ai-agents-vllm-bug-workaround--multi-turn-chat-with-structured-outputs-json-schem-5b577dd5

THEN Modify the chat history by changing the role of any existing assistant message to 'user' and prefix its content with '[ASSISTANT message]:\n\n'. This prevents the bug while preserving the conversation context.

messages = [
    {"role": "system", "content": "You are a helpful assistant!"},
    {"role": "user", "content": "What is 2 + 5?"},
    {"role": "user", "content": "[ASSISTANT message]:\n\n6, of course!"},
    {"role": "user", "content": "No, try again. Respond with {'response': '<your answer>'}"}
]