multica-ai / andrej-karpathy-skills

2026-01-27 · Source: Github Trending: All languages · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

A `CLAUDE.md` file, inspired by Andrej Karpathy's observations on large language model (LLM) coding pitfalls, provides four core principles to improve Claude Code's behavior. These principles address common issues such as LLMs making wrong assumptions, overcomplicating code, performing orthogonal edits, and lacking verifiable success criteria. The solution outlines "Think Before Coding" to surface assumptions and confusion, "Simplicity First" to prevent overengineering, "Surgical Changes" to limit modifications to only what is necessary, and "Goal-Driven Execution" to define clear success criteria and enable iterative verification. The guidelines can be installed as a Claude Code plugin or integrated per-project via a `CLAUDE.md` file, and also apply to Cursor projects.

Key takeaway

For AI Engineers and Machine Learning Engineers developing with Claude Code or similar LLM agents, adopting these Karpathy-inspired guidelines can significantly reduce costly coding mistakes and improve code quality. You should prioritize defining explicit success criteria and encourage agents to surface assumptions and simplify code. This approach biases toward caution, leading to cleaner, more focused code changes and fewer rewrites, especially for non-trivial tasks.

Key insights

LLMs excel at meeting specific goals; define clear success criteria for optimal coding agent performance.

Principles

• Explicitly state assumptions and clarify confusion.
• Prioritize minimum viable code over speculative features.
• Limit code changes strictly to the task at hand.

Method

Transform imperative coding tasks into verifiable, declarative goals with explicit success criteria. This enables LLMs to loop independently until criteria are met, reducing constant clarification needs.

In practice

• Install as a Claude Code plugin for project-wide application.
• Integrate into `CLAUDE.md` for project-specific guidelines.
• Use Cursor's project rules for consistent application.

Topics

• LLM Coding Pitfalls
• Claude Code Guidelines
• Andrej Karpathy's Observations
• Think Before Coding
• Simplicity First

Code references

• multica-ai/multica
• multica-ai/andrej-karpathy-skills

Best for: AI Engineer, Machine Learning Engineer, Prompt Engineer

Related on AIssential

• forrestchang / andrej-karpathy-skills — Explicit guidelines can significantly improve LLM code generation by addressing common pitfalls like overcomplication and hidden assumptions.
• Stop Just Vibe Coding: The Karpathy Teardown of AI Agents — LLMs fail in production due to wrong assumptions, scope creep, over-engineering, and weak success criteria.
• Turning Andrej Karpathy’s LLM Coding Thoughts into Claude.md — LLMs are transforming coding workflows, shifting developers from direct coding to guiding and reviewing AI agents.
• Claude Opus 4.8: not ambitious enough? — The model's output, while functional, lacked ambition and failed to push agentic coding boundaries despite explicit prompting.
• Claude Code + Karpathy's NEW Self-Evolving System = 10x Code Generation — AI agents can build and maintain self-evolving knowledge systems using structured wikis and schema rules.
• 60,000 Developers Are Using a Markdown File to Fix How AI Writes Code — Andrej Karpathy, former head of AI at Tesla and OpenAI co-founder, rapidly transitioned to 80% AI agent coding, describing it as the "biggest change to his basic coding workflow in two decades." Howev…

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Github Trending: All languages.