zisshh/computer-use ? reverse-engineered prompt

Build me a Mac voice assistant called Alfred that I can talk to hands free.

It should listen through the microphone, turn my speech into text locally, then send the meaning of what I said to a model that picks the right action and arguments. I want it to open apps, switch windows, type text, search the web or YouTube, scroll, use keyboard shortcuts, control volume and media playback, and handle a few common system actions like screenshots or locking the screen. It should work with a wake word like Alfred, but also support push to talk and a hold to talk mode.

Please make it feel fast and practical, with a small floating on screen status so I can see what it heard and what it is doing. It should run on macOS, ask for the right permissions when needed, and keep most of the work local except for the model decision step. If you need to check current docs for any Mac APIs or speech tools, go ahead and look them up.