Tailored Primitive Initialization is the Secret Key to Reinforcement Learning