The previous prompt was utterly broken with Anthropic models, they would just try to continue the conversation