The seven ROIP criteria
Direction, Context, Format, Constraints, Examples, Evaluation, and Task Decomposition.
- Direction: what the task is and what it's for.
- Context: the facts about your situation the AI can't know.
- Format: the shape of the answer: its structure, required parts, and length.
- Constraints: what's off-limits, and the bounds the answer has to stay inside.
- Examples: a sample of what good looks like.
- Evaluation: how to tell the answer worked.
- Task Decomposition: splitting work too big for one prompt into steps.
Which criteria count
Not every criterion counts on every prompt. For each kind of task, each one is Core (always needed), Situational (needed only when the grader judges that prompt needs it) or not used, and each has a weight. A Debug prompt, for example, always needs Direction and Context; Evaluation and Constraints count only when that prompt needs them; Format, Examples and Task Decomposition don't count.
What a prompt does changes this too. See Why a short prompt can score 100.
The Prompting guides cover each criterion in depth, with weak vs. strong examples.