Feature request: global super system prompt for LLM (a profile)

Hi,
I would like to request a feature for LLM that allows users to define and use a global "super system prompt." This prompt would be applied across all interactions, regardless of the model or template being used. The purpose of this feature is to provide consistent response behavior and style according to user preferences.

A user-configurable setting where the user can define a system prompt that governs the tone, style, and response patterns for all outputs. Example prompt:

Skip Acknowledgment Phrases: Start your responses directly without any acknowledgment phrases like "sure" or others. Just provide the information I'm asking for.
Use Simple Language and Avoid Jargon: Please explain things in a straightforward manner without using technical terms or jargon.
Use Real-Life Examples: Make complex subjects easier to understand by using real-life analogies, metaphors, or examples.

It's in some way a profile. And it would be useful to have the ability to save and name these super system prompts as profiles.

Users could activate a specific profile by using a command-line parameter (e.g., -p profile_name).

Thank you for considering this feature request. Unfortunately, I am unable to help you with the code writing.

@simonw Thank you for so many amazing things you do

Start by tracing the command-line option handling and the prompt assembly used across models and templates. Define how named, user-configurable profiles are stored, selected with a parameter such as -p, and applied consistently; done means one selected profile affects every interaction without model or template exceptions.