karpathy-guidelines

To reduce LLM coding mistakes, apply strict behavioral guidelines that enforce stating assumptions, planning step-by-step, and making minimal surgical changes. This surfaces hidden tradeoffs and prevents overcomplication by requiring verifiable success criteria for every task.

Best practices for LLM code review involve checking for minimal surgical changes, verifying measurable outcomes, and surfacing hidden assumptions. Goal-driven execution ensures tasks transform into measurable goals, looping on verification until defined success criteria are met.

Yes, you can use these guidelines to prevent unnecessary abstractions by enforcing a simplicity-first rule. They require the minimal, robust solution for coding tasks, explicitly avoiding overcomplication and collateral edits to preserve surrounding context.

You need to define verifiable success criteria whenever executing LLM coding tasks to ensure goal-driven execution. Transforming tasks into measurable goals allows you to loop on verification until criteria are met, preventing complex, untraceable code generation.

Unintended collateral edits occur when LLMs lack strict surgical change guidelines. Applying behavioral rules ensures the model modifies only what is needed, preserves surrounding context, and avoids unnecessary abstractions during refactoring or writing code.