Claude code agents.md support finally landed this week — and if you maintain repos that more than one AI coding agent touches, this small change kills a genuinely annoying bit of duplication. Here’s what we’re going to cover: exactly what shipped in version 2.1.278, how the CLAUDE.md fallback order works, and how I’d restructure your agent instruction files now.
Short answer:
- Claude Code v2.1.278 (19 September 2026, per the official changelog) now reads AGENTS.md in projects that have no CLAUDE.md.
- AGENTS.md is the open standard already used by 60,000+ open-source projects and read by Codex, Cursor, Copilot, Aider and more (per agents.md).
- CLAUDE.md still takes priority when it exists — nothing breaks for existing setups.
- No configuration needed: just update Claude Code.
- Multi-agent teams should consolidate shared instructions into one AGENTS.md.
Claude Code AGENTS.md Support: What Actually Shipped
The official Claude Code changelog entry for version 2.1.278, dated 19 September 2026, is one line long, and it is the line a lot of us have been waiting for: projects without a CLAUDE.md now read AGENTS.md instead.
That is the whole mechanism, and the ordering matters. Claude Code still looks for CLAUDE.md first. If it finds one, nothing about your setup changes. If it does not find one, it now falls back to AGENTS.md rather than starting the session with no project context at all. No flag, no setting — you update Claude Code and the behaviour is just there.
The same 2.1.278 release also changed auto mode to default to a server-side classifier for Claude API and Enterprise users (so you are not charged for classifier overhead), but the AGENTS.md line is the one that changes how you structure your repos.
What AGENTS.md Is (and Why It Beats Tool-Specific Files)
AGENTS.md is an open standard with a simple pitch, quoting the official spec at agents.md: it is “a dedicated, predictable place to provide the context and instructions to help AI coding agents work on your project”. Think of it as a README for robots: build and test commands, code style rules, testing instructions, security considerations, PR conventions — as the spec puts it, anything you would tell a new teammate belongs there too.
The adoption numbers are what make this update matter. Per the agents.md site, over 60,000 open-source projects already ship an AGENTS.md, and the compatibility list reads like a roll call of every serious coding agent: OpenAI Codex, Cursor, Aider, GitHub Copilot, VS Code, Google Jules, Devin, Zed and plenty more. Until this week, Claude Code was the notable absentee — teams running multiple agents had to duplicate the same instructions into a CLAUDE.md just for Claude.
🔥 Want this set up without the guesswork? If you are wiring up AI coding agents to actually ship SEO sites and content systems, this file structure is step one. Inside the AI Profit Boardroom (3,700+ members, four live calls a week, daily tutorials, done-for-you templates and a 30-day roadmap) we build systems like this together, step by step. And if you would rather talk it through one on one first, book a free SEO strategy session and I will map it out with you.
How Claude Code AGENTS.md Support Works Day to Day
Here is how I would set up a repo now, based on the fallback rule in the changelog and the precedence rules in the standard:
- Multi-agent teams: keep one AGENTS.md at the repo root as the single source of truth. Delete or skip CLAUDE.md entirely, and Claude Code 2.1.278+ reads AGENTS.md like every other agent.
- Claude-only repos: keep CLAUDE.md if you like — it still wins whenever it exists, so nothing breaks.
- Claude-specific overrides: this is the sharp edge. Because CLAUDE.md takes priority rather than merging, a CLAUDE.md means your AGENTS.md is ignored by Claude Code in that scope. If you want shared instructions plus Claude extras, put the shared rules in AGENTS.md and have a minimal CLAUDE.md that references them explicitly.
- Monorepos: the standard supports nested AGENTS.md files per package, and the closest file to the edited code wins, with explicit chat prompts overriding everything.
The migration itself is about as low-risk as repo changes get: rename CLAUDE.md to AGENTS.md, run a session, and check the agent still respects your build commands and style rules. The standard has no required fields — it is plain Markdown.
Should You Delete CLAUDE.md? My Take
If your codebase is touched by more than one AI agent — and in 2026, most serious teams are at least trialling two — consolidating on AGENTS.md is the obvious move. One file, every agent reads it, and your instructions stop drifting apart between tools. That drift is a real failure mode: I have seen repos where the Codex instructions said one test command and the Claude file said another, and the agents merrily did different things.
If you are Claude-only, there is genuinely no action needed. This update was deliberately designed not to break anyone: it only fires when CLAUDE.md is absent.
The Bottom Line on Claude Code AGENTS.md Support
Claude Code AGENTS.md support is a small changelog line with a big practical payoff: as of version 2.1.278 (19 September 2026, per the official changelog), one open-standard file can now drive Claude Code, Codex, Cursor, Copilot and the rest of your agent stack. CLAUDE.md still wins when present, so nothing breaks — but if you run a multi-agent workflow, move your shared instructions into AGENTS.md and stop maintaining duplicates. Fewer files, less drift, and every agent on your team reading from the same playbook.
FAQ: claude code agents.md support
Which Claude Code version added AGENTS.md support?
Version 2.1.278, dated 19 September 2026 in the official Claude Code changelog. Update Claude Code and the behaviour is on automatically — there is no flag or setting to flip.
Does CLAUDE.md still work after this update?
Yes. Nothing changes for existing projects. The changelog wording is precise: projects without a CLAUDE.md now read AGENTS.md instead. If CLAUDE.md exists, Claude Code keeps using it as before.
What is AGENTS.md exactly?
It is an open standard: one predictable Markdown file that gives AI coding agents the context they need to work on your project — build and test commands, code style, testing instructions, security notes and PR conventions. Per the official agents.md site, over 60,000 open-source projects already use it.
Which other tools read AGENTS.md?
The agents.md compatibility list includes OpenAI Codex, Cursor, Aider, GitHub Copilot, VS Code, Google Jules, Devin and Zed, among others — and now Claude Code joins them via version 2.1.278.
How does AGENTS.md handle monorepos?
You nest AGENTS.md files in subdirectories. The standard’s precedence rule: the closest AGENTS.md to the edited file wins, and explicit user chat prompts override everything.
Should I convert my CLAUDE.md to AGENTS.md?
If your repo is only ever touched by Claude Code, there is no rush — CLAUDE.md still takes priority. If several agents work on the same codebase, one AGENTS.md beats maintaining near-identical per-tool files.
Related reading
About the author: Julian Goldie is an SEO agency owner with 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, 10+ years in SEO and a best-selling SEO book. He runs the AI Profit Boardroom community, and you can book a free SEO strategy session with his team any time. For agency work, book a call for a custom quote.
Where to go from here: if you want AGENTS.md running across your whole agent stack working inside a real AI SEO system rather than as a one-off experiment, join us in the AI Profit Boardroom — or book a free SEO strategy session and we will look at your setup together.
Last updated September 2026. This is the living guide to claude code agents.md support — it gets updated as the tools change.