fixed: when a .check reaching max attempt count interrupts LLM conversation by throwing an error, that error does not trigger the retry mechanism for individual OpenAI calls #promptgun