agent-browser
Installation
Summary
Deterministic browser automation for AI agents with snapshot-based element references and multi-session support.
- Interact with web pages using stable element refs (@e1, @e2, etc.) generated from snapshots, enabling reliable automation across DOM changes
- Core commands cover navigation, form filling, clicking, waiting, screenshots, PDFs, and visual regression testing via baseline comparison
- Supports parallel isolated sessions, network-aware waits (networkidle), and selector-based targeting for dynamic single-page applications
- Optional hardening includes domain allowlists, action policies, and content boundaries to reduce injection risk in sensitive workflows
SKILL.md
agent-browser - Browser Automation for AI Agents
When to use this skill
- Open websites and automate UI actions
- Fill forms, click controls, and verify outcomes
- Capture screenshots/PDFs or extract content
- Run deterministic web checks with accessibility refs
- Execute parallel browser tasks via isolated sessions
Core workflow
Always use the deterministic ref loop:
agent-browser open <url>agent-browser snapshot -i- interact with refs (
@e1,@e2, ...) agent-browser snapshot -iagain after page/DOM changes