After Prokka installation, go to the db/hmm directory and either: 1) delete or temporarily move all databases except COGs; 2) rename the COG database so that its name is first alphanumerically (say, 1-COG). That way all your genes will have only COG annotation (case 1), or they will be first annotated according to COG database so that all possible COG annotations are included (case 2).

A somewhat relevant topic has been discussed recently.