Product1 publisher2 min readPublished
Anthropic finally lets Claude read the config file other agents already use
Claude can now read AGENTS.md, the file other agent tools already look for. XDA-Developers reports developers called the change small and long overdue, and the Claude-specific syntax that forced a second file has not moved.
The Product Desk · Product desk
What happened
- Anthropic announced that Claude can now read AGENTS.md files instead of relying only on its own CLAUDE.md file.
- AGENTS.md is the file most agents read for their parameters, so one copy can be ported between models without restating your preferences to each one.
- Until now, developers running Claude alongside other agents spent years on workarounds to get one set of config files working everywhere.
- XDA-Developers reports that Anthropic pinned the change as a big one while developers called it a small, long-overdue tweak.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Whoever maintains the repo now has to justify the second file at review, because the Claude-specific syntax is the only thing still holding CLAUDE.md in place.
- constraint One vendor honouring the shared filename leaves a mixed fleet's instructions short of interchangeable; the convention holds only while every agent reads the same file and acts on the same entries.
- contradiction The gap between Anthropic's framing and the developer response sorts the audience for the feature: it lands for people running several agents, and Claude-only users get nothing out of it.
A repo with AGENTS.md at the root and CLAUDE.md beside it is a sync problem before it is a standards problem. Someone adds a rule about how the test suite runs, puts it in one file, and the agent reading the other one never sees it. Native support for AGENTS.md removes that failure mode in the repos where both files said roughly the same thing [1].
One file is the saving. A repo that kept AGENTS.md for every other agent and CLAUDE.md for Claude can now cover both with a single file, provided nothing in it depends on Claude-specific syntax [7].
XDA-Developers advises that people who mostly rely on Claude are still better off with a CLAUDE.md, because the special syntax lives there [5]. The people this helps are the ones maintaining instruction files in repos where more than one agent actually runs. A team that only runs Claude stays where it was.
Other AIs could read CLAUDE.md if you pointed them at it, according to XDA, but it contained entries the model may not know what to do with [4]. That was the ceiling on the older workaround. A shared filename settles where an agent looks. On XDA's account, the syntax split survives the filename fix [5].
Read CLAUDE.md line by line and sort each instruction into one of two buckets: it means the same thing to any agent, or it only means something to Claude Code [4]. One pass is enough. If that second bucket is empty, the repo can keep one file. If anything lands in it, the split stays, and those lines are what you are paying the sync cost for. XDA's write-up does not say what Claude does when both files are present, so a team that wants to consolidate has to check the precedence itself [8].
What to watch
- Whether Anthropic documents precedence for repos that contain both AGENTS.md and CLAUDE.md.
- Whether Claude-specific syntax gains a place inside AGENTS.md so the portable file can carry Claude-only instructions.
- Whether other agent vendors keep AGENTS.md as their primary filename or add house variants the way Claude did.