Agent Browser

This is a proper browser automation CLI designed for AI agents to interact with websites. It uses a snapshot and reference workflow where you navigate to a page, grab snapshots that assign refs like @e1 and @e2 to interactive elements, then use those refs to click, fill forms, and extract data. The refs invalidate when the page changes, so you snapshot again after navigation. It handles the usual automation tasks like login flows with state persistence, parallel sessions, screenshots, and even iOS simulator testing. The workflow is clean and the ref system makes it easy for an agent to target elements without brittle selectors, though you need to remember to re-snapshot after DOM changes or you'll be working with stale references.