Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• Pleas do not modify this template :) and fill in all the required fields.

Provide a description of the help you need

How to reproduce:

• Create an agent type app and keep talking to it in a conversation until it output an incomplete answer.

Debuging:

• Agent type app use assistant_base_runner to run. According to Refactor agent history organization and initialization of agent scrat… #2495, it will take all the history messages as prompt messages. Therefore, prompt messages will become larger and larger when keep talking in a same conversation.
• When the prompt_tokens is greater than model_context_tokens, max_tokens will be fixed to 16. It limits the llm model output and result in an incomplete answer.
  

  dify/api/core/app_runner/app_runner.py

  Lines 114 to 115 in 16af509

  	if prompt_tokens + max_tokens > model_context_tokens:
  	max_tokens = max(model_context_tokens - prompt_tokens, 16)

Any ideas to solve this problem?