AWS Builder Center
Turn the AWS Exam Questions You Got Wrong into Hands-On Labs

Turn the AWS Exam Questions You Got Wrong into Hands-On Labs

An open-source agent skill that turns missed certification practice questions into Microcredential-style labs in your own AWS account, with cost and security guardrails, an automated grader, and a self-test that deploys every lab before you study with it.

Student Builder Campus Leader, Software Engineer | Distributed Systems & Cloud | Go & TypeScript
This is how most of us review a practice exam:
  1. open the question we got wrong.
  1. read the explanation.
  1. nod.
  1. move on.
Two weeks later, the same concept shows up with different wording and we miss it again.
Reading why an answer is right is not the same as knowing it and AWS clearly agrees: AWS Microcredentials have no multiple choice at all. You get a live environment, a business scenario, and about 90 minutes to fix or build things in the Console, with no step-by-step.
So I built a tool that does that for the questions you got wrong.

What it does

Wrong Answer Labs is an open-source Agent Skill, installable as a Claude Code plugin. You paste the questions you missed along with the exam code, and the agent:
  1. Checks every answer against the official AWS documentation. If you don't have the answer key, it works the answer out and cites the page that proves it.
  1. Groups related questions into scenarios, one lab per scenario.
  1. Writes each lab: a short business scenario, a CloudFormation template for the starting environment (often broken on purpose), challenges that state the outcome instead of the steps, and a grader script that checks your work.
  1. Writes documentation-only notes for topics a study account can't reproduce, like Direct Connect, Organizations policies, or sharing with a second account.
Each lab also ships a cleanup script and an annotated solution that explains why each wrong option was wrong.

The hard part: trusting AI-written infrastructure

An LLM writing CloudFormation that runs in your account is a real risk. The design rule I followed: the model writes the lab, but deterministic tools decide whether it ships.
  • Cost: CloudFormation Guard rules that are default-deny. Only resource types someone reviewed for price are allowed, sizes are capped, and nothing may survive the cleanup. A template with a 16 TB volume or a 10-node cache cluster simply fails.
  • Security: no SSH or RDP open to the internet, no admin permissions, private buckets and databases, container images pinned by digest.
  • Correctness: cfn-lint on every template, a grader that may only call read-only AWS operations, a check that every AWS CLI command in the scripts actually exists, and a link check that catches invented documentation URLs.
  • Prices from the Price List API, not from the model's memory.

Why the self-test matters

Static checks are not enough, and my own test proved it. The web tier lab passed every one of them, then scored 8 out of 11 with its own reference solution in a real account:
  • the cleanup script called an AWS CLI waiter that does not exist;
  • one grader query had a JMESPath precedence bug, so it always returned nothing;
  • the solution gave up attaching the WAF after two minutes, while the WAF docs say propagation can take several minutes.
Each of these became a new automated check. The lesson stays: before you study with a lab, let the agent run the self-test. It deploys the lab, confirms the grader scores zero before the solution and full marks after, then deletes everything. It usually costs cents. After the fixes, that lab scored 11 out of 11 and left nothing behind.

Try it

Inside a Claude Code session:
/plugin marketplace add jhermesn/wrong-answer-labs
/plugin install wrong-answer-labs@jhermesn
Then install the AWS CLI, cfn-lint, cfn-guard, ShellCheck and jq (or ask the agent to), sign in with  aws login , and paste your questions:
I'm studying for SAA-C03. These are the questions I got wrong: …
Use a study account, never production, set an AWS Budget, and always run the cleanup. Most labs cost well under a dollar.
The project is MIT-licensed on GitHub:  github.com/jhermesn/wrong-answer-labs . If a lab comes out wrong, the issue form asks for exactly what I need to fix it, and every report becomes a new check. It's an independent project, not affiliated directly with AWS.
Which topic do you keep getting wrong no matter how many explanations you read?
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article