Skill-Based Reinforcement Learning with Intrinsic Reward Matching