System Prompt

​

What goes in a system prompt

• The AI’s role and primary purpose
• Persona (guardrails, edge case resolution, tone & style, etc.)
• Any policies or rules that must always apply

​

Tips

• Define the structure: role -> rules -> constraints -> examples.
• Define output format explicitly. If you need it to return JSON, write it explicitly with example outputs.
• Include error-handling instructions: what should the AI say when it doesn’t know?
• Test with adversarial inputs (prompt injections) to find edge cases your system prompt doesn’t handle.
• System prompts are the highest-trust context — keep them tight and deliberate.

​

API value

"assetType": "system_prompt"

​

Example

{
  "name": "Customer support agent",
  "assetType": "system_prompt",
  "content": "
  You are a customer support agent for Versuno.

  Your job is to help users resolve issues with the Versuno platform.

  You are helpful, concise, and professional.

  Only answer questions related to Versuno — redirect off-topic questions directly.

  When you don't know the answer, say so clearly and offer to escalate.

  Always respond in plain text, no markdown, no json, no xml.
  "
}