karpathy-guidelines

To prevent common LLM coding mistakes during refactoring, apply guidelines that enforce explicit assumption surfacing, write minimal necessary code, and make surgical modifications to avoid premature abstractions and overcomplication.

Code review principles focus on simplicity and explicit assumptions by requiring developers to state tradeoffs clearly, define verifiable success criteria, and ensure all code modifications are surgical and directly address the requested changes.

To stop an AI agent from adding speculative features when writing software, enforce behavioral guidelines that drive goal-oriented execution, advocate for the minimum necessary code, and restrict modifications to only the requested changes.

Applying strict coding guidelines works for both human developers and AI agents by providing a shared behavioral framework that mitigates common mistakes, enforces simplicity, and reduces complexity during software development.

The best way to define verifiable success criteria for code refactoring tasks is to establish goal-driven execution metrics upfront, ensuring the refactored code maintains functionality while strictly reducing complexity and avoiding unnecessary features.

You should avoid deviating from surgical code modifications only when a broader architectural overhaul is explicitly requested; otherwise, strict guidelines require touching only necessary code and cleaning up solely your own modifications to prevent overcomplication.