Structured Output Designer

The Problem

When AI output needs to feed into code — populating databases, driving UI components, or triggering automation — free-form text is useless. You need predictable, type-safe structured output that your parser can handle without breaking. Most attempts at getting structured output fail because the schema is ambiguous, edge cases are not covered, or the model adds prose around the structure.

The Prompt

Generate output that EXACTLY matches this schema. No text before or after the structure.

SCHEMA:
```json
{
  "fieldName": "type | description | constraints",
  "items": [{
    "requiredField": "string | always present",
    "optionalField": "string | null if unknown",
    "enumField": "one of: option1, option2, option3",
    "numericField": "number | range: 0-100"
  }],
  "metadata": {
    "total": "number | count of items",
    "confidence": "number | 0.0-1.0"
  }
}

RULES:

• Output ONLY valid JSON, no markdown code fences
• Use null for unknown values, never empty strings
• Arrays must contain at least one item or be empty []
• All enum values must be from the specified options
• Numbers must be within specified ranges

INPUT: [your data or question]

## Example Output

```json
{
  "vulnerabilities": [
    {
      "id": "vuln-001",
      "severity": "critical",
      "location": {"file": "auth.ts", "line": 42},
      "description": "JWT secret hardcoded in source",
      "cwe": "CWE-798",
      "fix": "Move to environment variable"
    },
    {
      "id": "vuln-002",
      "severity": "high",
      "location": {"file": "api.ts", "line": 15},
      "description": "SQL query built with string concatenation",
      "cwe": "CWE-89",
      "fix": "Use parameterized queries"
    }
  ],
  "metadata": {
    "total": 2,
    "confidence": 0.92
  }
}

When to Use

Use this skill when building AI-powered tools that need machine-readable output — code generators, data pipelines, automated testing frameworks, or any system where AI output is consumed by code rather than humans. It is essential for production-grade AI integrations.

Pro Tips

• Provide a complete example — a filled-in schema with realistic data teaches the model better than a type description alone.
• Test with edge cases — ask for output when there are zero items, null fields, or very long strings to verify the schema holds.
• Use enum fields aggressively — constrained choice fields (severity: “low” | “medium” | “high”) produce more consistent output than free text.
• Add a metadata object — including count, confidence, and processing notes in a metadata field helps you validate the output programmatically.