A permitting chatbot's prompt template needs to be versioned and promoted from a test environment to production separately from the underlying foundation model, so prompt edits can be reviewed and rolled back on their own. Which approach fits this need?

Bedrock Prompt Management treats a prompt template as a first-class, versioned artifact independent of the foundation model it's paired with, so a team can create a new prompt version, test it, and promote a specific version to production — and roll back to a prior version — without touching the model or redeploying the surrounding application. Hard-coding the prompt in application source code couples prompt changes to full code deployments, which loses the independent review and rollback the county wants for prompt edits specifically. Storing prompt variants as comments in training data confuses two unrelated artifacts: fine-tuning data shapes model weights during training, it isn't a live template the running application reads and isn't a mechanism for tracking prompt versions at all. Swapping the foundation model to change prompt behavior conflates the model and the prompt — a model change affects far more than the wording of one template and isn't a controlled way to iterate on prompt text specifically. Scope caveat: even with independent prompt versioning, a prompt change can still shift model behavior enough to warrant the same evaluation rigor as a model change before it reaches production. Operational check: promote a new prompt version to a test environment, verify its outputs against expected results, and confirm the production environment still runs the prior version until it's explicitly promoted.