A referral coordinator needs a tool that can both read data from a Case record and take the deterministic action of updating the Case status field once a referral hospital confirms receipt. Which combination correctly describes how this should be built?

The correct approach separates reading and narrating Case data from committing a deterministic change: a Flow updates the status field, either when run directly or as the backing action behind an agent topic. A Prompt Builder template alone can generate text grounded in Case data, but its output does not itself perform record DML; it cannot directly write the determined value to the field. A record-summary prompt template is also for generating narrative text, not for database updates, and it has no write-back permission setting that makes it commit changes. An Einstein Trust Layer masking rule serves a different purpose: masking can hide sensitive values from being sent to or returned by the LLM, but it does not perform a status update. Exam caveat: an agent may invoke an action backed by automation, but that does not make the prompt template itself the update mechanism. For a practical check, confirm that the Flow has the Case record and sets its Status field to the required value when the receipt confirmation occurs. Remember the distinction: prompt templates and generative grounding produce language; Flows and actions built on them produce deterministic record changes.