[Question] Clarification on the intended usage of System/User/Assistant roles in Advanced Prompt Mode

[Zedthm]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

Content

Question

I'm looking for official clarification on the intended design and correct usage of the three message roles (System, User, Assistant) in the Advanced Prompt editing mode.

Background

After analyzing the codebase, I've observed that:

1. In base_app_runner.py, the AssistantPromptMessage is exclusively used to wrap the LLM's generated output
2. In simple_prompt_transform.py, there's no handling for Assistant role messages at all
3. In advanced_prompt_transform.py, while Assistant role messages are processed, they seem to be treated as historical conversation turns

What I'd Like to Confirm

1. Is it intended that users should NOT write prompts, instructions, or guidelines in the Assistant role?
2. Is the Assistant role specifically designed to store only AI-generated responses, not user-provided instructions?
3. What is the correct way to provide system instructions, guidelines, or few-shot examples in Advanced mode?

Additional Context

I've noticed that when users write prompts in the Assistant role, it often leads to:

• The LLM responding to the prompt itself rather than the user's actual query
• Role confusion where the AI mixes up questioner and responder identities
• Broken conversation history structure when memory is enabled

I want to document the correct best practices for our team and would appreciate official guidance.

Thank you for your time and clarification!