Dispatch keyboard input to a host app window’s focused control.
Accepts {window_id?, action} where action is a tagged object
such as {kind:"type", text} or {kind:"press", key, modifiers?}.
Capture a PNG of the host app’s window. Accepts an optional
window_id (returned by WINDOWS) so multi-window desktop
apps can pick a specific surface; mobile platforms ignore it
since they have a single foreground window. Returns the unified
screenshot envelope {target, kind, coordinate_space, format, width, height, size_bytes, image:{mime, encoding, data}}.