My prompt does: [task]. Good output means: [your goals in plain words]. Create:
1. A rubric with 4–6 criteria, each scored 1–5 with a one-line description of a 1, 3 and 5.
2. 10 test inputs covering normal, edge and adversarial cases.
3. A grading prompt I can give to another model to score outputs against the rubric, returning JSON {criterion: score, notes}.