Skip to content

Challenge

Overview

Put your knowledge tools skills to the test! This 5-minute timed challenge asks you to use your agent to research an AWS architecture question that requires combining documentation search, API reference lookup, and regional availability checking.

What You'll Learn

  • How to chain multiple knowledge tool queries together
  • How to solve real-world architecture research tasks with your agent
  • How quickly you can gather comprehensive AWS information with agent assistance

Instructions

The Challenge

Time limit: 5 minutes

Scenario: Your team is planning to deploy a serverless image processing pipeline that uses Amazon S3, AWS Lambda, and Amazon Rekognition. The pipeline must be deployed in a region that supports all three services and must handle images up to 15 MB.

Your task: Use your agent to answer all of the following questions:

  1. Which AWS regions support all three services (S3, Lambda, and Rekognition)?
  2. What is the maximum image size that Amazon Rekognition DetectLabels can process when the image is stored in S3?
  3. What IAM permissions does a Lambda function need to call Rekognition:DetectLabels and read from an S3 bucket?
  4. What is the maximum Lambda function timeout, and is it sufficient for Rekognition to process a 15 MB image?

Start your timer and go!

Info

Tip: You can ask all four questions in a single prompt, or break them into focused queries. Experiment with what gets you accurate answers fastest.

Try solving the challenge using a single, well-crafted prompt that asks your agent to research the full scenario. Then try breaking it into individual questions. Which approach gives you better results?

Info

Hint 1: A combined prompt like "I'm building a serverless image processing pipeline with S3, Lambda, and Rekognition..." lets the agent reason about the full context at once.

Info

Hint 2: For the image size limit, ask specifically about the DetectLabels API input constraints — the limit differs depending on whether the image is passed as bytes or referenced from S3.

Walkthrough

Click to see the step-by-step walkthrough

  1. Start your 5-minute timer.

  2. Ask your agent the combined research question:

"I'm building a serverless image processing pipeline using S3, Lambda, and Rekognition. I need to know: (1) which regions support all three services, (2) the maximum image size for Rekognition DetectLabels when reading from S3, (3) the IAM permissions a Lambda needs for DetectLabels and S3 reads, and (4) whether Lambda's maximum timeout is sufficient for processing 15 MB images with Rekognition."

  1. Review the response for completeness. If any answer is missing or vague, follow up:

"What is the exact maximum image file size in MB for Rekognition DetectLabels using an S3 input?"

  1. If needed, ask for the IAM policy in a structured format:

"Write the minimal IAM policy JSON for a Lambda execution role that can read from an S3 bucket and call Rekognition DetectLabels."

  1. Stop your timer when you have all four answers.

Validation

This is a self-graded challenge. Compare your answers against the reference below.

Click to see the reference answers

  1. Regions supporting all three services: Nearly all standard regions support S3 and Lambda. Rekognition is available in a subset — including us-east-1, us-west-2, eu-west-1, eu-west-2, ap-northeast-1, ap-southeast-1, and ap-southeast-2 (among others). For this workshop, us-east-1 supports all three.

  2. Maximum image size for DetectLabels (S3 input): When the image is stored in S3 and referenced via the S3Object parameter, the maximum file size is 15 MB. When passed as raw bytes via the Bytes parameter, the limit is 5 MB.

  3. Required IAM permissions:

  4. rekognition:DetectLabels on resource *
  5. s3:GetObject on the specific bucket/prefix (e.g., arn:aws:s3:::my-bucket/*)

  6. Standard Lambda execution role permissions (logs:CreateLogGroup, logs:CreateLogStream, logs:PutLogEvents)

  7. Lambda maximum timeout: The maximum Lambda timeout is 15 minutes (900 seconds). Rekognition DetectLabels typically processes images in 1–5 seconds regardless of size, so the timeout is more than sufficient for 15 MB images.

Scoring: - 4/4 correct: ⭐ Excellent — you're ready for the capstone! - 3/4 correct: Good — review the missed area - 2/4 or below: Try refining your prompts for more specific answers

Agent-Specific Tips

Claude Code handles multi-part research questions well. Try prefixing your prompt with "Using the AWS documentation tools..." to encourage it to use MCP Server lookups rather than relying on training data.

Kiro works well with structured questions. Try numbering your questions in the prompt — Kiro will typically structure its response to match, making it easy to verify each answer.

For timed challenges in Cursor, Agent mode gives the best results since it can make multiple MCP tool calls in sequence. If one answer seems incomplete, ask a targeted follow-up rather than repeating the full question.

Codex can handle the combined query efficiently. If you want to speed up the challenge, ask Codex to output its answers in a numbered list format matching your questions.