A generic CLAUDE.md with coding methodology.

CLAUDE.md

Methodology

Code like a hacker: concisely, with self-doubt, without fluff, without repeating yourself, keeping code as orthogonal as possible.

DRY

• Repeating oneself is unacceptable.
• If your LOCs look suspiciously similar, consider a loop or a lambda.
• If you can refactor to follow a "data-driven" approach (e.g. list of dicts instead of ad-hoc code), consider doing so.
• Don't be afraid of using tiny, local abstractions.

The Cost of Code

• Every line of code has a cost.
• Code, by itself, has no value.
• Code that has been verified abundantly, through unit and integration tests, might have some value.

Breaking Things That Work

• If you're afraid of breaking something that currently works, it's probably too brittle, so you should break it.

Functions and Variables

• Functions are your friends. Top-level, lambdas. Use and abuse.
• Purity is good. Testability is good.
• A one-line function is better than repeating a complex line twice.

Comments

• Code should be self-documenting using clear variable names and function names.
• If you need comments to explain the "what" or "how", that's usually a red flag.
• Comments are there for "business" logic / field-specific knowledge.

Classes

• Prefer naked pure functions over classes whenever possible.
• Only use stateful constructs if it is clearly the idiomatic, sensible thing to do.

Workflow

• Plan out API -> Write a test -> Write implementation -> Lint and run test -> Assess
• Start with something basic, get it to work, refactor early, often, aggressively.

Language-Specific

• When coding in Python, use ruff for linting, uv for virtualenv management.