Agent Browser

•
3
•

Automates headless browser workflows using accessibility-tree-based element selection optimized for AI agents.

Agent Browser is a high-performance browser automation skill designed specifically for AI agents to interact with complex web applications and Single Page Applications (SPAs). By utilizing accessibility tree snapshots and reference-based element selection (e.g., @e1, @e2), it provides a deterministic and efficient way to navigate the web, handle multi-step workflows, and manage isolated browser sessions. This skill is ideal for tasks where performance is critical and deterministic interaction is required over visual screenshot-based analysis, allowing for faster execution and more reliable automation in development and data collection environments.

主要功能

01Deterministic element selection via accessibility tree snapshots and unique refs
02State persistence to save and load cookies/storage, skipping repetitive login flows
03Headless automation optimized for speed and reduced token usage in AI workflows
04Multi-session support for isolated browser contexts and simultaneous testing
053 GitHub stars
06Advanced network control for blocking ads or mocking API responses

使用场景

01Performing high-performance web scraping and data extraction without visual overhead
02Automating complex, multi-step workflows in Single Page Applications (SPAs)
03Running simultaneous cross-user testing using isolated browser sessions