Agent Browser
Automates headless browser interactions using accessibility tree snapshots for deterministic AI-driven web navigation.
Agent Browser is a high-performance automation skill designed for AI agents that require precision when interacting with complex web environments. Unlike traditional browser tools that rely on fragile DOM selectors, this skill utilizes accessibility tree snapshots and reference-based element selection to ensure deterministic actions. It excels at multi-step workflows, handling single-page applications (SPAs), and managing isolated sessions, making it an essential tool for developers building robust web-scraping, testing, or automation pipelines within Claude Code.
主要功能
01Interactive JSON output optimized for AI parsing and multi-step reasoning
02Isolated session management for simultaneous multi-user or multi-role workflows
03State persistence for cookies and localStorage to bypass repetitive login sequences
043 GitHub stars
05Ref-based deterministic element selection via accessibility tree snapshots
06Comprehensive network control including request mocking, blocking, and filtering
使用场景
01End-to-end testing of user permissions using concurrent isolated browser sessions
02Automating complex, multi-page data entry and form submission workflows
03Extracting data from dynamic, JavaScript-heavy applications where traditional scraping fails