Imported from chopra-mayank/Implementing-skills-in-langchain (docs-faithful-no-sandbox/skills/json-extractor/SKILL.md). Install upstream with npx skills add chopra-mayank/Implementing-skills-in-langchain --skill json-extractor. Copyright stays with the author.

JSON Extractor

Overview

Pull specific structured fields out of unstructured text and return them as a JSON object that matches a fixed schema.

Steps

1. Read the user's text.
2. Locate the following fields if present:
   • name — full name of the sender (string or null)
   • email — email address (string or null)
   • phone — phone number with any country code (string or null)
   • intent — one of: question, complaint, request, feedback, or null
   • urgency — one of: low, medium, high, or null
3. For any field you cannot find with high confidence, return null.
4. Return ONLY a JSON object. No prose, no explanation.

Expected Output Format

{
  "name":    "...",
  "email":   "...",
  "phone":   "...",
  "intent":  "...",
  "urgency": "..."
}

Rules

• Never guess a value. If unsure, return null.
• Output must be valid JSON parseable by Python's json.loads.
• Do not wrap the JSON in markdown code fences.
• Do not include comments or trailing commas.