Checklist: Prompt & Assistant Testing

Use this checklist when validating a prompt or assistant before wider deployment.

1. Functional correctness
   • Does the assistant produce the required fields or format on >90% of test inputs?
   • Are outputs factually consistent with provided context or retrieval snippets?
2. Robustness and edge cases
   • Test ambiguous inputs — does the assistant ask clarifying questions or safely refuse?
   • Test adversarial inputs (malformed, contradictory, or abusive). Does it behave appropriately?
3. Safety & compliance
   • Does the assistant avoid forbidden content and follow domain-specific compliance rules?
   • Are sensitive data handling rules enforced (never leak PII in generated examples)?
4. Performance
   • Are latency and token-cost within acceptable bounds for the workflow?
   • Does chaining add unacceptable delay or cost?
5. Maintenance
   • Is the system instruction and key examples stored in version control?
   • Is there a simple regression test suite that can be re-run after changes?
6. User experience
   • Is the tone appropriate for the audience and consistent across responses?
   • Does the assistant provide next steps and clearly indicate uncertainty when present?

If any checklist item fails, prioritize fixes by user impact: safety and compliance first, then functional correctness, then cost/latency.