agentic-karpathy

To prevent LLM coding mistakes, enforce behavioral guardrails like reading existing code before writing, defining success criteria, and making surgical changes scoped to the request. This ensures modifications remain controlled.

Hard gates in code review are strict rules enforcing read-before-write, success definition, and scope restriction. They prevent unverified assumptions and ensure modifications are surgical and limited to the requested changes.

Enforce scope discipline by applying behavioral guardrails that restrict changes to the explicit request, verify assumptions, and mandate reading code before writing. This limits modifications to surgical, necessary updates.

Applying read-before-write rules improves code writing safety by forcing verification of existing code and assumptions before implementation. This prevents unintended side effects and limits changes to surgical modifications.

Use surgical change guardrails before implementing features, during code reviews, and when modifying code. They enforce reading before writing, verify assumptions, and restrict scope to prevent common LLM coding mistakes.