s06_context_compact--about return #131
Description
line 118-122
return [
{"role": "user", "content": f"[Conversation compressed. Transcript: {transcript_path}]\n\n{summary}"},
{"role": "assistant", "content": "Understood. I have the context from the summary. Continuing."},
]
agentloop run auto_compact return "user:" and "assistant:" and reset messages list, But doesn't this list structure actually cause an error when calling the LLM-API? Does this violate the principle of alternating turns between user and assistant?