AI Agent Notebook

Jupyter-style .aanb notebooks for vibe coding and agent workflows: Prompt cells send instructions to your LLM (Ollama, OpenAI, Codex, Client / OpenAI-compatible servers, Claude, Gemini); Markdown cells hold curriculum and notes; optional Code cells run small local JavaScript. Chat via @AI-Agent uses the same LLM settings as the notebook.

English summary: Teach and ship agentic notebooks in VS Code—ordered prompts, streaming cell output, cancellation, session resume, and a built-in education guide.

Features

• .aanb format — YAML frontmatter + markdown cell headers; serializer round-trips with VS Code notebook APIs.
• Cell execution — streaming output and interrupt/cancel for Prompt cells (agentNotebook.streamNotebookOutput).
• Notebook UX defaults — line numbers, cell toolbar placement, .aanb opens as the AI Agent Notebook editor.
• Chat participant — @AI-Agent with shared LLM configuration (short name matches VS Code’s allowed pattern for handles; fullName stays a readable title in Chat).
• Commands (category: AI Agent Notebook)
  .aanb를 노트북으로 열기 (Explorer context on .aanb); 바이브 코딩 교육 가이드 열기; 최근 .aanb 노트북 이어서 열기; 샘플 노트북 열기 (calculator.aanb); Execute Cell (when an AI Agent Notebook is focused); 실행 결과 정리 (removes Output cells and clears run outputs on Prompt/Code/Markdown while keeping source — notebook toolbar, editor title, cell menu, Command Palette, Explorer on .aanb).
• Shortcuts (optional): Ctrl+Shift+Alt+R (resume last), Ctrl+Shift+Alt+S (sample).
• Samples — samples/simple-calculator/calculator.aanb, samples/prompt-first/demo.aanb.
• Template AI Agent Notebook — 번들 templates/template_*.aanb로 바이브 코딩·창작 워크플로를 시작합니다. 파일 → 새 파일 하위의 **Template으로 AI Agent Notebook 시작하기…**에서 템플릿을 고르거나, template_*.aanb를 연 뒤 노트북 도구 모음·편집기 제목의 **「새 작업용으로 저장… (템플릿)」**으로 복사본을 만듭니다. 개념·설계: education/TEMPLATE_AI_AGENT_NOTEBOOK.md.

Privacy

Prompts and context go only to the LLM you configure. Do not put secrets in .aanb you plan to share. This extension does not send telemetry to the publisher; details: PRIVACY.md.

Quick start

1. Install — From Releases, download the latest agent-notebook-*.vsix, then VS Code → Extensions → … → Install from VSIX…. Reload the window once.
2. New notebook — Command Palette: AI Agent Notebook: 새 AI Agent Notebook (Untitled .aanb). You get a Jupyter-style layout: a Markdown explanation cell and a Prompt cell. Empty saved .aanb files open with the same starter when they contain no cells.
3. Open a notebook — Command Palette: AI Agent Notebook: 샘플 노트북 열기, or open samples/simple-calculator/calculator.aanb. If it opens as plain text: AI Agent Notebook: .aanb를 노트북으로 열기 or right-click the file in Explorer.
4. Configure LLM — Settings → agentNotebook (Ollama URL; keys for OpenAI / Codex; optional client.baseUrl + client.apiKey for LM Studio–style servers; Claude; Gemini).
5. Run cells — Use the cell Run button, or the notebook toolbar 전체 셀 실행. Toolbar also has + Markdown / + Prompt / + Code, 모든 출력 지우기 (clears outputs on every cell), and 실행 결과 정리 (Output-type cells removed + Prompt/Code/Markdown outputs cleared; confirms in a dialog).
6. Teaching — Command Palette → 바이브 코딩 교육 가이드 열기.
7. Templates — 파일 → 새 파일 → Template으로 AI Agent Notebook 시작하기… (또는 명령 팔레트 동일 이름). 저장 후 안내는 3초 후 자동으로 사라지거나 확인으로 바로 닫을 수 있습니다.

Settings (summary)

Azure OpenAI (Chat Completions–compatible): set defaultLLM.provider to client, point client.baseUrl at your resource (API version segment + /deployments/<deployment-id> as required by Microsoft’s REST shape), set defaultLLM.model to that deployment name, and set client.apiKey to the API key. Paths differ from https://api.openai.com/v1; adjust baseUrl until POST …/chat/completions matches your endpoint.

Build & version

• Shipped artifact — Single bundled out/extension.js (esbuild). Runtime dependencies in package.json are empty; axios and js-yaml are compiled into the bundle.
• npm run compile — Increments semver patch in VERSION and package.json (build counter: 0.1.n → 0.1.n+1), then runs esbuild.
• npm run bundle — esbuild only; no version bump (used by tests, vscode:prepublish, and CI).
• npm run watch — esbuild watch mode.
• npm run typecheck — tsc --noEmit.

Windows release script

From the repo root, release.bat runs npm ci → compile (bump + bundle) → test:coverage → packages dist\agent-notebook-<version>.vsix → creates or updates a GitHub release with gh when authenticated (gh auth login or GH_TOKEN).

CI

GitHub Actions (.github/workflows/ci.yml): npm ci → typecheck → lint → bundle → vsce package sanity → pytest gate (tests_python, runs Jest + line coverage check). If that job fails on a push to main in yskwon-wayforyou/agent_notebook, a follow-up job files a CI … failed GitHub Issue, posts a comment with steps to download the ci-failure-debug-logs artifact (workflow log ZIP + failure-context.txt), and uploads that artifact on the same run. To verify issue automation without breaking the build, run Actions → GitHub issue automation smoke (github-issue-automation-smoke.yml). Tag push v* triggers Release workflow (bundle, tests, VSIX, release notes).

Extension E2E (slow, downloads VS Code): run manually via Actions → E2E → Run workflow (.github/workflows/e2e.yml). Same as locally: npm run test:e2e.

Pytest gate (venv)

python -m venv .venv
. .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install -r requirements-ci.txt
pytest tests_python -q

Development

npm install
npm run typecheck
npm run lint
npm test                      # Jest (pretest: typecheck + bundle, no version bump)
npm run test:coverage         # Jest + coverage thresholds
npm run test:e2e              # Extension host smoke test (bundle + @vscode/test-electron; downloads VS Code on first run)

Install this checkout into Cursor / VS Code on your machine (run tests, build VSIX, --force install — then reload the window):

npm run apply:local

Requires cursor and/or code on your PATH. More on VSIX, optional skips, and releases: PUBLISHING.md; history: CHANGELOG.md.

License

MIT — see LICENSE.

Author

yskwon — publisher wayforyou — GitHub