Don't do it: Safer Reinforcement Learning With Rule-based Guidance