An option to disable automatic system instruction wrapping

I noticed TypingMind wrapping AI Agents' system instruction inside <ASSISTANT_PROFILE name=>
<INSTRUCTION></INSTRUCTION>
</ASSISTANT_PROFILE>, but if the Agents (AI Characters) aren't actually assistant-like characters, this wrapping might confuse the model, therefore let it produce lower quality output (e.g. in creative writing or roleplay). During my testing, mistral-medium-2508 produced significantly lower quality output when entering the same prompt (a non-assistant fictional character) in Agent compares to just entering the prompt in plain system prompt (in the model global setting page).

Maybe you could give users an option to let them choose to disable the "<ASSISTANT_PROFILE name=>
<INSTRUCTION></INSTRUCTION>
</ASSISTANT_PROFILE>" wrapping, and if this option is disable, you could just send the Agents' instruction as a plain system prompt. I know we can enter the prompt in model setting page, but that is a global setting and very inconvenient if user has lots of agents.