Despite recent advances in reinforcement learning (RL), significant barriers remain when applying it to a broad spectrum of safety-critical domains. Fundamentally, there is a notable absence of a standard definition of safe and robust RL and standardised checklists or guidelines to ensure the safe and robust design and deployment of RL systems. This paper addresses these gaps by compiling existing definitions of safe and robust RL, suggesting a consensus definition for each concept. We also propose a comprehensive checklist designed to enhance the safety and reliability of reinforcement learning applications, while remaining flexible to be tailored to the specific applications at hand. The checklist comes in the form of essential items and actions for practitioners and policy makers to follow, categorised for simplicity. Additionally, this checklist aims to present a starting point for further discussions and collaboration within the community. We hope this position paper serves as a foundation for developing detailed principles for reinforcement learning across safety-critical fields and sequential decision-making.