🔥 Hot Repo: Karpathy's 1-File Fix Hits 86K Stars in 12 Days

A single CLAUDE.md file inspired by Andrej Karpathy's observations on LLM coding pitfalls has accumulated 86,000 GitHub stars in 12 days, topping the global trending chart with +44,000 stars in one week.

By OMC Editorial on 2026-04-25

A single 400-word Markdown file has accumulated 86,000 GitHub stars in 12 days, making forrestchang/andrej-karpathy-skills the 1 trending repository globally for the week of April 22, 2026 — with +44,394 new stars in that seven-day window alone. On April 13, the file earned 5,828 stars in a single day, claiming the 2 global spot for new stars on that date. Seven platforms — GitHub, Hacker News, Reddit, X, AI YouTube, Tech YouTube, and Substack — trended it simultaneously. The Origin The file does not come from Andrej Karpathy. Developer Forrest Chang @forrestchanghttps://github.com/forrestchang built andrej-karpathy-skills after Karpathy published a January 26, 2026 X post naming specific failure modes he observed while using LLM coding agents. Karpathy identified three recurring problems: models silently assuming away ambiguity and running with wrong interpretations; models over-complicating solutions with unnecessary abstractions and dead code; and models making collateral edits to code outside the scope of the task. Chang distilled those observations into four directives wrapped in a CLAUDE.md file. What Is Inside The repository ships a single CLAUDE.md file with four operational principles: Think Before Coding — Surface ambiguity rather than assuming. State uncertainties explicitly, acknowledge multiple valid interpretations, and request clarification before writing code. Simplicity First — Implement only what was requested. No premature abstractions, no speculative configurability, no extra error-handling for edge cases that were not mentioned. The evaluation test: would an experienced engineer call this overcomplicated? If yes, simplify. Surgical Changes — Touch only the code the task requires. Do not improve adjacent sections, do not refactor working code, do not remove pre-existing dead code unless explicitly asked. Goal-Driven Execution — Replace vague directives "fix the bug", "add validation" with measurable success criteria and explicit verific