Scaffold tracks how complete your work is against assignment requirements as you write. Point it at an assignment (manually or auto-discovered from Canvas, Notion, or Google Classroom), and Claude breaks the rubric into measurable tasks and scores your draft 0–100% in real time — in a web dashboard and a browser sidebar for Google Docs / Notion.

clients (web + extension)  ──►  FastAPI  ──►  Claude (rubric analysis + scoring)
                                   │
                                   ├──►  Redis      (cache + progress history)
                                   ├──►  Supabase   (persistent storage)
                                   ├──►  Browserbase + Stagehand (LMS discovery)
                                   ├──►  Sentry      (errors + AI monitoring)
                                   └──►  Arize Phoenix (LLM tracing)

1. Create an assignment — manually via the API/UI, or auto-discovered from a platform.
2. Analyze the rubric — Claude decomposes the prompt + rubric into measurable tasks.
3. Track progress — as you write, Claude scores each task and surfaces what's missing.
4. Cache smartly — Redis skips re-calling Claude unless the document changed ≥100 chars.

• Docker & Docker Compose
• Node.js 18+
• Python 3.12 (only if running the backend without Docker)
• API keys: Anthropic and Supabase are required; Browserbase, Sentry, and Arize Phoenix are optional.

Run backend/supabase_schema.sql in the Supabase SQL editor (or via psql). It creates the assignments table the backend expects.

cp backend/.env.example   backend/.env
cp web/.env.example       web/.env.local
cp extension/.env.example extension/.env

Fill in the real values. At minimum the backend needs ANTHROPIC_API_KEY, SUPABASE_URL, and SUPABASE_SERVICE_KEY.

docker compose up --build

Backend

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload   # needs a Redis instance reachable at REDIS_URL

Web dashboard

cd web
npm install
npm run dev          # http://localhost:3000
# production: npm run build && npm start

Browser extension

cd extension
npm install
npm run dev          # or: npm run build

Then in Chrome: Extensions → Developer mode → Load unpacked and select the Plasmo output folder (extension/build/chrome-mv3-dev for dev, chrome-mv3-prod for build). The sidebar injects on docs.google.com/document/* and notion.so.

cd backend
pip install -r requirements.txt -r requirements-dev.txt
pytest

The suite covers pure logic (rubric/JSON parsing, change detection, scraped-data normalization) and the no-dependency API routes (/health, /api/discovery/supported). It runs without any live external services.

cd web
npm install
npm test

# Health
curl http://localhost:8000/health

# Create an assignment (triggers Claude rubric analysis)
curl -X POST http://localhost:8000/api/assignments/ \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Essay on Climate Change",
    "prompt": "Write a 1000-word essay analyzing three causes of climate change.",
    "rubric": [
      {"criterion": "Thesis", "description": "Clear thesis in intro", "points": 20},
      {"criterion": "Evidence", "description": "Three cited sources", "points": 30}
    ]
  }'

# List assignments
curl http://localhost:8000/api/assignments/

# Update progress (replace {id})
curl -X POST http://localhost:8000/api/assignments/{id}/progress \
  -H "Content-Type: application/json" \
  -d '{"assignment_id": "{id}", "document_content": "Climate change is driven by..."}'

backend/    FastAPI app, services (Claude, Redis, Supabase, Browserbase, Sentry, Arize)
web/        Next.js dashboard
extension/  Plasmo Chrome extension (Google Docs / Notion sidebar)
docker-compose.yml