karpathy-guidelines

To reduce LLM coding mistakes, enforce surgical, minimal-change guidelines that touch only necessary code and avoid unrelated edits. This requires surfacing assumptions upfront and executing plan-driven changes to maintain transparent decision-making.

The best coding guidelines for LLM code review require writers to surface explicit assumptions, propose minimal changes, and outline a clear success plan with defined verification criteria. This prevents overcomplication and ensures transparent decision-making.

Defining verifiable success criteria for LLM generated code involves establishing measurable acceptance criteria and tests before implementation. This plan-driven execution approach ensures changes can be objectively verified against expected outcomes.

Yes, you need to surface assumptions and tradeoffs before starting LLM coding tasks. Explicitly stating assumptions upfront enforces plan-driven execution and reduces common mistakes caused by overcomplication or unnecessary refactoring.

You should avoid unnecessary refactors when reviewing LLM code to maintain surgical, minimal changes that do not alter unrelated code. Overcomplicating edits introduces errors and obscures transparent decision-making during the verification process.