What problem does it solve?
LLMs frequently make common coding mistakes such as overcomplicating solutions, making unnecessary edits to existing code, and failing to define clear success criteria when writing, reviewing, or refactoring code, leading to extra rework and technical debt.
Core Features & Use Cases
- Pre-Coding Guardrails: Requires surfacing assumptions, clarifying ambiguities, and stating tradeoffs before writing any code to avoid misalignment with requirements.
- Simplicity-First Implementation: Enforces writing the minimum viable code that solves the stated problem, with no unrequested features, abstractions, or speculative error handling.
- Surgical Edit Rules: Mandates only modifying code directly related to the user's request, matching existing code style, and only removing unused code created by your own changes.
- Goal-Driven Execution: Requires defining verifiable success criteria for every task, with clear checkpoints for multi-step work to enable independent validation.
- Use Case: Ideal for developers using AI assistants for code generation, PR reviews, bug fixes, and refactoring tasks to reduce costly rework and improve code quality.
Quick Start
Use the karpathy-guidelines skill to review your recent pull request changes and simplify any overcomplicated sections while ensuring all edits directly align with the original request.
