latent-sre/save-toolkit/python-craft

Python craft

Make Python easier to understand, use, and change. Keep clear code when a change offers no benefit. The task and project constraints set scope; this skill adds no tools or authority. Explanation, design, and review requests stop at those outputs unless implementation is requested. For a defect, establish the cause and expected behavior, reproduce it, and verify the repair. Keep small changes scoped; report material adjacent findings separately.

Build or improve

1. Choose the workflow: new code establishes the contract, consumer, and delivery shape;

refactoring maps existing consumers, behavior, and ownership before changing structure. Use both when a refactor introduces a new capability; distinguish its new behavior from the contracts the restructuring preserves.

1. Choose the design or transformation that fits: representations, algorithms, control flow,

interfaces, extraction or inlining. Compare direct code, the standard library, existing dependencies, and maintained packages where relevant.

1. Deliver a complete path through its entrypoint, then cover remaining cases and affected callers.

Stage work by risk; a justified rewrite may replace all in-scope implementation. Preserve supported contracts; separate intentional behavior changes from restructuring and resolve uncertain requirements rather than guessing.

1. Check required behavior through intended entrypoints. For refactoring, show the benefit: one policy

owner, fewer facts callers must coordinate, removed machinery, or an easier real change. Remove obsolete code/dependencies; passing tests or reducing line count alone does not establish improvement.

Load relevant detail

Load matching references only. Run project checks in the intended environment on supported Python versions; report intentional behavior differences, verification, and remaining uncertainty; measure performance claims separately.