python-expert

What this skill does

# Python Expert

Elite Python 3.13+ expertise for backend development, testing, async programming, and type systems.

## When to Use

- Async/await, asyncio, concurrency patterns
- Type errors or complex annotations
- Writing/debugging tests (pytest, async, mocking)
- Performance optimization
- Security review
- Backend architecture (FastAPI, Django, SQLAlchemy)

## Core Expertise

**Python Mastery**: Decorators, context managers, metaclasses, descriptors, generators, coroutines, data model, GIL internals

**Backend**: FastAPI/Django/Flask, PostgreSQL/Redis/MongoDB, SQLAlchemy/Django ORM, REST/GraphQL/WebSockets/gRPC, OAuth2/JWT, microservices

## Code Standards

- Full type hints, Google/NumPy docstrings, 88-char lines
- PEP 8 naming, SOLID principles, secure by default
- Use f-strings for formatting, focused small functions

## Testing

**pytest**: Use `setup_method`, `pytest.raises`, `@patch` for mocking
**Async**: Use anyio for test fixtures, `AsyncMock` for mocking async functions
**Integration**: In-memory SQLite fixtures with proper cleanup
**All network calls must be mocked**

## Async/Await

- `asyncio.run()` for entry, `TaskGroup` for structured concurrency (preferred over `gather()`)
- `asyncio.timeout()` for timeouts, `Semaphore` for rate limiting
- Handle cancellation with try/finally, use `ExceptionGroup` for multiple errors
- Type: `async def foo() -> T` or `Awaitable[T]`

## Type System

**Modern syntax** (Python 3.10+): `list[str]`, `dict[str, int]`, `str | None`
**Variance**: dict invariant, Mapping covariant—use `Mapping[K, V]` when needed
**Advanced**: `Self` for fluent methods, `ParamSpec` for decorator typing, `TypedDict`

**Minimize `Any`**:
- Use `Protocol` for structural typing instead of `Any`
- Use `TypedDict` for dicts with known structure instead of `dict[str, Any]`
- Document why `Any` is necessary when it must be used

**Common fixes**: Mixed type ops, SQLAlchemy column assignments, API response access
**Atomic processing**: Fix ALL type errors in file with single edit

## Patterns

```python
# Dataclass with slots (memory efficient)
@dataclass(slots=True)
class User:
    name: str
    email: str
    tags: list[str] = field(default_factory=list)
    def __post_init__(self):
        if not self.name: raise ValueError("Name required")

# Pattern matching (3.10+)
match response.status:
    case 200: return response.json()
    case 404: raise NotFoundError()
    case _: raise APIError(response.status)
```

**Prefer**: Dependency injection over singletons, `@cache` for memoized instances

## Security

- Validate/sanitize all inputs, parameterized SQL queries only
- Rate limiting, CORS/CSRF protection, secure sessions
- Avoid dynamic code evaluation and unsafe serialization with untrusted data

**Cryptography**:
- Forbidden: MD5, SHA-1, DES/3DES, RC4, custom crypto
- Required: SHA-256+ for hashing, AES-256-GCM for encryption, Argon2/scrypt for passwords
- Use `secrets` module for tokens, `cryptography` package for crypto operations

## Performance

- Profile first (cProfile, timeit), optimize real bottlenecks
- Sets for O(1) lookup, deque for queues, Counter for counting
- Generators for large data, `__slots__` for memory
- `@cache` (unbounded) or `@lru_cache` (bounded) for memoization
- Eager loading (N+1), connection pooling, async I/O

## Pitfalls

```python
# Mutable defaults: use None, then check identity
def f(items=None):
    if items is None:
        items = []  # Don't use `or []` - empty list is falsy!
    return items

# Late binding: capture with default arg
funcs = [lambda x=i: x for i in range(3)]
```

**Avoid**: God classes, spaghetti code, magic numbers, copy-paste, bare `except:`

## Error Handling

Custom exception hierarchies, structured JSON logging, circuit breakers, retry with backoff, graceful degradation

## Tooling

```bash
ruff check .    # lint
ruff format .   # format
pyright .       # typecheck
```

**Stack**: uv, httpx/aiohttp/anyio, pydantic

## Cleanup

Remove before completion: `debug-*.py`, `test-*.py`, `__pycache__/`, `*.pyc`, `*_REPORT.md`