Bugnik

Amazon OA code repository question (2026): format, AI assistant and how to practise

By Bhimana Sai Pradeep · Published

Bugnik is an independent practice site. It is not affiliated with Amazon, HackerRank or any employer, and its problems are original, written in the same format. This guide is based on the public sources listed at the end.

In short: some Amazon software engineering online assessments (OAs) now include a code repository question. Instead of writing one function, you get a small multi-file project in a browser IDE on HackerRank. You read the README, find and fix bugs (or build a small feature), and tests decide your score. An AI assistant is built into this part of the test. Candidates usually describe about 60 minutes for it, after a normal algorithm question.

This guide collects what HackerRank and Amazon say publicly, what candidates have reported, and how to practise for it. Formats change, so the section on what isn't known matters as much as the rest.

What is the code repository question?

HackerRank, the platform the Amazon OA runs on, describes code repository questions as “multi-file, real-world coding problems that simulate actual development work”. You use one repository for all tasks in the question. HackerRank lists two kinds of task: bug fixing and feature building. Candidates mostly report bug fixing: something in an app is broken (a request fails to save, a page shows the wrong data) and you trace it to the code.

It tests a different skill from the algorithm question. You start from code someone else wrote, so most of the time goes into reading: finding where a request enters, which service handles it and which line is wrong. For a longer explanation of the format itself, see what repository-based questions are and how they're scored.

How the assessment is structured

HackerRank's page for Amazon candidates says the test has two sections:

  1. Coding: a classic algorithm question. You pick the language inside the test. HackerRank lists 2 sample and 13 hidden test cases for it, and generative AI is not permitted.
  2. Project (Code Repository): the repository task, with an embedded AI assistant.

Finish the coding section first: once you move to the code repository section, you cannot go back. Amazon's own OA page lists a coding assessment (about 70 minutes on average), a workstyles assessment (15), a work simulation (60, full-time roles only) and a feedback survey (5), and says the structure depends on the country. It doesn't mention the repository question by name. Candidates who got it describe the coding part as about 40 minutes of algorithms followed by about 60 minutes of repository work. Check the email that comes with your own invite: it is the only description that is certain to match your test.

Which languages and frameworks can you choose?

You choose the framework for the repository section when you open the invite, and HackerRank says you cannot switch once the test starts. Candidate reports list these options:

HackerRank's January 2026 release notes also mention full-stack MERN repositories and Node.js, Django and Spring Boot projects. Pick the stack you can read fastest, not the one that sounds most impressive: the clock goes into navigating unfamiliar files.

Bugnik's practice problems are in C++ and Java Spring Boot today. If you need MERN or Django, request it on the problems page; requests help decide which stack comes next.

Is AI allowed? How the AI assistant works

Only the assistant built into the code repository section. HackerRank's candidate page says:

Candidates' opinions of the assistant vary from useful to “not very helpful”. Treat it as a fast way to ask what an unfamiliar file or annotation does, and verify anything it suggests by running the tests. The skill being tested is still yours: finding the bug and knowing the fix is right.

How it's graded: tests, including ones you can't see

You check your work with the Run Tests button. The IDE saves automatically, you can resubmit as often as you like until time runs out, and if time ends first, your most recent saved code is evaluated. Do not edit the README or any file marked read-only.

Candidates describe the score as coming from tests, and say passing the tests you can run isn't always the whole story. Reported bug counts range from 3 to 8 per repository. Plan for more failures than the first test run shows: fix one, run again, and keep going.

A 60-minute plan

  1. Minutes 0–5: read the README fully and run the tests once before touching anything. The failing test names are your map.
  2. Minutes 5–10: if the project doesn't build, fix that first. A build error blocks every test, and the compiler message names the file and line.
  3. Minutes 10–45: take one failing test at a time. Read the test, follow the call into the code (controller, then service, then repository or helper), find the line that disagrees with the README, make the smallest fix, run the tests again.
  4. Minutes 45–55: when the visible tests pass, re-read the README rules and check the edges they mention: boundaries, empty input, duplicates, error codes.
  5. Last 5 minutes: run everything one more time and make sure your last change is saved.

What isn't known yet

We'll update this guide when that changes. The date at the top shows when it was last updated.

How to practise for it

Frequently asked questions

Is the code repository question in every Amazon OA?

No. Many 2025–2026 reports include it, but some candidates got two algorithm questions instead, and Amazon says the OA structure depends on the country and role.

How long is the code repository section?

Candidates usually report about 60 minutes, after about 40 minutes of algorithm coding. Your invite email has the timing for your test.

Can I use ChatGPT or another AI tool?

HackerRank says generative AI is not permitted in the coding section, and the code repository section has its own embedded assistant. Use that one; nothing on HackerRank's page allows outside tools.

Does using the built-in AI assistant lower my score?

HackerRank says you won't be penalized and that interactions are not considered in your score, but the transcript is visible to the evaluator.

Are Bugnik's problems real Amazon questions?

No. They are original problems written in the same format: a multi-file repository, seeded bugs or a feature to finish, visible tests you can run and hidden tests on submit.

Sources

Practise the format

Free multi-file C++ and Spring Boot problems with seeded bugs, visible tests and hidden tests, in the browser.

Browse the practice problems →