Claude now reads AGENTS.md
Like, I have an idea of the reason why they didn't read before, but it's such a petty reason I would love to be proven wrong. Anthropic and OpenAI are always competing on being the most megalomaniac lab, and not adopting standards is one way to say Our Product Is Different, it's not like the others.
It is like the others. Coding harnesses all share the same core. They may have different bells and whistles, some may fit better into your workflow than others, but it's still just a loop with a prompt hooked to an inference API. It's a very simple concept. If you are getting into AI for development, I strongly recommend implementing one. It's like implementing a shell: you can get something done in an afternoon and learn the fundamentals of a tool you'll use every day.
The CLAUDE.md thing is dumber still because the workaround is so simple: write AGENTS.md for everyone else, and put @AGENTS.md on CLAUDE.md, done. Glad I can delete that file now.
Now Claude, please, read the .agents/ folder!