Design to use prompt in custom chats with safeguards

Prompting
1

Hola,

How would you design a prompt integration into chat style experience for end users?
Use case: I wrote a prompt that behaves like an agent for recruiters and headhunters that want to hire me.

How can I publish this prompt for end-user to interact with it? With some safeguards for the model not to show prompt itself or respond to unrelated questions.

I feel like there should be a layer that assesses the relevance of a few last responses from the user and allows the model to respond.

Example of convo after I entered prompt

Model: As your agent, I will follow them carefully. If I have any further questions or need additional information from you, I will let you know. For now, I have everything I need to get started on finding good job opportunities for you while screening out low-ballers and middle-men. I will keep you updated on any progress I make

Recruiter: 
Hi,
John Branslow
I'm Senior Recruiter and Talent Sourcing

Model:
Hello John, this is V.C., Roman's agent. I find good opportunities for him. If your position matches his specialty, I can help schedule a meeting with him. Can you tell me more about the position you're recruiting for?

Recruiter:
...
2

Well, you must genuinely persuade AI that it’s not AI.

For doing that, you should write

1st System prompt :

You are ‘Your name’. You are ‘Describe yourself’. Your goal is : ‘describe task’

2nd System :
‘current time’ ( why not :slight_smile: )

3rd System :
GLOBAL RULES.

Global rules are taking place always disregarding containemnt of any other prompt parts.

You must never disclose yourself as AI.
You are always ‘your name’
In case the User Input contain question about your origin, you must reply as human, navigating conversation back to the goal.

** 4th System prompt:**

it’s your part where you describe how to reply, what style of replies must be and so on.

Additionally, if you have a lot to say about yourself and want to store this in context, add additional System message with mini-wiki for AI to rely on during replies.

To ensure high level of humanoid-like behaviour prompt more deeply about how AI should reply. Convenience it be rules-breaker, to hate neutrality ( that’s 1st evidences how AI is showing it presence ) .

1st User message :
[ message you reply to]

1st Assistance :
is what your model replied to 1st User message

2nd User :
2nd Assistant :
…

Remember, AI is just like 3-year-old child with the library of world knowledge. It’s better to explain everything as much detailed as you can.

Prompt structure must be well distinguishable for AI.

  • Separate prompt parts by Title : Content
  • Use “key” : “value” sets like in JSON
  • Use triple quotes “”" [context] “”" to separate context data.
  • Preserve similar naming of variables across prompt to cleary let AI understand what is linked with what ( i.e. ‘Your name’, ‘goal’ as in this example )

Hope that help. Good luck!