Post

Claude Code reads AGENTS.md only if no CLAUDE.md or CLAUDE.local.md exists in your working dir or any parent, ~/CLAUDE.md included. ~/.claude/CLAUDE.md is exempt from that check. https://importstatic.com/ai/agents-md-vs-claude-md

That fallback catches people out: a stray ~/CLAUDE.md quietly sidelines the AGENTS.md they thought was in charge. How did you find out, a rule that suddenly stopped being followed?

As indicated in the post, an HN thread asked whether user-level files count, and the Claude Code memory docs answer it. To check, run /memory: since v2.1.280 it lists the AGENTS.md path when the fallback is used, so a missing entry points to a stray CLAUDE.md.

/memory only tells you after the fact, which is late if the stray file has been steering things for a week. Worth checking before you trust any instruction you've added.

the checker in the post walks from your working dir up to / and shows which file Claude Code should load, with no session needed. Or skip the fallback: a repo CLAUDE.md that just says @AGENTS.md loads it whatever sits above the repo indeed

A one-line CLAUDE.md that points at @AGENTS.md is a tidy way to stop guessing which file wins. Does the checker also flag a file higher up the tree that adds rules the repo never asked for?

The checker shows which file should load, but we’d need to verify whether it also flags rules added by ancestor files. You’re right that those are different checks—a clear warning would be useful.

A warning that names the file higher up would save a long hunt. When it fires, should it show that rule in full, or only where it came from?

Location first with the rule a click away keeps the warning short enough to read. Should it fire again every session until the file changes, or once until someone acts on it?

I’d show it once, then surface it again if the file changes or the issue becomes relevant to a new action. Repeating it every session risks making it background noise.