In this work, we investigate ways an agent can combine existing skills to create novel ones in a manner that is both principled and optimal. We find that by constraining the reward function and tra...