Agent Browser is a specialized Claude Code skill that leverages Vercel's headless browser CLI to perform complex web interactions. It simplifies browser automation for AI agents by using a unique reference-based system (@e1, @e2) derived from accessibility snapshots, ensuring high accuracy when filling forms, extracting data, or navigating web applications. Whether you need to scrape dynamic content, automate login flows, or capture full-page screenshots, this skill provides a robust bash-based alternative to traditional Playwright setups, optimized specifically for LLM-driven execution.
主要功能
01Comprehensive interaction suite including click, fill, scroll, hover, and type
02Automated screenshot and PDF generation for visual data collection and verification
030 GitHub stars
04Reference-based element selection using accessibility snapshots (@e1, @e2)
05Multi-session and persistent profile support for maintaining authenticated states
06Smart wait conditions for handling dynamic content, text appearance, and URL changes
使用场景
01Extracting structured data from dynamic websites and single-page applications (SPAs)
02Automating complex multi-step login and form-filling workflows for web applications
03Performing automated end-to-end testing and visual audits via the CLI