Fastapi patterns

Skill ai-enhanced-engineer/aiee-skills/skills/fastapi-patterns

Modern FastAPI 0.111+ production patterns for Python services — async lifespan, Pydantic v2 schemas, dependency injection, middleware ordering, structured exception handlers, router composition. Use for FastAPI endpoint design, service composition, API contract evolution, and migrating from `@app.on_event` / Pydantic v1.From its SKILL.md

Install

npx -y skills add ai-enhanced-engineer/aiee-skills --skill fastapi-patterns

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, as published. Nobody here has run it

FastAPI Production Patterns

Modern FastAPI 0.111+ patterns for production microservices. This skill covers the patterns that change across versions and trip up teams: lifespan replacing @app.on_event, Pydantic v2 schema patterns, Depends() caching/scoping, middleware LIFO order, and structured error handling.

When to use

Designing or refactoring a FastAPI service (new endpoints, lifespan resources, DI graph)
Migrating from @app.on_event → lifespan or Pydantic v1 → v2
Debugging "CORS missing on 500 errors" or "blocking async endpoint" issues
Evolving API contracts (versioning, OpenAPI customization, response schemas)

For modern Python idioms throughout, see arch-python-modern. For service-layer separation (app/routers/ ↔ app/services/ ↔ app/crud/), see arch-ddd.

Core patterns

Pattern	One-line rule
Lifespan	`@asynccontextmanager async def lifespan(app)` — startup before `yield`, shutdown after; reverse-order resource teardown
`Depends()`	Cached once per request; use `Annotated[T, Depends(get_x)]` aliases at module level
Pydantic v2	`model_config = ConfigDict(...)`; `@field_validator`/`@model_validator`; `model_dump()` not `.dict()`
Schema separation	Three classes per resource: `XCreate` (request), `XRead` (response), `XUpdate` (PATCH, all optional)
Middleware order	LIFO — `add_middleware` last = runs first on request. CORSMiddleware sits outermost so OPTIONS preflight reaches it before auth runs and 500 responses still pass through it (otherwise CORS headers are lost — see reference.md)
Exception handlers	Register `@app.exception_handler(Exception)` so errors pass through CORS layer (else 500s lose CORS headers)
Router composition	`APIRouter(prefix=..., tags=..., dependencies=[Depends(Auth())])` — apply auth at router, not per-endpoint
Test overrides	`app.dependency_overrides[get_db] = fake_db`; clearing in fixture teardown prevents cross-test leakage when the app instance persists across tests

Quick reference

# Lifespan (replaces @app.on_event)
@asynccontextmanager
async def lifespan(app: FastAPI):
    create_mongo_client()              # mandatory — let exception abort startup
    await initialize_http_client()
    yield
    shutdown_http_client()              # reverse order
    shutdown_mongo_client()

app = FastAPI(lifespan=lifespan)

# Pydantic v2 schemas
class ProductCreate(BaseModel):
    model_config = ConfigDict(extra="forbid")
    name: str
    sku: str

class ProductRead(BaseModel):
    model_config = ConfigDict(from_attributes=True)
    id: str
    name: str
    sku: str
    created_at: datetime

# Dependency injection (Annotated alias style)
DbSession = Annotated[AsyncSession, Depends(get_db)]

@router.get("/products", response_model=list[ProductRead])
async def list_products(db: DbSession): ...

Anti-patterns (top 4)

Blocking I/O in async def → wrap with await asyncio.to_thread(...) or use httpx.AsyncClient
Mutable default in Depends (tags: list = []) → use None + reassign, or Query(default_factory=list)
Single schema for request + response → split into XCreate / XRead / XUpdate
Missing HTTP client timeouts → httpx.AsyncClient(timeout=httpx.Timeout(connect=2.0, read=5.0))

See reference.md for the full pattern catalogue and examples.md for project-grounded implementations.