Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
**Big update: Expanded documentation and new contribution guidelines.** - Added a CONTRIBUTING.md file with guidelines for contributors. - Significantly revised and expanded SKILL.md: - Clearer structure, concise “Quick start,” and streamlined “Core workflow.” - Added or reorganized sections: video recording, mouse control, semantic locators, browser settings, cookies/storage, network, tabs, frames, dialogs, JS, and session state management. - More comprehensive examples and improved command explanations. - Introduced allowed-tools field in SKILL.md for clearer operational context.
What This Skill Does
A headless browser automation CLI that lets AI agents control web pages through structured commands. It navigates URLs, clicks elements, fills forms, and extracts data using ref-based targets from accessibility snapshots. Built in Rust with a Node.js fallback.
Accessibility-tree refs from snapshots replace CSS selectors or XPath, making element targeting more resilient to layout and class changes.
When to use it
- Automating login flows with reusable saved session state
- Filling and submitting multi-step web forms programmatically
- Extracting text or attributes from dynamic page content
- Recording video demos of web app workflows
- Running parallel browser sessions for multi-site testing
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Submit the contact form at https://example.com/form with my email and message
- 1Run `agent-browser open https://example.com/form` to navigate to the page
- 2Run `agent-browser snapshot -i` to get interactive element refs
- 3Run `agent-browser fill @e1 "user@example.com"` to fill the email field
- 4Run `agent-browser fill @e2 "Hello, I'd like more info"` to fill the message field
- 5Run `agent-browser click @e3` to submit, then `agent-browser wait --load networkidle` to confirm
Form submitted; agent snapshots the result page to confirm the success message