What's this about?
A GitHub issue asking Claude Code to support `AGENTS.md` — the cross-tool convention other coding agents have adopted — rather than requiring its own `CLAUDE.md`. The issue has accumulated roughly 700 comments, and reaching HN gave it another 78.
What exactly are these files?
Both are plain markdown files you put in your repository containing project-specific instructions for an AI coding agent: the conventions, the build commands, the things it should never touch. Every agent reads one, and each vendor chose a different filename — `CLAUDE.md`, `AGENTS.md`, and others. Since the file is just text, supporting a second name is close to trivial, which is precisely why the refusal reads as a decision rather than a limitation. It's a standards fight in miniature, with the twist that the standard in question is "what to call a text file."
What's the community saying?
- The advertising theory dominated. A `CLAUDE.md` in every repository is free brand placement — "the 'Sent from my iPhone' of our time" — and commenters connected it to auto-added commit attribution.
- "This is a nothingburger." The most-repeated practical response: a one-line `CLAUDE.md` saying "read AGENTS.md", or a symlink, solves it in ten seconds.
- A cleaner variant several people use: keep a git-ignored `.claude/CLAUDE.md` whose only content is a directive to only ever update `AGENTS.md`.
- The disproportion is the joke. "700 comments on an issue about reading a markdown file" — with the pointed addition that the issue tracker rendering it is a React app that should be a static page.
- A real technical question nobody answered: how does `AGENTS.md` behave with subagents and swarms — is a single project-level file even the right abstraction once multiple agents with different roles are running?
- The harshest framing: "they want to be the Apple of AI — non-standard interfaces and grandiose claims."