The recurring move in this pack is forbidding the default behaviour. Left to themselves, assistants rewrite when you wanted an edit, praise when you wanted criticism, and add adjectives when you wanted them removed. Most of these prompts are mostly constraints for that reason.
The one that people underuse
The criticism prompt. It is uncomfortable and it is the one that most reliably improves the draft, because "what would a hostile reader challenge" is a question about your argument rather than your prose.
Adapting the voice prompt
The voice sample has to be your writing, and it has to be more than two sentences. A description of your voice ("friendly but professional") does almost nothing; an example does almost everything.
What changes
Nothing model-specific. If a model update makes edits heavier-handed, add the line "return the text with the same paragraph count"; that constraint is usually enough to pull it back.