A production-ready playbook for executing deterministic, AI-friendly web automation via a high-performance CLI.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-browser-core-temp
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-core-temp 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 Core (Local)?
The agent-browser-core-local skill is a specialized framework designed to bridge the gap between AI agents and web environments. By utilizing a Rust-based CLI with a Node.js fallback, it provides a fast, reliable method for agents to interact with the DOM using structured commands, compact snapshots, and reference-based actions. This skill ensures that AI-driven browsing is both machine-readable and highly efficient.
Built for the Openclaw Skills ecosystem, it prioritizes performance and safety. It allows developers to implement complex web workflows—such as data extraction, form filling, and multi-step navigation—without the overhead of full SDKs. The core philosophy centers on a loop of snapshotting, acting via unique element references, and verifying state changes to maintain high accuracy in automated tasks.
Agent Browser Core (Local) Use Cases
- Automating complex multi-step web workflows for autonomous AI agents.
- Generating compact, machine-readable DOM snapshots to reduce LLM token usage.
- Performing deterministic actions on web elements using stable reference IDs.
- Scraping dynamic web content where standard HTTP requests fail.
- Integrating browser-based tasks into local development CLI tools and scripts.
How Agent Browser Core (Local) Works
- Initialization: The AI agent initializes the browser environment and checks for the required CLI runtime.
- Navigation: The agent navigates to a target URL using specified profile or session strategies for persistence.
- Snapshotting: A compact DOM snapshot is generated, assigning unique refs to interactable elements.
- Command Execution: The agent issues structured commands (click, type, etc.) targeting specific refs to ensure precision.
- State Verification: Post-action snapshots are taken to confirm the DOM has updated as expected.
- Resource Cleanup: Tabs and sessions are closed to release system resources and finalize the automation loop.
Agent Browser Core (Local) Setup
To get started with this skill within the Openclaw Skills framework, ensure you have the agent-browser CLI installed.
# Install the agent-browser CLI
npm install -g agent-browser
# Verify the installation
agent-browser --version
# Run a test snapshot to ensure runtime is configured
agent-browser snapshot https://example.com --json
Agent Browser Core (Local) Data Schema & Taxonomy
The skill organizes its operations through structured JSON outputs and a specific file hierarchy to manage context and security:
| Component | Type | Description |
|---|---|---|
| Snapshots | JSON | Compact representation of the DOM including element refs and accessibility labels. |
| Commands | Object | Structured input including action, target ref, and value parameters. |
| References | Markdown | Documentation files for command mapping, safety rules, and troubleshooting. |
| Logs | Text | Local execution logs for debugging automated interactions. |
Agent Browser Core (Local) Advanced Features
- JSON Machine Parsing: Full support for --json flags to facilitate seamless integration with AI agent scripting.
- Reference-Based Actions: High-precision interaction model using refs instead of fragile CSS selectors.
- Safe Mode Enforcement: Built-in guardrails that restrict high-risk operations like eval or local file access.
- Session Persistence: Strategies for managing authentication and user profiles across multiple automation runs.
- Domain Allowlisting: Security controls to restrict automation to approved web targets and block private network access.
SKILL.md
Loading
Related Openclaw Skills
Twitter User Profile API
A specialized skill for fetching comprehensive Twitter profile metadata, audience statistics, and verification details using a Rest ID.
Media.io Text-to-Video API
A powerful AI skill for generating high-fidelity videos from text descriptions using the Media.io OpenAPI and Vidu Q3 models.
Twitter User Published Posts API
A specialized skill for programmatically retrieving Twitter user posts, engagement metrics, and timestamps using the JustOneAPI service.
Exponential Growth System
An advanced self-evolution methodology that enables AI agents to distill problem-solving experiences into reusable skills and systematic knowledge.
NS Fate
A multi-system metaphysical forecasting tool that transforms traditional divination into a rigorous, data-driven decision support protocol.
Session Logs
A powerful utility for searching and analyzing historical AI conversation transcripts using jq and ripgrep.
Featured*
