karpathy-guidelines

Behavioral coding guidelines reduce common LLM coding mistakes by enforcing surgical changes, surfacing assumptions, and defining verifiable success criteria during code generation and refactoring tasks.

You can prevent LLM overcomplication by consulting guidelines that prompt the model to ask clarifying questions before coding and strictly avoid unnecessary refactoring during implementation.

Apply code review guidelines by checking for surfaced assumptions, verifying that changes are surgical, and ensuring verifiable success criteria and tests are defined to guide the review process.

Yes, these coding guidelines are designed to be generalizable, safe, and easy to integrate into standard workflows with minimal friction, requiring no additional dependencies or specific environment setup.

You should reconsider using these guidelines if your development task requires broad architectural refactoring rather than surgical changes, or if verifiable success criteria cannot be clearly defined beforehand.