Exams / AIP-C01 / Study guide / Prompt templates, versioning and governance (Bedrock Prompt Management)
Prompt templates, versioning and governance (Bedrock Prompt Management): AIP-C01 study notes and practice questions
What to know about Prompt templates, versioning and governance (Bedrock Prompt Management) for the AIP-C01 exam, from the Foundation Model Integration, Data Management, and Compliance domain (31% of the exam guide), then 1 free question to test it. Part of the AIP-C01 study guide.
AIP-C01 asks how teams share prompts safely, release changes with control, and prove later who changed what and which prompt produced an answer.
What Bedrock Prompt management provides
Amazon Bedrock Prompt management stores prompt templates with named variables, so one template serves many applications. You can create versions of a prompt, and variants that compare different wording or models side by side. Prompts work with the Converse API and with Amazon Bedrock Flows, so a prompt can be reused across workflows.
Governance patterns
Need
Approach
Teams reuse one approved template
Bedrock Prompt management with variables
Changes need review before production
Prompt versions promoted through a pipeline with approval
Prompts stored with application code
Keep templates in Git or S3 with pull request review
Who changed a prompt
CloudTrail records the management API calls and caller identity
Which prompt text produced an answer
Model invocation logging, which is off by default and must be enabled
Role and safety rules applied everywhere
Put the role in a centrally managed system prompt, and enforce safety with a guardrail
CloudTrail records API activity, such as who created or updated a prompt. It does not record prompt or response content. Model invocation logging can send requests and responses to CloudWatch Logs or S3, and it must be turned on first. Plan for the log access and retention this data needs.
How to choose
Scenario wording
Choose
“Several teams share approved prompts with variables”
Bedrock Prompt management
“Prompt change must be reviewed like code”
Git or S3 storage with review and versioned releases
“Audit who edited a prompt”
CloudTrail
“Investigate exactly what the model received”
Model invocation logging to CloudWatch Logs or S3
“Compare two prompt wordings on the same inputs”
Prompt variants, or an evaluation job
Common traps
Hard-coding prompts in many services, so a fix must be copied everywhere.
Expecting CloudTrail to show prompt text. It shows API calls, not content.
Assuming invocation logs are on. They are off by default.
Editing the production version directly instead of publishing a new version after testing.
Relying on the system prompt alone for safety. Use a guardrail for enforcement.
Storing prompts with sensitive data in logs without a retention and access plan.
An online bank has five teams that each write prompts for summarizing customer messages, drafting replies and classifying complaints. Today the prompts are copied into each team's code, so the same instruction exists in several slightly different versions.
Audit asks which prompt version produced which reply, and the compliance team must approve each change before any app uses it. A bad change must be rolled back within minutes, without a redeployment of every app.
The prompts use shared variables such as customer name, product type and language, which all teams fill in the same way.
Which TWO actions should the team combine to meet these requirements? (Choose two.)
Choose two answers.
AA. Keep each prompt in the team's Lambda code, and tag each release in Git, so that the version of every prompt can be traced to a commit.
BB. Create the prompts in Amazon Bedrock Prompt Management with declared input variables, and publish an immutable prompt version for each change that compliance approves.
CC. Store the prompt text in a shared S3 bucket, overwrite the file each time a prompt changes, and have every app read the newest file at startup, so all teams always use the same wording.
DD. Let each team store its prompts in its own DynamoDB table with no fixed schema, so teams can add or rename fields quickly, and read the latest item in each app.
EE. Keep the approved prompt version ARN for each app in AWS AppConfig, deploy changes with a staged strategy and a CloudWatch alarm, and have apps read it through the AppConfig agent.
Show the answer and explanation
Answers: B, E
A. Wrong: Git tags trace code history, but the prompts would still be duplicated across five codebases, and every change would need a redeployment. The shared variables and fast rollback requirements are not met.
C. Wrong: Overwriting the file erases the earlier versions, so audit cannot see which prompt produced a reply. Apps also read at startup, so a rollback needs a restart of each app.
D. Wrong: Each team's table has its own fields, so variables are not shared in a consistent way. No approval step or version history is built in, so the compliance requirement is not met.
E. Right: Apps read the version ARN at runtime, so moving to a new prompt version or back to an earlier one is a configuration deployment, not an app redeployment. AppConfig can roll back automatically when the alarm fires, which meets the minutes-long rollback target. Learn more: Working with deployment strategies (opens in a new tab)
Explanation
Prompt Management gives one shared template per task with variables, and its immutable versions let audit trace each reply to an approved prompt. AppConfig then controls which version each app uses, so a rollout or rollback is a configuration change with alarm-based rollback instead of a redeployment. Code-held prompts, overwritten files and per-team tables each lose version history, sharing or control.
Want more on Prompt templates, versioning and governance (Bedrock Prompt Management)? The 5 full-length practice sets have 5 more Prompt templates, versioning and governance (Bedrock Prompt Management) questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →
Study Prompt templates, versioning and governance (Bedrock Prompt Management)