When an Agent re-executes due to a retry mechanism or interrupt recovery, tool calls with side effects are triggered twice—a real pain point encountered in production environments by users of mainstream Agent frameworks such as LangGraph and CrewAI.