You don't need a hundred support prompts. You need five you trust and actually reuse.
Reuse the recipe
Write down what you never want as explicitly as what you do. The exclusions often do more work than the instructions.
- Compare two outputs instead of trusting one
- Keep one ask per prompt — chaining beats cramming
- Have the model flag what it could not cover
Anchor to one reader
Treat the first output as a draft, not a verdict. The prompt does the typing; you keep the judgment about what ships.
Treat the first output as a draft, not a verdict. The prompt does the typing; you keep the judgment about what ships.