There’s a bit of a conversation going on about node gizmos for D9969, that I think is worth expanding on here.
As a general pattern, I think that generic gizmos for inputs would be both easier to work with and easier to maintain than unique gizmos for operations. A boolean on a float/vector/transform node that toggles a viewport gizmo for it, that you could hook up to group node inputs, would be about as generic as you could get without requiring an overly complex set of options for configuring the gizmo.
An alternative to that might be a dedicated gizmo-node that inputs are routed through. It’d open the door to more complex configurations (routing multiple inputs through one gizmo, etc) and it’d shift where the gizmo might sit in a chain of nodes.
Either option could prove to be fairly powerful, really, but we do need an approach that’ll minimize the amount of single-node gizmos that can’t be generalized beyond the specific node they’re designed for.