Stanford CS336 Publishes CLAUDE.md AI Agent Guidelines — 328 Points as Educators Debate Enforceability

Stanford CS336 / Hacker News·high signal

Stanford's CS336 'Language Modeling from Scratch' course published a CLAUDE.md file that instructs AI coding agents to teach rather than solve — explicitly banning implementation of core assignment components like tokenizers, transformer blocks, and training loops. The 328-point HN discussion centers on whether such guidelines are enforceable (students can modify them locally) versus whether intent-based honor systems are sufficient. The guidelines build on Carson Gross's earlier AGENTS.md framework.