GIT_FEED

forrestchang/andrej-karpathy-skills

A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.

View on GitHub

What it does

This project provides a single configuration file that makes Claude, Anthropic's AI coding assistant, behave more like a careful senior engineer — asking questions before diving in, keeping code simple, and only touching what it's supposed to touch. It translates AI researcher Andrej Karpathy's widely-shared frustrations about AI coding tools into practical guardrails you can drop into any project in seconds.

Why it matters

As AI coding assistants become standard in software teams, the gap between a well-configured and a poorly-configured AI can mean the difference between shipping clean code and accumulating costly technical debt. With over 61,000 stars, the massive community response signals that silent AI assumptions and bloated AI-generated code are pain points serious enough for builders to actively seek fixes.

Why it's trending

The frustration with AI coding tools doing too much, too fast, and in the wrong places has clearly hit a nerve — this single configuration file pulled in nearly 40,000 stars in one week, suggesting it spread virally through developer communities who recognized the problem immediately. The premise is simple: Andrej Karpathy's public complaints about LLM coding behavior, translated into a file you can drop into any project to make Claude act more like a careful collaborator than an overeager intern. With over 3,600 new forks this week, builders aren't just starring it to bookmark it later — they're actively pulling it into their own workflows right now.

32Active

On the radar — signal detected

Stars
76.4k
Forks
7.1k
Contributors
5

Score updated Apr 23, 2026

Related projects

Multica is an open-source platform that lets teams assign tasks to AI coding agents the same way they'd assign work to a human teammate — the agents pick up the work, write code, flag problems, and update the project board on their own. It acts as a coordination layer between your team and multiple popular AI coding tools, so you can manage AI workers alongside human ones without being glued to the screen.

// why it matters As AI coding tools become standard, the bottleneck is shifting from 'can AI write code' to 'how do you manage and trust AI workers at scale' — Multica is building the workflow infrastructure for that next phase. Being vendor-neutral and self-hosted means teams avoid lock-in while getting a repeatable, auditable system for delegating real work to AI, which is a foundational capability for any software company trying to build faster with smaller teams.

TypeScript18.9k stars2.3k forks31 contrib

Web Platform Tests (WPT) is a massive shared test suite that checks whether all major web browsers — Chrome, Firefox, Safari, Edge, and others — behave consistently when displaying websites and web apps. Think of it as a universal quality checklist that browser makers run to confirm their software follows the agreed-upon rules of how the web should work.

// why it matters When browsers behave differently, developers must build workarounds that add cost and slow down shipping — WPT is the industry's shared mechanism for reducing that friction, making the web a more reliable platform for products to run on. For builders, broader browser consistency means less money spent on cross-browser bug fixes and greater confidence that web-based products will reach users as intended, regardless of what device or browser they use.

HTML5.9k stars3.8k forks3245 contrib

LLVM is the foundational technology that turns code written by developers into programs that computers can actually run, and it's used to build compilers for languages like C, C++, and many others. Think of it as the invisible translation engine that sits behind many of the world's most popular programming languages and development tools.

// why it matters Nearly every major tech company — Apple, Google, Meta, and more — relies on LLVM to power their software development pipelines, meaning it underpins the tools that build most modern apps and systems. For builders, understanding LLVM matters because it's the infrastructure behind emerging programming languages and performance-critical software, making it a key leverage point for teams building developer tools, new languages, or platform technologies.

LLVM38.0k stars16.9k forks8791 contrib

The Supabase CLI is a command-line tool that lets developers manage their Supabase projects — an open-source alternative to Google Firebase — directly from their computer, including setting up local development environments, managing database changes, and deploying serverless functions. It essentially gives builders a fast, scriptable way to control their entire backend infrastructure without touching a web dashboard.

// why it matters As more startups choose Supabase over Firebase or custom backends to move faster, having a robust CLI means entire backend workflows can be automated, version-controlled, and reproduced — reducing errors and speeding up shipping. With nearly 2,000 stars and 163 contributors, this is a well-adopted tool in a growing ecosystem, signaling strong developer momentum behind Supabase as a serious Firebase competitor.

Go2.2k stars436 forks169 contrib
// SUBSCRIBE

The repos that moved this week, why they matter, and what to watch next. One email. No noise.