(not OP but..) I personally am not very into "prompting", you just need to figure out how these models work

it's best when you ask a well known problem/thing they can reference (vs. a niche way to solve exactly what you want to solve)

then you work backwards, I e. why is it like this, what is this for, what are the alternative ways to accomplish this etc...

it's a big query engine after all.

don't try to ask like "what is the exact right way" or etc. because it will try to generate that and likely hallucinate if there is no such answer in its training corpus.

instead ask what the model does know, or doesn't.