A topic's instructions currently say only "answer the owner's question about lab results." Owners have occasionally been given results after only loosely confirming which pet they mean. What should the specialist add to the instructions to fix the sequencing, without relying on a guardrail or a filter?
Select an answer to reveal the explanation.
Short Explanation
Vague instructions produce vague behaviour. "Be careful" isn't a step. "Confirm the patient first, then look up results" is a step, and steps are what actually change what the agent does in what order.
Full Explanation
Instructions are the place to specify not just what a topic should accomplish but the order operations should happen in, so adding an explicit step that requires confirming the specific patient before the lab-results action is invoked directly fixes the sequencing gap: the agent is told to gather and verify an input before acting, rather than acting first and hoping the right pet was implied. A vague instruction to "be more careful" gives the model no concrete behaviour to follow and is the kind of under-specified instruction the exam expects you to recognise as a design smell; instructions should read like a procedure, not a mood. Removing the topic and pushing the work to staff abandons automation entirely over a fixable configuration gap and is a disproportionate response. Duplicating the topic per species does not address patient confirmation at all, since an owner with two dogs is still ambiguous within a single species-specific topic, so it solves a different, unrelated problem while leaving the actual identity-confirmation gap untouched.