karpathy-guidelines

To prevent LLM coding mistakes during refactoring, enforce assumption clarity, simplicity, and surgical edits. This ensures minimal code changes that directly map to your request, avoiding overengineering and accidental orphaning of variables or imports.

You constrain an AI code assistant to minimal code changes by applying surgical edit guidelines. This limits modifications to only what is necessary for the requested problem, preventing unrelated refactors and speculative abstractions.

Goal-driven execution for AI generated code defines explicit success criteria and loops by checking measurable outcomes, such as tests passing. This verifies each step of the software change against the requested problem.

An AI coding assistant overengineers solutions due to ambiguous interpretations and insufficient verification. Enforcing clearer thinking and simpler implementations with minimal code directly solves the task without speculative abstractions.

You verify success criteria after an AI code refactoring by defining a concrete verification step, such as checking tests passing. This goal-driven approach ensures each modification directly maps to the user's request.

Yes, you can use surgical edit guidelines for code review to limit edits to what is necessary. This prevents unrelated refactors and ensures each modified line directly maps to the requested behavioral change.