Behavioral guidelines to reduce common LLM coding mistakes by making agents surface assumptions, prefer simple implementations, keep edits focused, and define verifiable success criteria.