Agent Skills: Python Expert — Gotchas & Decisions

AI-consumed reference. Optimized for Claude to read during execution. Human-readable explanation: see docs/architecture/HIERARCHICAL_PLANNING.md or docs/getting-started/ depending on topic.

Python Expert — Gotchas & Decisions

Use Context7 for FastAPI/Django/Flask docs.

Key Decisions

decisions[4]{choice,use_when}:
  FastAPI vs Django vs Flask,"FastAPI: async APIs + auto-docs. Django: full-featured + ORM + admin. Flask: minimal/micro"
  Pydantic vs dataclass,"Pydantic for validation/serialization (API boundaries). dataclass for internal structs"
  SQLAlchemy vs Django ORM,"SQLAlchemy: standalone/FastAPI. Django ORM: Django projects only"
  sync vs async,"async for I/O-bound (HTTP/DB). sync for CPU-bound. Don't mix blocking calls in async"

Gotchas

• Mutable default args: def f(items=[]) shares list across calls — use def f(items=None): items = items or []
• asyncio.run() creates new event loop — can't nest. Use await inside existing async context
• FastAPI Depends(): new instance per request by default. Use @lru_cache for singletons
• Django N+1: use select_related (FK/OneToOne) and prefetch_related (M2M/reverse FK)
• isinstance(x, int) catches bool too — bool is subclass of int. Check type(x) is int if needed
• Type hints are NOT enforced at runtime — use Pydantic or beartype for runtime validation
• dict.get('key') returns None silently — use dict['key'] when key must exist
• requirements.txt vs pyproject.toml: prefer pyproject.toml (PEP 621) for modern projects
• with statement for resource cleanup (files, DB connections) — never rely on __del__