Input for computer_act — ONE step of a semantic pipeline: either an
accessibility action on the element matched by selector, a keyboard
action (key/text), or a wait pause — chained via then with
shell && semantics.
Input for computer_control — one step of a desktop-control pipeline:
EITHER a pointer action (click/scroll/move/drag — key/text absent)
OR a keyboard action (key or text), at raw coordinates or on an
accessibility-tree element.
Which OS shell surface to target instead of an application — the parts
of the desktop OUTSIDE any app: menu bar, Dock/taskbar/panel, tray
status items, the desktop itself, transient flyouts (Notification
Center, Quick Settings, shell context menus). Mirrors xa11y’s
xa11y::ShellSurfaceKind exactly (same snake_case wire spelling).