ICE Nattee Grader for VS Code

Everything from the NatteeGrader website inside VS Code, so you never alt-tab:

• Sidebar listing every problem by chapter, with your progress from the grader: a tick for solved, a filled dot for in progress (with your best score), and solved/total per chapter. Exams are hidden by default.
• Statement viewer: click a problem to read its PDF beside your code. Switching to a linked solution file re-opens its statement automatically.
• Auto-link: a file named after a problem code (e.g. 05_List_15.py) links itself, no setup.
• Run tests (▶ in the editor title bar or status bar, or automatically on save) for .py, .cpp, .c files, with PASS / FAIL / TLE / RE per case. For a wrong answer it points at the first difference (line and column) and makes spaces visible, so mistakes like Hello Python. vs Hello Python. are obvious.
• Run one test or your own input: Nattee: Run One Test or Custom Input (type it, or use the clipboard).
• Submit to the grader without leaving VS Code (Nattee: Submit to Grader, the cloud button). It always asks for confirmation first, never submits on save, then shows the grader's points and per-test verdicts live.
• Test Results panel in the sidebar keeps results visible next to the statement.
• Explain a failure with AI (Nattee: Explain Test Failure (AI), the ✨ button): sends your code and the failing tests (up to 3 at once) to Google's Gemini, using your own Gemini API key, and shows why it fails. It's told to explain the bug, not to hand back corrected code - the point is to help you fix it yourself, not to do the assignment for you. Check your course's policy on AI assistance before using this.
  • Get a free key at aistudio.google.com → Get API key, then run Nattee: Set Gemini API Key. A Gemini app subscription (e.g. Google AI Pro) is separate from the API and isn't needed. School Google accounts may have AI Studio switched off; a personal Google account works.
  • On the free tier Google may use what you send to improve its products. Nothing is sent until you run the command. The model is set by nattee.geminiModel (default gemini-3.8-flash). If Google says it's overloaded, the extension retries twice and then asks nattee.geminiFallbackModel (default gemini-3.5-flash).
• Test cases are read from the grader's test case page and cached locally. The grader shows only the first 2 KB of each file: a test whose input is longer shows as SKIP (submit to check it), and a longer expected output is checked on the part that's shown. Credentials (grader and Gemini key) are kept in VS Code's secret storage, not a temp file.

Install

In VS Code open Extensions (Ctrl+Shift+X), search ICE Nattee Grader and click Install. Needs VS Code 1.90+. Cursor, Windsurf and VSCodium get it the same way, from Open VSX.

Or download nattee-grader.vsix from Releases and run code --install-extension nattee-grader.vsix. To build it yourself: python tools/build_vsix.py.

Use

1. Open the ICE Nattee Grader icon in the activity bar → Sign in.
2. Right-click a problem → Create Solution File. It creates <chapter>/<code>.py in your workspace (e.g. 05_List/05_List_15.py), links it and shows the statement. If a file with that name already exists anywhere in the workspace it opens that one instead. Set nattee.organizeByChapter to false to keep files flat. Or open any file and run Nattee: Link Current File to Problem.
3. Click ▶ to run the tests, or just save the file: tests run automatically on save (nattee.runOnSave).

Settings (nattee.*): rootUrl, hideExams, defaultExtension, organizeByChapter, pythonPath, cppCompiler, cCompiler, timeLimitSeconds, runOnSave, geminiModel, geminiFallbackModel.

Layout

• extension/ — the VS Code extension (plain JS, no build step). lib/client.js talks to the grader, lib/runner.js runs solutions, lib/statement.js is the PDF viewer.
• tools/build_vsix.py — packages the extension into a .vsix.

Platforms

Written to be cross-platform (Windows, macOS, Linux). The automated tests run on all three in CI (.github/workflows/test.yml), and it has been used by hand on Windows 11 only. Requires VS Code 1.90+ and Python (or g++/gcc for C/C++ solutions). On macOS the Python command is usually python3; the extension finds it automatically.

Development

cd extension
node --test          # unit tests, no VS Code needed
python ../tools/build_vsix.py
python ../tools/publish_openvsx.py   # Open VSX upload; token from OVSX_PAT or ~/.ovsx_token

The VS Code Marketplace upload is done by hand at marketplace.visualstudio.com/manage (Update → the .vsix).

License

MIT. Bundles pdf.js (Apache-2.0), see THIRD_PARTY_NOTICES.md.

This is an unofficial student tool. It only reads what your own grader account can already see, and stores it on your machine. Check your course's rules before using it.