Anthropic finally adds AGENTS.md support to Claude Code
TL;DR Highlight
Claude Code 2.1.277 now reads AGENTS.md when no CLAUDE.md exists in a folder
Who Should Read
A developer using several AI coding tools who is tired of maintaining both CLAUDE.md and AGENTS.md
Core Mechanics
- Falls back to AGENTS.md without CLAUDE.md
- It starts in Claude Code version 2.1.277.
- If a folder has no CLAUDE.md, Claude checks for and uses AGENTS.md.
- Toggle it in /config
- You can turn this behavior on or off in /config.
- Built on Claude Code mods
- AGENTS.md support is built on Claude Code mods.
- Mods are described as an upcoming way to customize the Claude Code harness (the runtime wrapped around the model).
- Built-in mod, custom ones coming
- AGENTS.md support is a built-in mod.
- You will be able to build your own custom versions of project instructions.
- Mod source is public
- The author shared a GitHub link to the mod's source.
Community Reactions
How to Apply
- Check your Claude Code version and /config
- Update Claude Code to 2.1.277 or later.
- Open /config and confirm the AGENTS.md toggle is on.
- Test in a folder without CLAUDE.md
- Run Claude Code in a repo that has only AGENTS.md.
- Ask about a rule to confirm the instructions are applied.
- Decide how to dedupe instruction files
- Find repos where CLAUDE.md and AGENTS.md overlap.
- CLAUDE.md takes precedence when present, so pick which file to keep.
Try It
# Run in the Claude Code terminal to check AGENTS.md is picked up (in a folder without CLAUDE.md)
{{one rule that exists only in AGENTS.md, e.g. write commit messages in Korean}}
Tell me which file the project instructions you are following in this folder came from, and quote the rule above verbatim.
Next: turn the toggle off in /config, ask the same question again, and compare the answers.Terminology
Related Papers
Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.
Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.
Harness Engineering 101: The LLM is the brain, the harness is the body
An agent is a loop managing a JSON array; 16 chapters build one from scratch
Harness2 by Google Research – Recursive Agent Harnessing
Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.
Ousted Savannah police officer gave ICE agent husband access to Flock system
A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
Running Claude Code agents in tmux works well, but limits appear around four agents.