Multi-Monitor / Virtual-Desktop Geometry

snap_window, arrange_grid and the layout planner all take a single primary (width, height) — they are monitor-blind: they cannot tile on the second display or cope with a negative-origin virtual desktop, and coordinate_space only rescales a model grid. This adds the missing physical layer: enumerate the monitors, compute the union virtual bounds, ask which monitor contains a point or a window, convert between virtual and per-monitor-local coordinates, and remap a point to the equivalent spot on another display.

The geometry is pure arithmetic over plain Monitor dataclasses, so it is fully unit-testable; only enumerate_monitors’ default provider touches the OS (via mss) and it is injectable. Imports no PySide6.

Headless API

from je_auto_control import (enumerate_monitors, monitor_at_point,
                             virtual_bounds, to_local, remap_point)

monitors = enumerate_monitors()
print(virtual_bounds(monitors))            # (x, y, w, h) spanning all displays

here = monitor_at_point(monitors, x, y)    # which monitor owns this point
idx, lx, ly = to_local(monitors, x, y)     # virtual -> (monitor, local x, local y)

# Move a point to the equivalent relative spot on another monitor.
second = remap_point(monitors[0], monitors[1], 960, 540)

Monitor carries index, x, y, width, height, scale, primary and a work area (.bounds / .contains(x, y) / .to_dict()). virtual_bounds returns the union box (origin may be negative); primary_monitor picks the primary; monitor_for_window(rect, monitors) returns the display a window mostly occupies (max overlap); to_virtual is the inverse of to_local; remap_point preserves the fractional position so it works across differing resolutions and DPI.

Executor commands

AC_enumerate_monitors → {count, monitors, virtual_bounds} and AC_monitor_at_point (x / y) → {found, monitor}. They are exposed as the MCP tools ac_enumerate_monitors / ac_monitor_at_point and as Script Builder commands under Window.