Agent Browser Local for Openclaw

A high-performance headless browser automation CLI designed specifically for AI agents to interact with web pages using deterministic accessibility tree snapshots.

alexbingquanxu-cpu
v1.0.0
Mar 22, 2026
0
752
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install agent-browser-custom

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install agent-browser-custom 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 Local?

Agent Browser Local is a specialized automation tool that bridges the gap between AI reasoning and web interaction. Unlike traditional scrapers, it leverages accessibility tree snapshots with unique reference identifiers (refs), allowing AI agents to perform deterministic element selection. This ensures that when an agent decides to click a button, it targets the exact functional element identified in its previous internal state.

Built for speed and reliability, this tool is a core component for developers building sophisticated web agents within the Openclaw Skills ecosystem. It excels at handling complex Single Page Applications (SPAs) and multi-step workflows where session isolation and performance are critical requirements.

Agent Browser Local Use Cases

  • Automating multi-step web workflows that require high precision.
  • Building AI agents that need to interact with complex JavaScript-heavy SPAs.
  • Running parallel browser sessions (e.g., Admin vs. User) for simultaneous testing or data processing.
  • Extracting structured data from web pages using deterministic element references.
  • Managing authenticated states to bypass repetitive login flows in automated tasks.

How Agent Browser Local Works

  1. The AI agent initializes a session and navigates to a target URL using the open command.
  2. A snapshot is generated with the -i --json flags to capture interactive elements and their accessibility metadata.
  3. The agent parses the returned JSON to map visual or functional elements to specific refs (e.g., @e1, @e2).
  4. Interaction commands like click, fill, or hover are executed targeting these specific refs.
  5. The agent re-snapshots the page after interactions to observe state changes and plan the next move.

Agent Browser Local Setup

To integrate this capability into your Openclaw Skills environment, install the package and its dependencies via npm:

npm install -g agent-browser
# Download the necessary Chromium binaries
agent-browser install
# On Linux, install additional system dependencies if needed
agent-browser install --with-deps

Agent Browser Local Data Schema & Taxonomy

Agent Browser organizes interaction data through structured JSON snapshots. This allows Openclaw Skills to maintain a clear map of the DOM.

Key Description
success Boolean indicating if the command executed successfully.
data.snapshot The raw text representation of the accessibility tree.
data.refs A dictionary mapping IDs (e.g., @e1) to element roles and names.
auth.json A persistable state file containing cookies and localStorage for session continuity.

Agent Browser Local Advanced Features

  • Isolated Session Management: Run multiple independent browser contexts simultaneously using the --session flag.
  • State Persistence: Save and load authentication cookies and storage to skip login walls.
  • Network Interception: Route, abort, or mock network requests to control the environment or block ads.
  • Detailed Inspections: Access deep element attributes, computed styles, and visibility states through the CLI.
  • Frame & Tab Navigation: Seamlessly switch between iframes and browser tabs within a single agent workflow.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*