Really just promoting this question to an "idea".
External attributes which are exposed on a custom transformer input port get applied to all input ports in that custom transformer. It would be nice if the exposed attributes could be tied to specific input ports.
As was noted in the original question, if different features are coming into different input ports, then they will not all necessarily contain all of the attributes being exposed and this can lead to confusion within the custom transformer as to what is really available on a feature.
Similarly for output ports. If features exiting a specific output port only contain certain attributes, then invalid attributes should not be made available to the caller.