karpathy-guidelines

To prevent common LLM coding mistakes during refactoring, apply surgical guidelines that enforce minimal changes, surface explicit assumptions, and define verifiable success criteria before executing modifications. This approach avoids overcomplication and keeps changes strictly targeted.

Surgical coding guidelines are behavioral rules that ensure LLMs touch only necessary code, state explicit tradeoffs before writing, and provide a verifiable plan. They reduce overcomplication by forcing the model to define goals and assumptions upfront before generating changes.

Review LLM-generated code for overcomplication by checking if the changes are minimal and surgical, verifying that assumptions are explicitly stated, and ensuring the implementation matches a predetermined plan with measurable success criteria.

Yes, you can use these coding guidelines for any LLM refactoring task regardless of platform. They are behavioral guidelines applied to writing or reviewing code, requiring no specific dependencies or environment setup to implement within your prompts.

The best way to define verifiable success criteria for LLM code changes is to specify clear goals and state explicit assumptions upfront. This provides a measurable plan that guides the LLM to execute surgical changes without altering unrelated code.

LLM generated code alters unrelated functions during a refactor because it lacks explicit surgical constraints. By applying behavioral guidelines that enforce minimal changes and require verifiable success criteria, you prevent the model from overcomplicating the output.