← ClaudeAtlas

occam-pythonlisted

Python guidance on async and cancellation, shared mutable state, data-pipeline memory, time, money and text semantics, typing, packaging, modules, file size, large codebases, names, docstrings and comments, pytest and mocks, moved APIs, security sinks, notebooks. Use alongside occam to write, debug, audit or review Python where these matter.
stephenoffer/occam · ★ 1 · Code & Development · score 77
Install: claude install-skill stephenoffer/occam
# Python Tie choices to the project's supported versions and tools. Inspect `pyproject.toml`, lockfiles, packaging, interpreter versions, CI, and the established formatter, linter and type checker, then reuse the existing environment. Don't migrate tooling or raise the Python version to express a preference, or install a stack that isn't established. ## What goes wrong most These failed in real runs, with or without it: - **New module names.** `random.py`, `email.py` or `logging.py` beside a script shadows the standard library and breaks that script. - **Partial splits.** Asked to break a module up, moving a third isn't done. Say what stayed and why. - **Audits that trust tests.** A test asserting only on a mock can stay green when behavior breaks. Report it. - **Notebooks.** Only a fresh kernel running top to bottom verifies one. ## Style and structure Follow PEP 8, PEP 257 and the repository's public API conventions, sensibly. Avoid unnecessary metaclasses, decorators, dynamic attributes, import tricks, and classes wrapping stateless functions. ## Types Type meaningful interfaces and data boundaries with the project's chosen checker. Avoid both untyped public contracts and ornate generics that add nothing. An annotation isn't input validation or a security check: data crossing a trust boundary still needs validating at runtime. ## Values and state Watch for these common defects: - Mutable default arguments and accidental aliasing. Know whether you need a shallow