Code Quiz — Do You Actually Understand This?

A VS Code extension that watches your code in real time and quizzes you on it — instead of explaining, autocompleting, or fixing it for you.

Built in response to a simple observation: AI coding tools are excellent at generating, completing, and explaining code—but they rarely verify that the developer actually understands what was written. Code Quiz closes that gap by asking adaptive questions as you write or paste code, turning passive AI assistance into active learning. Instead of remembering to ask an AI to explain every 20 or 30 lines, understanding becomes part of the development workflow. Whether you wrote the code yourself or pasted it from an AI assistant, the extension continuously checks your understanding by asking targeted questions about the code's intent, edge cases, complexity, and potential bugs before you move on. It also asks multiple questions and makes it part of active learning which is often better for understanding and retention.

Not autocomplete. Not Copilot. Not "explain this code." Questions only.

---

Demo

You paste or finish writing code → the sidebar opens with a Socratic question about a specific part of it → you answer → it tells you if you're right → it asks the next one. A typical session is a 5-question batch covering different functions, edge cases, and bug risks across the file.

Toggle Professor Mode and it stops giving you the answer entirely, even when you ask for a hint — it just asks another question instead.

---

How it works

Trigger	What happens
Paste a block of code (120+ characters)	Assumes you copied it without reading it closely — generates a multi-question batch about it
Finish writing a function	Waits a few seconds of inactivity, then asks about what you just wrote
Right-click → "Quiz Me On This File Now"	Generates 5 questions covering the entire file, not just the last thing typed

Each trigger generates a batch of 2–5 questions (scaled to code size), shown one at a time with a progress indicator ("1 of 5"), a verdict after each answer, and a score summary at the end.

---

Setup

1. Install

git clone https://github.com/your-username/code-quiz-extension.git
cd code-quiz-extension
npm install
npm run compile

# Package it
npm install -g @vscode/vsce
vsce package --allow-missing-repository

# Install the .vsix into VS Code
code --install-extension code-quiz-0.1.0.vsix

Or press F5 inside the project in VS Code to launch an Extension Development Host immediately, no packaging required.

2. Add a free Gemini API key

1. Get one at aistudio.google.com — free tier, no credit card
2. In VS Code: Cmd+, (or Ctrl+,) → search Code Quiz → paste it into Gemini Api Key

Without a key, the extension still runs — it falls back to a small set of built-in offline questions per language.

---

Settings

Setting	Default	Description
codeQuiz.geminiApiKey	""	Google Gemini API key for AI-generated questions
codeQuiz.triggerOnPaste	true	Generate questions when a large code block is pasted
codeQuiz.triggerOnFunctionComplete	true	Generate questions when a function is finished
codeQuiz.debounceSeconds	6	Idle time after typing before a function-complete trigger fires
codeQuiz.professorMode	false	Never reveal answers or give real hints — only ask more questions
codeQuiz.difficulty	"standard"	gentle / standard / rigorous

---

Professor Mode

With Professor Mode on, every "give me the answer" path — hints, follow-up requests — gets intercepted and replaced with another probing question instead. The goal is to make the aha moment of finding your own bug or gap in understanding the only way out, not a button click away.

---

Why questions instead of explanations

There's a well-documented effect in learning research called the testing effect: being asked to retrieve information produces dramatically better retention than being told the same information, even when the explanation is perfect. Reading an explanation feels like learning. Being forced to produce an answer is what actually builds it.

Asking an AI to "explain this code" is still passive consumption. This flips it — the model holds the answer, you have to produce it first.

---

File structure

code-quiz-extension/
├── src/
│   ├── extension.ts   ← entry point, command registration, question queue logic
│   ├── watcher.ts      ← detects paste / function-complete events from the editor
│   ├── claude.ts       ← Gemini API calls (question generation, answer verification, hints)
│   ├── panel.ts         ← sidebar webview UI (questions, verdicts, progress, summary)
│   ├── session.ts       ← tracks answer history, streaks, stats for the session
│   └── types.ts          ← shared TypeScript interfaces
├── media/
│   └── icon.svg
├── package.json
├── tsconfig.json
└── LICENSE

---

Question categories

Each generated question is tagged with one of:

• Intent — what does this code actually do?
• Edge case — what happens with empty/null/zero/extreme input?
• Complexity — what's the time/space cost?
• Bug risk — what could silently break here?
• Alternative — why this approach instead of another?
• Side effect — what else does this touch or mutate?

---

Known limitations

• Trigger detection (paste size, function-end regex) is heuristic and language-specific — it won't catch everything in every language equally well
• Requires a Gemini API key for real AI-generated questions; the offline fallback set is intentionally small
• No persistence between VS Code sessions yet — stats and history reset on reload

---

Roadmap

• Persist session history across reloads
• Spaced repetition — resurface categories you've gotten wrong before
• Per-language question packs tuned for that language's common pitfalls
• Configurable question count for "Quiz Me On This File" (currently fixed at 5)
• Export a session as a markdown learning log

---

License

MIT — see LICENSE