This is one of those things that won't ever be "solved" as people just want different things here, hence we can steer the models with the system prompt.

I'm mostly the same as you, I don't want the model to assume things, or act on implicit "directions". But then also, sometimes I do, and I myself might not always know when what approach is best.