ScaleCanvas

System design interview practice: React Flow canvas (left) + AI interviewer chat (right). See docs/MVP_PLAN.md for the implementation plan.

All coding rules in .cursor/rules/ are mandatory — not optional. CI and pre-commit enforce lint, typecheck, and tests so the rules are followed.

Stack

Next.js (App Router), Supabase, AWS Bedrock (Anthropic), Vercel AI SDK, React Flow, Tailwind CSS, TypeScript, shadcn/ui, Zustand, neverthrow, ts-pattern. Package manager: pnpm only.

Setup

pnpm install
cp .env.example .env.local
# Edit .env.local with your Supabase URL, publishable key, and (server-only) AWS/Bedrock vars
pnpm dev

Scripts

• pnpm dev — start dev server (Turbopack)
• pnpm build — production build
• pnpm start — start production server
• pnpm lint — run ESLint
• pnpm test — run tests (Vitest)

Enforcement

• CI (GitHub Actions): on push/PR to main, runs pnpm lint, pnpm exec tsc --noEmit, pnpm test, pnpm build. The build fails if any step fails.
• Pre-commit (husky): before each commit, runs pnpm lint, pnpm exec tsc --noEmit, pnpm test. Commit is blocked if any step fails. E2E tests (pnpm test:e2e) run in CI or manually, not in the commit hook.
• lint-staged: ESLint runs on staged *.ts/*.tsx when configured (e.g. for faster feedback).

Environment

Copy .env.example to .env.local and fill in:

• Client: NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_PUBLISHABLE_DEFAULT_KEY
• Server (optional): SUPABASE_SECRET_KEY
• Server-only (BFF /api/chat): AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION, BEDROCK_MODEL_ID — never use NEXT_PUBLIC_ for these.

IAM for Bedrock (inference profile)

The IAM user used by /api/chat must allow both the inference profile and the foundation model (e.g. for Claude Sonnet 4.6). When using cross-region inference, the foundation model is evaluated as a region-agnostic resource (arn:aws:bedrock:::foundation-model/...), so the foundation-model ARN in the policy must use a region wildcard:

{
  "Version": "2012-10-17",
  "Statement": [{
    "Effect": "Allow",
    "Action": ["bedrock:InvokeModel", "bedrock:InvokeModelWithResponseStream"],
    "Resource": [
      "arn:aws:bedrock:us-east-1:ACCOUNT_ID:inference-profile/global.anthropic.claude-sonnet-4-6",
      "arn:aws:bedrock:*::foundation-model/anthropic.claude-sonnet-4-6"
    ]
  }]
}

Replace ACCOUNT_ID with your AWS account ID.

Chat latency

Responses should stream and show the first tokens within a few seconds. If you see a long delay (e.g. 60–90s) before any output:

• Try production: pnpm build && pnpm start and test again (dev/Turbopack can behave differently).
• Verify streaming: The route sends Cache-Control: no-store and X-Accel-Buffering: no so the response isn’t buffered.
• Inference profile: A geographic profile in your region (e.g. us.anthropic.claude-sonnet-4-6 from US) can be faster than the global one; see AWS Bedrock inference profiles docs.

Folder structure

• app/ — routes, layouts, pages; API under app/api/
• components/ — UI; components/ui/ — shadcn primitives
• lib/ — pure functions, types, utils
• stores/ — Zustand stores
• services/ — Supabase client, API calls
• supabase/migrations/ — Supabase migrations (apply via CLI)

Coding rules: see .cursor/rules/ — mandatory for all implementation.