karpathy-coding-guidelines

Reduce LLM coding mistakes by enforcing behavioral guidelines that require explicit assumptions, minimal code, and verifiable success criteria before generating code. This keeps AI-assisted software projects clear and measurable.

Enforce minimalism in LLM code generation by applying surgical guidelines that touch only necessary code and surface trade-offs explicitly. This prevents unnecessary abstractions and keeps changes minimal and documented.

Apply surgical changes to LLM-generated code by keeping unrelated code intact and documenting modifications. This refactoring approach ensures minimum code solves the problem without introducing unintended side effects.

You need explicit success criteria for AI-assisted coding tasks when defining verifiable goals to iterate until met. This goal-driven execution ensures the generated code satisfies requirements and functions correctly.

Yes, you can use these coding guidelines for reviewing prompts and tool calls in LLM projects. They apply directly to coding tasks, code reviews, and refactoring across AI-assisted software projects to enforce clarity.

LLM-generated code includes unnecessary abstractions when simplicity-first guidelines are not enforced. Applying minimalism rules ensures the minimum code required solves the problem without adding extra layers.