AI & LLMs

System Prompt

The standing instructions a model receives before any user input — identity, rules, output contract.

The system prompt is the highest-priority instruction block: who the model is, what it must and must not do, and how to format output. Written well it sits at the right altitude — firm on outcomes and guardrails, not micromanaging steps. It is also a stable prefix, which makes it ideal for prompt caching. Models weight it above user turns, but it is guidance, not a security boundary.

Worked example: the system prompt is the standing instruction set — persona, rules, tools, format — prepended before the conversation and obeyed every turn; the harness’s ‘constitution’ for the model. Gotcha: keep it stable so prompt-prefix caching can skip re-encoding it, and remember it is not a hard security boundary — a strong enough injection or jailbreak can override it, so never put secrets in it.