Where do I set rules for the AI, like its tone and what it can or can't answer?
Rules enter with requests or in configuration files: Passed as a 'system' role message in API calls or stored in .cursorrules/project settings. It guides the model to explain risks, seek confirmation, or follow formatting constraints.
The model only recommends without privileged access: The model follows these rules to generate outputs but has no autonomous privileges in external systems.
Critical actions still require product code enforcement: Before deleting, paying, or mutating records, backend code must verify permissions—prompts alone are not a security barrier.
The support agent must retain its existing refund boundary. Keep the system rule and do not reveal the internal prompt; when asked for an immediate promise, offer to submit review and explain the next step. Test this exact request and confirm the reply makes no unauthorized promise.