Isolate trusted agent input from human focus on multi-monitor

Feature request: isolate agent input from the human on a multi-monitor session

Computer-use agents (Hermes, Claude Code, Codex, and similar) share Windows' single mouse and keyboard focus. GlazeWM already owns monitors, workspaces, and window focus, but an agent that clicks or types still steals the human cursor and keyboard. You cannot keep working on one monitor while a trusted agent operates another.

GlazeWM cannot invent a second Win32 hardware cursor if the OS does not provide one. The request is for WM-level isolation as far as the platform allows.

Please consider:

1. IPC so an orchestrator can target windows on an assigned workspace/monitor without stealing the human's current keyboard focus or warping the human mouse.
2. A documented "agent workspace" vs "human workspace" model on multi-monitor.
3. If Windows allows a second pointing device / overlay cursor, expose it.
4. A kill switch that cancels agent-driven focus changes and restores human focus.

Not requesting a second Windows session, RDP, or a nested desktop.

I searched issues for multiple cursor / virtual pointer / agent workspace and did not find this. CONTRIBUTING says to open an issue for feature requests; there is no feature-request template.