andrej-karpathy-skills

To prevent AI overengineering, apply coding guidelines that enforce cautious assumptions and minimal implementations. This approach limits modifications to surgical edits, ensuring changes remain strictly goal-oriented and avoid speculative features.

Effective LLM-assisted refactoring guidelines prioritize simplicity, clear tradeoff analysis, and verification-driven execution. They require the AI to state assumptions upfront and perform surgical edits, directly addressing the task without introducing unnecessary modifications.

You can enforce clear tradeoff analysis by instructing the AI to follow coding discipline that surfaces assumptions and ambiguity before implementation. This ensures the agent thinks through tradeoffs and establishes explicit success criteria prior to coding.

Yes, these coding guidelines are ideal for sensitive software engineering tasks including instruction merging and refactoring. They satisfy verification-driven execution by encouraging explicit success criteria and surgical edit discipline for precise code changes.

Surgical edit discipline in LLM-assisted development limits modifications to exactly what a task requires. It prevents speculative features by ensuring the AI cleans up only what it creates, maintaining simplicity and precise code changes.