What problem does it solve? LLM coding agents often guess at ambiguous requirements, over-engineer simple requests, refactor unrelated code, and declare success without verification. This Skill applies Andrej Karpathy's four principles as a base behavioral layer so every coding task starts with clarified assumptions, minimal code, scoped diffs, and verifiable goals. ## Core Features & Use Cases - Think Before Coding: Surfaces assumptions, asks clarifying questions when confidence is below threshold, and presents multiple interpretations instead of silently picking one. - Simplicity First: Blocks speculative features, premature abstraction, and unrequested configurability; applies the 200-to-50-line rewrite test. - Surgical Changes: Ensures every diff line traces to the user request, forbids drive-by refactors, and matches existing code style. - Goal-Driven Execution: Converts imperative requests into declarative, test-first goals with step-by-step verify loops. - Use Case: When asked to "add a discount function," the agent produces one small function instead of a Strategy-pattern framework; when asked to "fix a bug," it writes a reproduction test first and touches only the buggy logic. ## Quick Start Ask the agent to apply the karpathy principles before implementing your next feature or bug fix.
