and blocking redundant heavy assets or ads during automated runs.
How Agent Browser Skill Works
- Navigate & Initialize: The agent opens a target URL inside an isolated or global Chromium browser session using the CLI.
- Generate Accessibility Snapshot: It requests an interactive snapshot, generating a clean JSON map of the page's accessibility tree elements assigned with references (like @e1, @e2).
- Deterministic Interaction: The agent issues interaction commands (click, fill, hover) targeting those exact ref keys without needing fragile CSS selectors or XPaths.
- Re-evaluate State: After the page updates, the agent requests a fresh snapshot to sync the reference tree and repeat the interaction loop.
Agent Browser Skill Setup
To integrate this capability into your workflow, ensure Node.js 24+ is configured on your machine. This tool relies on a globally installed agent-browser package and Playwright Chromium.
# Install the agent-browser CLI globally via npm
npm install -g @vercel-labs/agent-browser
Next, configure Openclaw to use the correct browser executable path by editing your configurations in your home directory:
{
"browser.executablePath": "/root/.cache/ms-playwright/chromium-1223/chrome-linux/chrome"
}
This path should point to your local Chromium binary, allowing your Openclaw Skills agent to control the browser natively.
Agent Browser Skill Data Schema & Taxonomy
The skill returns clean, parseable JSON payloads when snapshots or queries are executed, making parsing trivial for AI pipelines.
Snapshot JSON Schema
When capturing active pages using the -i --json flags, the output structure maps as follows:
| Key | Type | Description |
|---|---|---|
success |
boolean | Indicates if the snapshot operation was successful |
data.snapshot |
string | Text representation of the rendered page layout |
data.refs |
object | Dictionary mapping unique IDs (e1, e2) to accessibility roles and element names |
State Storage Schema
When preserving authentication state, the skill exports a clean JSON configuration containing session keys, cookies, and local storage values, which can be easily re-injected to bypass login screens.
Agent Browser Skill Advanced Features
- State Persistence: Instantly save and load cookies or local storage schemas (
state save/load) to completely skip login gates. - Network Interception & Routing: Block media trackers or mock JSON endpoints entirely inside your flow (
network route --abortor--body). - Isolated Sessions: Run distinct profiles side-by-side (e.g., admin vs. standard user) without cross-contamination using the
--sessionflag. - Stability Control: Wait for specific states before executing actions (
wait --load networkidleor--text). - Tab & Frame Handling: Seamlessly switch between multiple browser tabs and nested iframes using standard CLI keys.
SKILL.md
Loading
Related Openclaw Skills
Gold Fundamental Analysis
An automated AI agent skill that aggregates and analyzes macroeconomic indicators, COT positioning, and ETF flows to score and forecast gold market fundamentals.
Financial Astrology AI Agent Skill
An advanced AI coding agent skill that analyzes market cycles, crowd psychology, and technical planetary alignments for automated day trading and financial research.
Gumroad Admin
An advanced automation skill designed to administer Gumroad e-commerce stores, streamline batch digital product publishing, and safely manage files using the official CLI.
Auto Bookkeeping Assistant
A smart AI-driven bookkeeping assistant that parses natural language inputs to automatically organize, log, and visualize your daily expenses locally.
A Tree Grows in Brooklyn Literature & Theme Analyzer
An interactive AI agent skill designed to analyze the themes, characters, and symbolic depth of Betty Smith's classic novel, A Tree Grows in Brooklyn.
Taiwan Fund
An advanced financial tool designed to query and analyze Taiwan domestic and offshore mutual funds with auto-currency conversion and benchmark comparisons.
Featured*
