46.2 Prompt checklist (inputs, outputs, edge cases, failure paths)

Prompt Checklist

## Prompt Checklist

### Structure
- [ ] Clear role defined (system instruction)
- [ ] Context provided (what does the model need to know?)
- [ ] Task clearly stated
- [ ] Output format specified

### Inputs
- [ ] How is user input sanitized?
- [ ] What if input is empty?
- [ ] What if input is too long?
- [ ] What if input is in wrong language?

### Outputs
- [ ] Schema defined (Zod, JSON Schema, etc.)
- [ ] responseMimeType set if JSON required
- [ ] Examples of correct output provided
- [ ] Maximum length specified if needed

### Edge Cases
- [ ] What if model doesn't know the answer?
- [ ] What if question is out of scope?
- [ ] What if input is ambiguous?
- [ ] What if input is malicious?

### Failure Paths
- [ ] How do we detect failure?
- [ ] What's the fallback?
- [ ] Is failure graceful for the user?
- [ ] Is failure logged for debugging?

Where to go next

• 46.3 Tool checklist