What problem does it solve?
This Skill reduces common LLM coding failures by enforcing defensive practices that prevent speculative complexity, minimize unnecessary diffs, and make outcomes verifiable through explicit success criteria.
Core Features & Use Cases
- Clarify before implementing: Surface assumptions, ask questions when ambiguous, and present tradeoffs when multiple interpretations exist.
- Simplicity-first implementation: Prefer minimal code that solves the requested problem without future-proofing or unnecessary abstractions.
- Surgical edits only: Modify only what the user requested, avoid changing adjacent code, and remove only orphaned code introduced by your changes.
- Goal-driven verification loop: Convert vague tasks into verifiable targets and iterate with a plan-act-verify structure until checks pass.
Use it when generating, reviewing, or refactoring code to ensure the result is maintainable, targeted, and testable rather than over-engineered.
Quick Start
Use the karpathy-guidelines skill when you need to implement or refactor a feature and want the AI to ask clarifying questions first, keep the diff surgical, and define concrete success checks before finalizing code.
