A specialist writes this line into a topic: "Never provide a diagnosis, and never tell an owner whether a symptom is an emergency; hand the conversation to a licensed staff member instead." How is this line best characterised in Agentforce configuration terms?
Select an answer to reveal the explanation.
Short Explanation
Instructions say "here is how to do the job." Guardrails say "here is the line you do not cross." A rule that starts with the word never, and that exists to stop the agent rather than to help it finish a task, is a guardrail every time.
Full Explanation
Guardrails are instructions whose purpose is to constrain the agent: prohibitions, mandatory escalations, and refusal conditions that hold for the whole topic regardless of which action is selected. A no-diagnosis rule paired with a mandatory hand-off to a human is the textbook shape. It is not a classification description, because classification descriptions are matched against an incoming utterance to decide routing; this line never influences which topic is selected and would be useless there. It is not a scope statement either, and that is the nearest miss: scope also constrains, but it states the boundary of what the topic is responsible for accomplishing in positive terms, while this line is a behavioural prohibition applied during response generation. It is certainly not a filter, since filters operate on data returned to the agent, such as narrowing retriever results or action query results, and this rule is about what the agent may say and when it must escalate, not about which records it can see. The exam distinction to hold onto: instructions and guardrails govern behaviour and language, filters and permissions govern data.