Testing a chat feature needs conversations, and writing forty of them by hand is nobody's afternoon. These are structurally plausible system and user messages for exercising storage, token counting and rendering — not for producing good model output.
What you can control
- System messages combine a role, a task and a configurable number of constraints, which is how most real ones are shaped.
- Pair mode produces a matching system and user turn for seeding a conversation record.
- The optional output schema gives you something to test structured-output parsing against.
- Worked examples exercise the few-shot section that many prompt templates include.
What this is not
These prompts are for filling a database, not for use with a model. They are assembled from templates and will not produce useful responses.
Questions
Will these prompts give good results with a real model?
No. They are shaped like prompts for testing your plumbing — storage, token counts, rendering. Write your actual prompts by hand.
Can I use these to test token counting?
Yes, that is a good fit. They vary in length the way real prompts do, so a token counter gets a realistic spread.
Why generate an output schema?
Structured-output parsing needs a schema to parse against. Having one alongside each prompt makes it easy to test the round trip.