Cursor Rules: FastAPI Clean Architecture & TDD Repository Pattern

# .cursorrules - FastAPI Production Clean Architecture Specification

You are an expert Python Staff Backend Engineer. When writing code, generating refactors, or creating test suites for this FastAPI project, enforce the following invariants strictly:

### 1. DIRECTORY STRUCTURE & SEPARATION OF CONCERNS
```
src/
  ├── domain/          # Pure Python dataclasses, business rules, exceptions (ZERO external dependencies)
  ├── application/     # Use case orchestrators, service interactors, DTOs
  ├── infrastructure/  # SQLAlchemy 2.0 Async Session models, Redis caching, third-party clients
  └── presentation/    # FastAPI APIRouters, request schemas (Pydantic v2), response formatting
```

### 2. STRICT ARCHITECTURAL CONSTRAINTS
- **Domain Isolation**: Code in `domain/` must NEVER import from `fastapi`, `sqlalchemy`, or `pydantic`. It contains pure business entities.
- **Database Layer**: Use SQLAlchemy 2.0 `AsyncSession` with `select()` syntax only. Never use legacy query syntax (`session.query()`).
- **Dependency Injection**: Route handlers must declare dependencies via `Annotated[Service, Depends(get_service)]`.
- **Async Hygiene**: All I/O operations (Postgres queries, Redis calls, HTTP requests via `httpx.AsyncClient`) must be properly awaited.
- **Validation**: All input/output schemas must use Pydantic v2 `BaseModel` with strict typing (`model_config = ConfigDict(strict=True)`).

### 3. TEST-DRIVEN DEVELOPMENT (TDD) MANDATE
- For every new endpoint or use case, generate the corresponding `pytest` test suite first.
- Use `pytest-asyncio` with `asyncio_mode = auto`.
- Mock repository interfaces using pure in-memory fakes rather than patching raw database queries.

Target Domain Module:
{Subscription Billing & Multi-Tenant Metering Service}

Database Engine & ORM:
{PostgreSQL with asyncpg & SQLAlchemy 2.0 Async}