How to include reasoning in conversation history?

[dstoc]

I'm using unsloth/gpt-oss-20b-GGUF with llama-server version: 6199 (f08c4c0)

I can access reasoning_content in the output of the v1/chat streaming API. However if I feed it back in using the same format, it does not make it into the prompt.

e.g. for request:

"messages":[
{"role":"user","content":"hello!"},
{"role":"assistant",
    "content":"Hello! 👋 How can I help you today?",
    "reasoning_content":"The user says \"hello!\" Probably want a greeting. No request to use tools. Just reply."},
{"role":"user","content":"test"}
]

the reasoning_content is missing in the relevant part of the prompt:

...
<|start|>user<|message|>hello!<|end|>
<|start|>assistant<|channel|>analysis<|message|>Hello! 👋 How can I help you today?<|end|>
<|start|>user<|message|>test<|end|>
<|start|>assistant

[aldehir]

The gpt-oss template only inserts reasoning for the last assistant message with a tool call. It does not insert reasoning for messages with content only.