I have created a Custom GPT that connects to an API (that I also wrote). The OpenAI schema I created is very explicit in all the parameters that are sent and the JSON data that is returned, so that my GPT knows exactly what it needs to send and what it gets back.
I have also made my instructions very specific, and despite there being no enforced structure, I have a pretty good system set up for params and return values.
But my question is, should I be including what are essentially duplicate API params and other details in the instructions when they are already clearly spelled out in the schema? Is the duplication overkill or does it help guide the GPT?
My GPT/API works pretty flawlessly right now. I will test without the instructions, but mainly I’m looking forward to new more complicated projects and really don’t want to dream up new “fake structure” in the instructions when it’s clearly laid out already in the schema.