Direct Action Surface

Phase 4 closes the main direct-action parity gap on top of the live runtime without changing the existing sikuli-go architecture.

Scope

The live API now represents the SikuliX-style direct action vocabulary across:

Supported actions:

The protocol only adds primitives where stateful or platform-sensitive behavior needs durable server-side semantics.

Primitive RPCs:

Composed helpers in pkg/sikuli:

This keeps the runtime protocol small while still exposing the full direct-action vocabulary at the public API layer.

Match direct actions now route through the live runtime instead of the local InputController.

That change is important for parity:

a live Match keeps using its resolved match target point
actions execute on the same remote/API-backed desktop context as the screen search that produced the match
image-only matches still return ErrRuntimeUnavailable for live-only actions

No backend architecture was changed.

Current platform notes:

macOS:
- mouseDown/mouseUp are implemented via cliclick
- keyDown/keyUp currently support modifier-key holds
- wheel scrolling remains backend-unsupported
Linux:
- primitives map to xdotool
- wheel scrolling maps to button-repeat scroll events
Windows:
- primitives map to PowerShell/user32-based input helpers
- wheel scrolling is supported

The SikuliX direct-action vocabulary is now represented at the API level without changing the current separation between: