karpathy-guidelines

To reduce LLM coding mistakes during refactoring, apply behavioral guidelines that enforce simplicity, surgical changes, and verifiable success criteria. This approach minimizes overcomplication and surfaces hidden assumptions.

The best guidelines for LLM-assisted code review require explicitly stating assumptions, surfacing tradeoffs, and defining verifiable success criteria before execution. This ensures the AI loops until the goal is satisfied.

To make LLM code changes surgical, enforce guidelines that prefer minimal, direct modifications and preserve existing style. The AI modifies only what is necessary, avoiding unnecessary abstractions.

Yes, applying LLM coding guidelines works across multiple languages and project sizes. The behavioral rules focus on iterative improvement and risk management rather than language-specific syntax.

Your LLM generates overcomplicated code because it lacks enforced simplicity constraints. By applying structured guidelines, the AI prioritizes minimal, direct changes and must think before coding.

To define verifiable success criteria for LLM coding tasks, enforce goal-driven execution guidelines that require the AI to loop and test iteratively until the explicit success conditions are satisfied.