Use this before asking an LLM for output.
What are we asking the model to do?
Which language operation does this use?
- summarize supplied material with a lens
- classify against explicit categories
- compare options against criteria
- rewrite for a known audience or voice
- translate between domains or vocabularies
- generate candidates inside constraints
- critique against a checklist
- extract decisions, risks, owners, or open questions
- other:
What context must be supplied before the model can do this well?
What should the model refuse to guess?
What artifact should come back?
- Format:
- Required sections:
- Evidence requirement:
- Missing-context section:
How will a person tell whether the output is good?