karpathy-guidelines

To reduce LLM coding mistakes during refactoring, apply surgical coding guidelines that enforce explicit assumptions, plan-first workflows, minimal targeted edits, and measurable success criteria before code generation begins.

Best practices for reviewing LLM generated code involve enforcing surgical edits that affect only the intended area, surfacing hidden assumptions, and verifying that measurable success criteria are met without introducing collateral changes.

You can enforce surgical code changes by instructing the AI to apply targeted editing guidelines, ensuring modifications affect only the intended area, avoiding collateral changes, and explicitly recording assumptions before code execution.

Surfacing explicit assumptions before writing code prevents overcomplication and reduces errors by forcing developers to define verifiable success criteria and plan-first workflows, ensuring clarity, safety, and minimal changes.

Yes, you can integrate these coding guidelines into code review prompts for LLM-powered development, applying structured guidance to refactoring tasks and reviews to improve correctness, reliability, and risk minimization across software tasks.

You should reconsider surgical minimal changes when a refactoring task requires systemic architectural updates, as these guidelines target targeted edits to reduce collateral changes rather than broad structural overhauls.