Multi-agent reply is incorrect, because the sub-agent’s final response set the 'user' role. #3902
Description
The sub-agent's final response is assigned the 'user' role, which causes the main agent in the next conversation turn to mistakenly interpret it as either genuine user input or the result of a tool query. As a result, the main agent unnecessarily summarizes this content again in the new turn. However, according to standard conversational flow, interactions should strictly alternate between user queries and model responses. The current approach deviates from this established interaction pattern.
