System message should be passed in subsequent inference if reflect_on_tool_use is True

[jercoh]

Hi,

Based on my understanding of the AssistantAgent and my observations in the logs, the _reflect_on_tool_use_flow does not include the system_message to the subsequent inference or "reflexion" on the tool results.

I think it should be set, as the Agent role doesn't change when reflecting on tool use. Also the system message can help on formatting the tool's results or summarizing the tool's result.

Please let me know your thoughts on this.

Thanks

autogen/python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py

Line 1040 in 4d8b97e

async def _reflect_on_tool_use_flow(

AutoGen version

0.4.8.2

[ekzhu]

It's likely a bug introduced in 0.4.8. We can submit a fix