@hd-agent-skills/andrej-karpathy-skills
v1.0.0
Published
Behavioral guidelines to reduce common LLM coding mistakes, packaged as a Claude Code plugin. Derived from Andrej Karpathy's observations on LLM coding pitfalls.
Maintainers
Readme
andrej-karpathy-skills
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls.
Ships the karpathy-guidelines skill, covering four principles:
| Principle | Addresses | |-----------|-----------| | Think Before Coding | Wrong assumptions, hidden confusion, missing tradeoffs | | Simplicity First | Overcomplication, bloated abstractions | | Surgical Changes | Orthogonal edits, touching code you shouldn't | | Goal-Driven Execution | Verifiable success criteria, test-first loops |
See skills/karpathy-guidelines/SKILL.md for the full rules.
Install
Option A — Claude Code marketplace
From the parent agent_skills repo:
/plugin marketplace add <git-url-of-agent_skills>
/plugin install andrej-karpathy-skillsOption B — npm
npm install @hd-agent-skills/andrej-karpathy-skills
# then register the installed directory as a local marketplace:
/plugin marketplace add ./node_modules/@hd-agent-skills/andrej-karpathy-skills
/plugin install andrej-karpathy-skillsThe postinstall script prints the exact marketplace add path to use.
Credits & References
This plugin is not original work — it repackages an existing skill for distribution through this marketplace.
- Primary source referenced for this import: multica-ai/andrej-karpathy-skills (commit on
main, imported 2026-04-19). - Original author: forrestchang — forrestchang/andrej-karpathy-skills.
- Inspiration: Andrej Karpathy's X post on LLM coding pitfalls.
- License: MIT (see LICENSE — original copyright preserved).
If upstream updates the guidelines, re-sync this plugin from the source above rather than editing in place.
