Meal Planner — Orientation

First stop for any agent resuming work. Read this, then docs/HANDOFF.md for the deep dive.

What this project is

Self-hosted meal planning for one family of 4. Pulls weekly grocery prices from Lucky California (San Pablo, store 757) via the Swiftly JSON API, generates a 7-day meal plan, emails per-member approval links, builds a shopping list grouped by aisle. Replaces meal-kit subscriptions (Blue Apron / Sunbasket / etc.) which marked up ingredients ~3× and produced repetitive meals.

Constraint that drives the design: 3 of 4 members do not like mushrooms; family is calorie/budget conscious; no allergies. Must work for non-technical wife + 2 kids; technical owner self-hosts.

Architecture (current)

email ─► SendGrid (Phase 6, not yet wired) ──┐
web   ─► nginx :80/:443 ─► React/Vite ───────┼─► FastAPI ──► PostgreSQL 15
                                             │      │
                                             │      └─► Swiftly JSON API
                                             │         (prod.swiftlyapi.net)
                                             │
                                             └─► /api/admin/scrape (BackgroundTasks)

scheduler ─► APScheduler (America/Los_Angeles)
              Fri 02:00 scrape → 05:00 generate → 06:00 email
              Fri 17:00 deadline → 18:00 finalize+shopping-list
              (same backend image; command: python -m app.scheduler)

backend (FastAPI 0.109, SQLAlchemy 2.0, Alembic) — internal only, expose: 8000
frontend (React 18 + TS + Vite + Tailwind) — internal only via nginx
db (Postgres 15-alpine) — internal only
nginx — sole external entry, ports 80/443

Phase status (2026-05-07)

#	Phase	Status
1	Infra (Docker, FastAPI, React, nginx, Postgres)	Complete
2	DB & models (Alembic, Pydantic schemas, API endpoints)	Complete (real, verified)
3	Lucky California ingestion (Swiftly JSON API)	Complete — 17 categories, ~10k products live
4	Recipe engine (CRUD, search, tagging, never-suggest filter)	Thin slice complete — recipe + ingredient CRUD, ingredient↔grocery match layer (rapidfuzz, manual override), NeverSuggest CRUD, 30-recipe seed. Ingestion source decision deferred (see spec).
5	Meal planner orchestration (generate → email → vote → finalize)	Complete — Friday Pacific cadence; APScheduler container; `weekly_run` state table; idempotent steps; stale-data fallback; configurable deadline policy.
6	SendGrid email integration (proposal/reminder/confirmation)	Stub only — `app/services/email.py::SendGridEmailBackend` raises NotImplementedError
7	Web UI core (Dashboard / Meal Detail / Pantry / Shopping List)	Complete (no auth UI yet)
8	Web UI feedback portal	Not started
9	Meal-planner generation algorithm	Complete — POST /api/admin/meal-plans/generate produces 3-dinner plans against seeded recipes + matched grocery prices. Filter (6 hard constraints), score (5 signals), top-K=20 set enumeration with diversity penalty.
10	Image strategy (scraped + AI fallback)	Not started
11	Polish (variety analysis, budget tracking, APScheduler)	Not started

Verification gate (current, 2026-05-07): 115/115 pytest green; alembic upgrade→downgrade→upgrade clean; frontend npm run build clean; live scrape persists 9,980 grocery_item rows in 44 s with auto-minted JWT (no env var); 29,779 ingredient_grocery_match rows produced; email approval round-trip (approve/deny/single-use) verified end-to-end; scheduler container starts with all 5 Friday jobs registered.

Auth model (R1-B+D, locked in)

Bearer token ADMIN_TOKEN for every /api/admin/* route.
Signed-cookie session (itsdangerous, key=SECRET_KEY) for all NON-GET routes on profile/pantry/recipes/meals/shopping-list. GET reads stay open inside the trusted network.
/api/auth/login accepts {password} matching SESSION_PASSWORD. Sets the cookie.
/api/meals/vote/{item_id}?token=… keeps its per-voter token flow; not session-gated.
Bootstrap hatch: when no family_profile row exists, login signs the literal string "bootstrap" instead of a UUID. First-run convenience only — replace with a real setup gate before any non-trusted exposure.

Database schema highlights

Core tables: family_profile, family_member, recipe, ingredient, meal_plan, meal_plan_item, meal_plan_vote, approval_token, home_pantry, feedback, grocery_item, scrape_log, email_log.

Conventions: UUID PKs everywhere, TIMESTAMPTZ, Postgres ENUMs (with values_callable=lambda obj: [e.value for e in obj] on every SQLEnum — name-mode silently breaks otherwise), ISO day-of-week (1=Mon).

Key relationships: family_profile→family_member; family_member→meal_plan_vote (per-voter); recipe.ingredients JSONB (no recipe-ingredient join table); grocery_item.(source, external_id) is the upsert key for scrape ingestion.

