karpathy-guidelines
The karpathy-guidelines skill provides a set of behavioral instructions for LLMs to minimize coding errors. It emphasizes simplicity, surgical code changes, explicit assumption surfacing, and goal-driven execution.
Is karpathy-guidelines safe to install?
Safe to install: our audit of karpathy-guidelines's source files found 0 shell commands, 0 external URLs, no file writes (none risk). Every command and URL listed appears verbatim in the skill's source. The skill consists of text-based behavioral guidelines and does not execute code or perform network requests.
How we audit skills: our security review methodology.
Who is this skill for?
Developers and AI agents performing coding, code review, or refactoring tasks.
What can you do with it?
- Reducing overcomplication in code generation
- Reviewing code for unnecessary abstractions
- Refactoring existing codebases with minimal impact
- Defining verifiable success criteria for coding tasks
How good is this skill?
Quality score: 10/10. The skill is well-documented and provides clear, actionable instructions for AI agents.
What does the skill file contain?
# Karpathy Guidelines Behavioral guidelines to reduce common LLM coding mistakes, derived from [Andrej Karpathy's observations](https://x.com/karpathy/status/2015883857489522876) on LLM coding pitfalls. **Tradeoff:** These guidelines bias toward caution over speed. For trivial tasks, use judgment. ## 1. Think Before Coding **Don't assume. Don't hide confusion. Surface tradeoffs.** Before implementing: - State your assumptions explicitly. If uncertain, ask. - If multiple interpretations exist, present them - don't pick silently. - If a simpler approach exists, say so. Push back when warran...
Frequently asked questions
Does this skill automatically refactor my code?
No. It provides guidelines for an agent to follow while performing tasks like refactoring.
Does this skill require external API access?
No. The skill is a set of static instructions.
How does this skill handle existing dead code?
The guidelines instruct the agent to mention unrelated dead code rather than deleting it.
Related skills
convex-quickstart
78.3KDevelopers and AI agents building applications with Convex who need to initialize projects or integrate Convex into existing codebases
This skill automates the setup and configuration of Convex in new or existing web applications. It handles project scaffolding, environment variable generation, provider wiring, and local deployment provisioning.
git-commit
37.6KDevelopers who want to automate the creation of semantic git commit messages based on their local code changes
The git-commit skill generates and executes standardized git commits following the Conventional Commits specification. It analyzes diffs to determine commit types, scopes, and descriptions, and manages file staging.
golang-code-style
33.3KGo developers and AI coding agents performing code reviews or writing new Go code
This skill provides a comprehensive set of Go code style conventions focused on clarity, maintainability, and idiomatic patterns. It covers variable declarations, control flow, function design, file organization, and string handling, serving as a guide for human-judgment-based code reviews that complement automated linters.
vue-best-practices
24.1KVue
This skill provides a structured workflow for Vue.js development, mandating the use of the Composition API, TypeScript, and specific architectural patterns. It requires developers to plan component boundaries, follow strict SFC structure, and apply core reactivity and data flow principles before implementing features.