Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Headless browser automation CLI built for AI agent workflows. Takes accessibility tree snapshots of web pages and exposes interactive elements as stable refs (like @e2, @e3) for deterministic targeting. Supports isolated sessions, saved auth state, and network interception.
Ref-based element selection from accessibility tree snapshots avoids the fragility of CSS selectors and XPath that break when layouts change.
When to use it
- Automating multi-step web form submissions without brittle selectors
- Running simultaneous admin and user browser sessions for testing
- Scraping dynamic single-page apps that require JavaScript execution
- Reusing saved login state to skip authentication on repeat runs
- Mocking API responses during automated browser workflows
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: search Google for 'AI agents' and return the first result link
- 1Run agent-browser open https://www.google.com
- 2Run agent-browser snapshot -i --json to get interactive element refs
- 3Identify search box ref from snapshot, run agent-browser fill @e1 "AI agents" then agent-browser press Enter
- 4Run agent-browser wait --load networkidle, then agent-browser snapshot -i --json again
- 5Run agent-browser get attr @e4 "href" --json to extract the first result URL
Returns the href value of the first search result as JSON
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.