Agentic Development Plan

Orchestration

Governing LLM (Claude) plans + reviews. Specialist agents build in parallel where safe.

Agents

  1. Backend Architect agent → Node/Express API, PG schema + migrations, JWT auth, WhatsApp reminder integration (reuse getaseed), validation, rate-limit. Seed + README.
  2. Mobile App Builder agent → Flutter app: theme/tokens, auth, home, customer detail, entry sheet, reminder preview, settings; Dio API client; sqflite offline cache + outbox.
  3. Code Reviewer (via /review) → correctness, security, SQL safety.
  4. Ponytail review → cut over-engineering.

Sequencing

  • Phase 1a: Backend agent builds API against DATABASE.md + API.md (can run first/parallel).
  • Phase 1b: Flutter agent builds UI against UI-UX.md + API.md (mock client first, wire later).
  • Phase 2: Wire app to live API; smoke test.
  • Phase 3: /review + /ponytail-review → fix blockers.
  • Phase 4: Deploy backend (Apache vhost + certbot SSL per global rule), build APK.

Build order checklist → see CHECKLIST.md

Conventions

  • Money = paise (BIGINT) end to end.
  • shop_id always from JWT, never client body.
  • Secrets in .env (root-only), never committed. .env.example checked in.
  • No new deps where stdlib / existing stack suffices (ponytail).