What are the Karpathy Guidelines for AI coding?

These are behavioral rules derived from Andrej Karpathy's observations that prevent LLMs from over-complicating tasks, making unnecessary changes, or assuming user intent without clarification.

How does it handle complex, multi-step tasks?

It uses goal-driven execution, requiring the AI to define a plan with verifiable success criteria (like passing tests) for each step before proceeding.

How does this skill prevent over-engineering?

It enforces a 'Simplicity First' rule, preventing the AI from adding speculative features, single-use abstractions, or unrequested configurability.

Will this skill refactor my entire file?

No. One of the core pillars is 'Surgical Changes,' which instructs the AI to touch only what is necessary and match the existing style without 'improving' adjacent code.

Karpathy Coding Guidelines

Name: Karpathy Coding Guidelines
Author: riseshia

byriseshia

•

生产力与工作流

Implements Andrej Karpathy's principles to reduce LLM coding mistakes through simplicity, surgical edits, and verifiable goals.

The Karpathy Guidelines skill transforms Claude into a disciplined and efficient coding assistant by applying best practices derived from Andrej Karpathy’s observations on AI coding pitfalls. It prioritizes caution over speed, ensuring that every change is surgical and every assumption is surfaced. By enforcing a 'simplicity first' mindset and requiring verifiable success criteria, this skill prevents common issues like over-engineering, unnecessary refactors, and speculative abstractions. It is ideal for developers who value clean, maintainable code and want to minimize the 'code churn' often associated with AI-generated contributions.

主要功能

013 GitHub stars

02Strict 'Simplicity First' logic to prevent over-engineering and bloat

03Explicit assumption surfacing to resolve ambiguity before coding

04Goal-driven execution with defined verification steps for every task

05Surgical code modifications that touch only what is necessary

06Automatic removal of orphans and unused code created during edits

使用场景

01Implementing minimalist features that require zero additional dependencies or abstractions

02Standardizing AI behavior during code reviews to ensure consistency with existing styles

03Refactoring critical legacy code where side effects must be strictly minimized

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add riseshia/dotfiles karpathy-guidelines

For use in Claude.ai and ChatGPT

主要功能

013 GitHub stars

02Strict 'Simplicity First' logic to prevent over-engineering and bloat

03Explicit assumption surfacing to resolve ambiguity before coding

04Goal-driven execution with defined verification steps for every task

05Surgical code modifications that touch only what is necessary

06Automatic removal of orphans and unused code created during edits

使用场景

01Implementing minimalist features that require zero additional dependencies or abstractions

02Standardizing AI behavior during code reviews to ensure consistency with existing styles

03Refactoring critical legacy code where side effects must be strictly minimized