Recently reading Google's guideline on how to write prompt for LLM since recently I'm trying to connect geimini to meshtastic channels, and due to limitations, I want specific behavior which is different from what you usually get from a LLM, and the prompt I wrote is not working well.

I also found the prompt leaks from different LLM services. From what I read, I think GPT 5 might learn well on how to fudge humans to void answering questions. Since the first paragraph of their system prompt is to tell the model does not avoid answering questions. LOL

----
Ref: https://www.kaggle.com/whitepaper-prompt-engineering
Ref: https://github.com/asgeirtj/system_prompts_leaks/blob/main/OpenAI/gpt-5-thinking.md