AutoBrowse

0•

Builds and refines reliable browser automation workflows through iterative execution, trace analysis, and automated skill graduation.

AutoBrowse is an autonomous browser skill builder that implements an iterative auto-research loop to create dependable browser automation routines. Operating as an outer agent that monitors an inner browsing agent, AutoBrowse inspects execution traces, failure screenshots, and decision logs to formulate targeted heuristics and update strategy instructions until tasks pass reliably. It supports both local browser execution and remote cloud browsers for bot-protected websites, parallel multi-task processing via sub-agents, and automatic graduation of proven strategies into standalone, production-ready Claude Code skills.

Key Features

01Automated graduation that compiles validated strategies into self-contained, reusable Claude Code skills
02Iterative auto-research loop that optimizes browser strategies based on failure traces and screenshots
03Detailed trace diagnostics capturing decision logs, turn-by-turn costs, and visual failure checkpoints
040 GitHub stars
05Dual-environment support for local browser control and remote anti-bot cloud sessions via Browserbase
06Parallel multi-task execution utilizing Claude sub-agents for concurrent browsing workflows

Use Cases

01Generating and maintaining a portfolio of modular, site-specific web scraping and interaction skills
02Overcoming anti-bot defenses and dynamic UI timing issues through automated heuristic refinement
03Developing resilient browser automation for complex multi-step web flows like booking or checkout forms