In the words of Thariq Shihipar

If you work with AI coding assistants across different projects, you've probably had to juggle multiple configuration file formats. That friction just got smaller. Anthropic has announced that Claude Code now understands AGENTS.md files, the emerging open standard for giving coding agents project-level instructions. Starting with version 2.1.277, Claude Code can read AGENTS.md directly, so teams using several AI tools no longer need to maintain separate instruction files just for Claude.

Here's what changed, how the fallback works, and why this update signals something bigger about where Claude Code is heading.

What the AGENTS.md Support Actually Does

The behavior is simple and predictable. When Claude Code starts working in a folder, it first looks for a CLAUDE.md file, the traditional way to give Claude project-specific context like coding conventions, build commands, or architectural notes.

If no CLAUDE.md exists in that folder, Claude now falls back to checking for an AGENTS.md file instead. If it finds one, it uses those instructions just as it would have used a CLAUDE.md file.

For developers, this means one less file to duplicate. If your repository already follows the AGENTS.md convention because you use other agent tools, Claude Code will respect those same instructions without any extra setup on your part.

Built on Claude Code Mods

The more interesting part of the announcement is what's underneath. According to Anthropic, the AGENTS.md feature is implemented as a built-in "mod" — part of Claude Code mods, an upcoming system for customizing the Claude Code harness itself.

In other words, rather than hardcoding AGENTS.md handling directly into the tool, the Anthropic team built it as a customizable layer. Anthropic says you'll eventually be able to write your own custom versions of project instruction handling, tailoring how Claude Code picks up and applies project context to suit your workflow.

The source code for this particular mod has been published, so curious developers can look at exactly how the fallback logic works today.

Why This Matters for Mixed-Tool Teams

Plenty of engineering teams don't rely on a single AI assistant. One developer might prefer Claude Code while a colleague uses a different agent that reads AGENTS.md. Until now, keeping both tools happy meant maintaining parallel instruction files that could easily drift out of sync.

With this update, AGENTS.md becomes a genuinely shared standard: write your project guidance once, and any compatible tool can use it. That reduces maintenance overhead and cuts down on the classic problem of stale, conflicting configuration files scattered across a repo.

It's also worth noting that CLAUDE.md remains fully supported. This isn't a replacement — it's an additional option. Teams already invested in CLAUDE.md don't need to change anything, though those who want maximum portability across tools now have a reason to consider the standard.

What to Watch Next

The mention of mods is arguably the bigger story here. If project instruction handling is just one built-in mod, the natural question is what else will become customizable once the mod system ships more broadly. Anthropic has indicated that custom mods will let developers build their own versions of features like project instructions, which could open the door to significant workflow tailoring.

For now, the practical takeaway is straightforward: update to Claude Code version 2.1.277 or later, and if your project uses AGENTS.md but no CLAUDE.md, Claude will pick up your instructions automatically. The announcement came from Thariq Shihipar, who also pointed to additional mods available for exploration.

Tags: claude-code, anthropic, coding-agents, generative-ai, llms