Migrations applied: 0001 initial, 0002 seed (idempotent via ON CONFLICT DO NOTHING), 0003 grocery_item.description, 0004 family_profile.calorie_target, 0005 grocery_item.external_id + source + composite index, 0006 thin Phase 4 (ingredient.aliases, recipe.calories_per_serving, ingredient_grocery_match), 0007 seed canonical ingredients (50) + 30 starter recipes, 0008 Phase 5 (weekly_run table + family_profile.pending_approval_policy).

Full schema: docs/database-schema.md.

Environment variables (verified)

# Database
POSTGRES_PASSWORD=...
DATABASE_URL=postgresql://mealplanner:${POSTGRES_PASSWORD}@db:5432/mealplanner

# Auth
SECRET_KEY=...                    # signs session cookies + approval tokens
ADMIN_TOKEN=...                   # bearer for /api/admin/*
SESSION_PASSWORD=...               # family-shared password for /api/auth/login

# Email
EMAIL_BACKEND=console              # 'console' (default) or 'sendgrid'
SENDGRID_API_KEY=...              # only when EMAIL_BACKEND=sendgrid (Phase 6)

# Phase 5 orchestration
ADMIN_EMAIL=...                    # alert destination; empty = alerts dropped silently
APP_BASE_URL=http://localhost      # base URL for vote links in emails

# Lucky / Swiftly
LUCKY_STORE_ID=757                 # Lucky California — San Pablo
SWIFTLY_API_BASE=https://prod.swiftlyapi.net
SWIFTLY_CATEGORIES_URL=https://luckysupermarkets.com/categories
# (no bearer-token env var — minted on demand by app/services/swiftly_auth.py)

# Other
LUCKY_CA_URL=https://luckysupermarkets.com
AI_IMAGE_ENABLED=false
LOG_LEVEL=INFO

The Swiftly bearer JWT is auto-minted at request time via Firebase REST anon-signUp (backend/app/services/swiftly_auth.py, spec docs/specs/2026-05-06-swiftly-token-auto-mint.md). Process-local cache; ~99% cache-hit rate in steady state. A mint failure surfaces as SwiftlyAuthMintError and lands verbatim in ScrapeLog.error_message.

Verification commands

# Full local stack (include scheduler for Phase 5)
docker compose --env-file .env.test up -d db backend scheduler
docker compose --env-file .env.test exec backend alembic upgrade head
docker compose --env-file .env.test exec -e TEST_DATABASE_URL=postgresql://mealplanner:${POSTGRES_PASSWORD}@db:5432/mealplanner backend pytest -q tests/   # → 115 passed

# Verify scheduler
docker compose --env-file .env.test logs scheduler | grep Registered
# Expected: 5 lines — scrape/generate/email/deadline/finalize

# Manual step trigger (useful for testing without waiting for Friday)
curl -X POST http://localhost:8000/api/admin/orchestrate/scrape \
  -H "Authorization: Bearer ${ADMIN_TOKEN}"

# Frontend
cd frontend && npm ci && npm run build

# CI: .github/workflows/ci.yml runs both jobs on push/PR.

A .env.test template lives in the repo root (gitignored) for local stack runs. Pytest uses TEST_DATABASE_URL; alembic uses DATABASE_URL.

Conventions

Python: Black + isort. TypeScript: Prettier + ESLint.
Conventional commits (feat:, fix:, docs:, refactor:, test:).
API paths: no trailing slash, no /list//planned suffixes.
Tests: pytest; requires_postgres marker auto-skips locally without TEST_DATABASE_URL.
Migrations: Alembic only. Never Base.metadata.create_all() at runtime.
Background work: FastAPI BackgroundTasks for on-demand scrape. APScheduler BlockingScheduler in the scheduler container for the weekly cycle (Phase 5 complete).

Where to look

docs/HANDOFF.md — comprehensive handoff for fresh agents (start here for non-trivial work).
docs/SPEC.md — product spec.
docs/ARCHITECTURE.md — system design.
docs/database-schema.md — full DDL reference.
docs/implementation-plan.md — original phased plan.
docs/RUNNING.md — local dev workflow.
docs/specs/2026-05-05-meal-planner-algorithm-design.md — Phase 9 + thin Phase 4 design.
docs/specs/2026-05-06-swiftly-token-auto-mint.md — Swiftly JWT auto-mint design (Implemented 2026-05-06).
docs/superpowers/plans/2026-05-07-phase5-orchestration.md — Phase 5 implementation plan (executed 2026-05-07).
.agent/plan.md, .agent/context.md, .agent/phase-summaries/ — recovery decisions and per-phase summaries from the R1+R2+R3-0 work.
Review/reviewconcensus.md — the adversarial review that drove the recovery.

Last updated: 2026-05-07 — Phase 5 weekly orchestration shipped; 115/115 pytest green; scheduler container live with 5 Friday Pacific jobs. Next pickup: Phase 6 SendGrid.

9.5 KiB Raw Blame History Unescape Escape