remorses/usecomputer ? reverse-engineered prompt

Build me a desktop automation tool called usecomputer that AI agents can use to control the screen from the command line.

I want it to work on macOS, Linux with X11, and Windows, and I want the main actions to be taking screenshots, moving and clicking the mouse, dragging, scrolling, typing text, and pressing keyboard shortcuts. It should feel fast and reliable, with a native binary behind it, and also expose a small JavaScript API for people who want to call it from code.

Please include the screenshot to action to screenshot workflow for agents, plus the coordinate mapping so clicks from screenshots land in the right place. Add clear examples and a simple install and usage flow, so someone can try it quickly from the terminal. If you need to look up current docs online, go ahead and do that.