In the system prompt for the charter-desk assistant, the engineer writes: 'You are a maritime contracts assistant. Answer only using the retrieved clause text provided below; if the answer is not in the retrieved text, say you don't know.' Why is this instruction placed in the system prompt rather than left for the user to type each time?
Select an answer to reveal the explanation.
Short Explanation
You don't want every planner typing 'and please don't make things up' before every single question - you set that rule once, in the system prompt, and it quietly governs every turn after that, like a standing house rule instead of a reminder you repeat at every meal.
Full Explanation
The system prompt is the correct home for this instruction because its role in the chain is to set persistent, cross-turn behavior - grounding discipline, refusal behavior, and persona - that should apply uniformly regardless of what any individual user happens to type, which is precisely what 'answer only from retrieved text, else say you don't know' needs to be. The claim that system prompts run at no token cost is false; system-prompt tokens are still processed and billed like any other prompt tokens, caching (where available) only reduces repeated-processing cost, it does not make them free. The claim that user prompts cannot contain instructions is false; users routinely give instructions in their turn (e.g. 'summarize this in three bullets'), the system/user split is about role and precedence, not about which one is allowed to contain imperative language. The claim that retrieved context can only legally be inserted via the system prompt is also false - retrieved context is commonly injected into the user turn or a dedicated context block, and there is no such restriction; the real reason for the system-prompt placement here is persistence of behavior across turns, not a hard technical constraint on where context can go.