Summary
Stanford's CS336 assignment repository now includes a CLAUDE.md policy that tells AI coding assistants to act as teaching aids, not solution generators. The rules prohibit writing Python or pseudocode, completing TODOs, editing student repos, running shell commands, or converting assignment requirements into finished code.
What changed
The CS336 repository includes a 74-line AI Agent Guidelines file governing how coding assistants should interact with students on implementation-heavy assignments.
Why it matters
This is a concrete example of AI-agent governance moving from abstract academic-integrity language into machine-readable repo instructions. It gives educators and hiring managers a template for preserving learning outcomes while still allowing debugging guidance, conceptual explanation, and code review feedback.
Evidence excerpt
The guideline states that AI agents should function as teaching aids, not solution generators, and should not write Python or pseudocode, complete TODOs, edit student repos, or run bash commands.