> ## Content Index > Fetch the complete content index at: https://www.notatechguy.com/llms.txt > Use this file to discover other available public pages before exploring further. # Karpathy-inspired CLAUDE.md file surges 333 stars in a day on GitHub - URL: https://www.notatechguy.com/karpathy-inspired-claude-md-file-surges-333-stars-in-a-day-on-github/ - Published: 2026-09-12T21:36:18.000Z - Updated: 2026-09-12T21:36:18.000Z - Description: The multica-ai/andrej-karpathy-skills repository distills Karpathy's warnings about LLM overcomplication and hidden assumptions into four principles - Author: Marcello Babbili - Tags: Technology & AI, Anthropic A single markdown file distilling Andrej Karpathy's blunt critiques of how large language models write code has gained 333 stars in 24 hours on GitHub, pushing the multica-ai/andrej-karpathy-skills repository past 196,000 total stars [P²](https://github.com/forrestchang/andrej-karpathy-skills?ref=notatechguy.com). The file, CLAUDE.md, is not code — it is a behavioural contract users drop into their projects to steer Anthropic's Claude Code agent toward simpler and more verifiable changes [S¹](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com). **My read:** This is the first time I've seen a community treat a prompt file as a shareable, version-controlled artifact the way they treat a library. The velocity suggests developers are hungry for guardrails that don't require fine-tuning — but the star count also reflects Karpathy's personal gravity, not proof that the guidelines work. I'd want to see a controlled comparison before adopting it wholesale. The repository's README quotes Karpathy describing models that "make wrong assumptions on your behalf and just run along with them without checking" and "really like to overcomplicate code and APIs, bloat abstractions, don't clean up dead code" [S¹](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com). Those quotes are attributed to "Andrej's post" without a link; the original source has not been independently verified [S¹](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com). The maintainer, who goes by jiayuan\_jy on X, maps each complaint to a principle: Think Before Coding, Simplicity First, Surgical Changes, and Goal-Driven Execution [S¹](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com). Claude Code itself — Anthropic's agentic terminal tool that reads, writes and runs code across a repository, has accumulated 143,000 stars [P³](https://github.com/anthropics/claude-code?ref=notatechguy.com). Developers have been swapping prompt files privately for months; as we found when [a new system spawned Claude Code agents with pre-loaded memory](https://www.notatechguy.com/new-system-spawns-claude-code-agents-with-pre-loaded-memory/), the bottleneck is often the instructions, not the model. The karpathy-skills file formalises that layer into a single MIT-licensed skill definition [P⁴](https://github.com/multica-ai/andrej-karpathy-skills/blob/HEAD/skills/karpathy-guidelines/SKILL.md?ref=notatechguy.com). The maintainer also points to Multica, an open-source platform for running and managing coding agents with reusable skills [S¹](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com). That project is still early, but it signals a shift: prompt engineering is becoming infrastructure. The next checkpoint is whether Anthropic adopts a native skill registry for Claude Code — something the company has not announced but which would render community files like this one either obsolete or first-class citizens. For now, any team using Claude Code can drop the CLAUDE.md into their repo and test whether the four principles cut the over-engineering Karpathy described. The file is 200 lines; the experiment takes minutes. --- *Sources: [S1 — multica-ai/andrej-karpathy-skills: A single CLAUDE.md file to improve ](https://github.com/multica-ai/andrej-karpathy-skills?ref=notatechguy.com) · [P2 — multica-ai/andrej-karpathy-skills](https://github.com/forrestchang/andrej-karpathy-skills?ref=notatechguy.com) · [P3 — anthropics/claude-code](https://github.com/anthropics/claude-code?ref=notatechguy.com) · [P4 — skills/karpathy-guidelines/SKILL.md](https://github.com/multica-ai/andrej-karpathy-skills/blob/HEAD/skills/karpathy-guidelines/SKILL.md?ref=notatechguy.com) · [P5 — Andrej](https://github.com/karpathy?ref=notatechguy.com)* ## Related reading - [New system spawns Claude Code agents with pre-loaded memory](https://www.notatechguy.com/new-system-spawns-claude-code-agents-with-pre-loaded-memory/) — our technology desk, 2026-08-29 - [LoopX adds durable memory to Claude Code and Codex agents](https://www.notatechguy.com/loopx-adds-durable-memory-to-claude-code-and-codex-agents/) — our technology desk, 2026-09-05 - [Claude Code v2.1.239 adds cost tracking, hits 142k stars](https://www.notatechguy.com/claude-code-v2-1-239-adds-cost-tracking-hits-142k-stars/) — our technology desk, 2026-08-22 --- *Written from 5 sourced items, 4 of them primary.* ![GitHub stars for multica-ai/andrej-karpathy-skills and anthropics/claude-code](https://storage.ghost.io/c/6e/89/6e896869-22ef-4281-a213-b4c462c17cff/content/images/2026/09/chart_4fb01ec244d6bc4ebb33.png)