LLM Output JSON Schema and Validator Suite

The JSON suite for the lead qualification agent has been generated with strict validation for the lead score and budget fields, including a multi-stage recovery protocol.

Lead Qualification JSON Suite

📐 Strict JSON Schema:

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "Full name of the lead"
    },
    "email": {
      "type": "string",
      "format": "email",
      "description": "Valid contact email address"
    },
    "budget": {
      "type": "integer",
      "minimum": 0,
      "description": "Total project budget in USD"
    },
    "lead_score": {
      "type": "integer",
      "minimum": 1,
      "maximum": 10,
      "description": "Qualification score from 1 (low) to 10 (high)"
    }
  },
  "required": ["name", "email", "lead_score"]
}

🛠️ Self-Repair System Prompt Snippet:
"CRITICAL: Output MUST be a single, valid JSON object. No markdown blocks, no pre-amble, and no trailing commas. Validate that budget is a raw integer and lead_score is between 1 and 10 before finalizing your response."

🛡️ Fallback Protocol:

Attempt 1 & 2: On parse failure, catch the exception and pass the raw error string (e.g., 'Expecting , delimiter') back to the model with the instruction: 'Your previous JSON was invalid. Fix the syntax and try again.'

Attempt 3 (Final Fallback): Return a static JSON object: {"status": "manual_review_required", "error": "schema_validation_failed"} and flag the record in the database for human audit.

⚡ Token Optimization Notes:

• Shortened 'qualification_score' to 'lead_score' to save 5 tokens per call.
• Removed redundant 'USD' from description; moved currency constraint to the system prompt to keep schema light.

[Self-Check]

• All properties have explicit types and descriptions
• Required fields are clearly defined
• Self-repair prompt snippet provided
• Fallback protocol defined for repeated failures
• Schema is optimized for token efficiency

Next steps

1. Implement the schema into your LLM provider's 'response_format' parameter.
2. Update your error handling middleware to pass exceptions to the Attempt 1 retry logic.
3. Set up a monitoring alert for the Attempt 3 fallback trigger.