When designing a system prompt for an agent, what is the primary purpose of defining explicit 'constraints' beyond 'goals' and 'persona'?

When designing a system prompt for an agent, 'goals' define what the agent should achieve, specifying its primary objectives or desired outcomes. 'Persona' establishes the agent's identity, tone, and style, dictating how it should present itself and interact. Beyond these, the primary purpose of defining explicit 'constraints' is to impose specific limitations, rules, or boundaries on the agent's behavior, actions, or output format, irrespective of its stated goals or persona. Constraints serve as critical guardrails that prevent undesirable, harmful, or non-compliant actions by specifying what the agent *must not do*, *must strictly adhere to*, or *must avoid*. For instance, a goal might be to provide helpful information, and a persona might be friendly, but a constraint would explicitly state, 'Do not share personal identifiable information,' or 'Responses must always be less than 200 words,' or 'Only use information from the provided document and do not hallucinate.' This ensures safety, ethical conduct, factual accuracy, compliance with policies, and adherence to specific technical requirements by tightly controlling the agent's operational scope, even if its general objective is to be helpful or knowledgeable.