minitap-ai/mobile-use ? reverse-engineered prompt

Build me a local tool that lets an AI control a phone app from a plain English prompt, basically like a person tapping through Android apps and, where possible, iOS simulators too. I want to be able to say something like open Gmail, find the first 3 unread emails, and return the sender and subject as JSON, and have it actually navigate the app screen by screen.

Please make it usable from the command line with a simple prompt plus an optional description of the output format. It should work well with Android phones over USB or an emulator, and if iOS is included, keep it to simulator support on Mac only. Add a straightforward setup using environment variables for API keys, sensible defaults for model config, and an easy Docker based quick start for Android. Include the scripts, docs, and basic tests so someone can clone it, connect a device, and run it without much fuss. If anything is unclear, look up the current docs online and follow the project README style.