Due to token restrictions on model output, the agent is unable to cyclically output tokens

[bllp12]

I have a requirement for the agent to call the tool to obtain the source code, which is about 6000 lines long. There is no problem with the model context and it will not exceed. However, when it comes to the final output, due to the 16K limit, it is not possible to output all the code. How can we solve this problem?