ghost-in-the-droid/android-agent ? reverse-engineered prompt

Build me an open source app that lets an AI control a real Android phone and an iPhone like a person would, by seeing the screen, tapping, swiping, typing, opening apps, and handling simple tasks end to end.

I want one system where I can connect a phone over USB or wireless, watch the live screen in a browser, and send jobs to one device or a bunch of devices at once. It should support reusable “skills” for common app workflows, plus a way to browse and install shared skills. I’d also like a simple dashboard and API so I can start tasks from the web, and the option to plug in either a local model or a cloud model depending on what I have available. If you need to look up current docs online while building it, go ahead.