Top AI Product

Forrest Chang’s andrej-karpathy-skills is the #1 trending repo on GitHub today with 1,955 stars in a single day — and over 220,000 combined across his personal account and the multica-ai organization mirror. It’s a single CLAUDE.md file encoding four behavioral rules derived from Andrej Karpathy’s documented frustrations with LLM coding agents.

## The core rules

“Think Before Coding”: the agent must state assumptions explicitly, present multiple interpretations when a request is ambiguous, and ask rather than guess. “Goal-Driven Execution”: transform imperative instructions into declarative goals with verification loops, so the agent checks its own work instead of charging ahead. The full set targets the specific failure modes — silent assumption, over-eager edits, charging past ambiguity — that make coding agents frustrating in practice.

## How to use it

Merge the guidelines into your existing CLAUDE.md or drop in a new one. Ships for Claude Code, Cursor (CURSOR.md), and as a plugin. No code, no dependencies — just prompt engineering distilled from one of the field’s most credible voices.

## Why it matters

The most valuable thing in agentic coding right now isn’t a bigger model — it’s the prompt scaffolding that makes the model behave. A 220,000-star CLAUDE.md proves the market knows this. The meta-signal worth noting: the highest-leverage AI coding “product” of the week is a text file.