globalbtc1/agent-browser ? reverse-engineered prompt

Build me a fast browser automation CLI for AI agents that feels natural to use from the terminal.

I want to be able to open a browser, go to a site, take a snapshot of the page, click and fill things by either simple selectors or AI friendly references, and then read text, HTML, titles, URLs, and screenshots back out. It should also support basic actions like typing, scrolling, hovering, checking boxes, uploading files, pressing keys, and waiting for elements or text to appear.

Make it work well with Chrome, including local installs and a first time setup command that can download what it needs. I also want a command that can connect to an existing browser session and a simple chat style mode where I can tell it what to do in plain English.

Keep it snappy, reliable, and friendly for automation workflows, and make sure it has good help text and sensible defaults so someone can use it right away from the command line.