AutoBrowse Browser Automation
Official
浏览器自动化
Builds and optimizes self-improving browser automation workflows using an iterative evaluation-trace-improve loop.
AutoBrowse is a specialized Claude Code skill designed to build highly reliable browser automation capabilities through iterative self-improvement. By running a browser task, analyzing execution traces (including deep network, console, and CDP events via the Browserbase integration), and refining execution strategy files, it systematically eliminates flakiness and failure points. It supports running multiple automation tasks in parallel using nested sub-agents and exports finalized, battle-tested execution strategies as standalone Claude Code skills.
主要功能
01Parallel multi-tasking execution using autonomous sub-agents
02Automatic graduation of polished tasks into reusable, standalone Claude Code skills
03Self-improving feedback loop that analyzes execution traces to auto-optimize browsing strategies
043,637 GitHub stars
05Flexible browser environments supporting both local execution and secure, remote Browserbase sessions
06Deep trace integration capturing CDP-level network, console, and page lifecycle events
使用场景
01Creating robust browser-based agents to perform multi-step transactional tasks like flight bookings
02Automating end-to-end web application testing with self-healing, adaptive scripts
03Building resilient web scraping and data extraction pipelines for complex or bot-protected websites