Skill Decomposition

In a cooperative multi-agent learning environment if a new agent enters the system the others can teach the new agent the high level task which it can then decompose to learn other useful skills.  In the single agent case an example would be teaching a robot to get from a to b which requires going through a door and the robot could learn the simpler activity of walking through a door by exploring its options space.  I think maybe task decomposition type algorithms could be augmented to decompose the option space.