What is a System Prompt and How to Write One for Agents?

A "System Prompt" (or System Message) is the most important part of an LLM call. It sets the "rules of the game" for the AI, defining its persona, its capabilities, and its constraints.

Defining the Persona and Role

An effective system prompt starts with a clear role definition (e.g., "You are a senior cybersecurity analyst specializing in threat detection"). This "primes" the model to use a specific vocabulary and follow specific professional standards, leading to more expert-level responses than a generic assistant.

Setting Hard Constraints

The system prompt is where you define what the AI *cannot* do. You should include explicit instructions on safety ("Never leak API keys"), tone ("Always be concise and avoid corporate jargon"), and formatting ("Only return output in valid JSON"). Clear constraints are the primary defense against hallucinations and unpredictable AI behavior in production.