Keyboard & mouse

windowctl type "hello world"
windowctl key cmd+shift+t
windowctl click --x 400 --y 300 --button left --count 2
windowctl move-mouse --x 960 --y 540
windowctl scroll --dy -120
windowctl drag --x1 100 --y1 100 --x2 600 --y2 400

These are real OS-level events (CGEvent on macOS), delivered to whatever has focus — not messages posted to a specific window. Combine with focus when you need to be sure where they land:

windowctl focus --app "Notes" && windowctl type "meeting notes"

The visual loop

Screenshot, decide, act — the loop an agent uses to drive a UI it has no API for:

windowctl screenshot --out /tmp/s.png     # look
windowctl click --x 812 --y 344           # act
windowctl screenshot --out /tmp/s2.png    # confirm

As a stream

For continuous control, windowctl start input --port N --token T accepts input events over a WebSocket instead of one process per action — which is what you want when the events are coming from a test harness at speed.

Permission

Input injection needs macOS Accessibility, same as moving windows. Availability is reported honestly: windowctl devices tells you whether the input surface is usable before you script against it.