So it makes sense to first send stable prompt, reasoning and files content, tool calls summary and actual tool calls at the very end?