Anatomy of a production prompt

Amateurs write a sentence. Professionals write a structured prompt with clear roles and sections. A strong production prompt usually has:

• Role / system prompt: who the model is and its high-level goal + constraints.
• Task instructions: precise, unambiguous steps. Tell it what to do and what not to do.
• Context: retrieved documents, user data, examples — clearly delimited.
• Output format: exact shape (JSON schema, sections). Show, don't just tell.
• Guardrails: how to handle unknowns ("If the answer isn't in the context, say 'I don't know'").

SYSTEM: You are a support-ticket classifier for an e-commerce company.
Classify each ticket into exactly one category and rate urgency.

Rules:
- Use ONLY these categories: [billing, shipping, returns, technical, other]
- If unsure, use "other". Never invent a category.
- Output valid JSON only, no prose.

Output schema:
{ "category": string, "urgency": "low"|"medium"|"high", "reason": string }

Ticket: """{ticket_text}"""