Claude Code now reads AGENTS.md if there is no Claude.md
TL;DR Highlight
Claude Code now reads AGENTS.md when a project has no CLAUDE.md
Who Should Read
A developer who uses Claude Code alongside Codex or Cursor and maintains CLAUDE.md symlinks to AGENTS.md
Core Mechanics
- Reads AGENTS.md when CLAUDE.md is missing
- Version 2.1.277 (Sept 18, 2026) added AGENTS.md support.
- In a project with no CLAUDE.md, Claude Code reads AGENTS.md instead.
- It is configurable
- Change it under 'Project instructions' in /config.
- Not yet available on Bedrock, Vertex or Foundry, per the changelog.
- This page is the 2.1.304 release notes
- 2.1.304 (Oct 19, 2026) adds /auto-mode-setup and turns tool search on by default behind a custom ANTHROPIC_BASE_URL.
- 2.1.302 adds the CLAUDE_CODE_AUTO_MODE_MODEL variable.
- Context-related fixes shipped too
- Self-paced /loop wakeups at the 1-hour maximum no longer fire just after the prompt cache expires.
- After compaction the skill list is sent again, and /reload-skills now sends only new or changed skills.
Community Reactions
- Complaints about how AGENTS.md is followed
- A commenter says Claude Code wraps both files in a system-reminder saying the context may not be relevant.
- They say Codex follows AGENTS.md better, and the related issue (anthropics/claude-code#18560) was closed Not Planned.
- Skills folder still unsupported
- Several commenters note .agents/skills is not included in this change.
- A news link was shared about Shopify's CEO threatening to ban Claude Code over it.
- Existing workarounds
- One commenter put just '@AGENTS.md' in CLAUDE.md and it worked well.
- Others used a symlink and say they can now delete it.
- Concern about settings precedence
- A commenter says missing settings.json values get defaults that override CLAUDE.md instructions.
- Their example was a 'no Made with Claude Code' instruction being ignored.
How to Apply
- Test in a folder without CLAUDE.md
- Put only an AGENTS.md in a temp folder and run Claude Code.
- Ask it to summarize the AGENTS.md rules to confirm it reads them.
- Decide whether to remove symlinks
- List the CLAUDE.md to AGENTS.md symlinks in your repos.
- Delete them if the updated version (claude --version) works without them.
- Keep '@AGENTS.md' on Bedrock or Vertex
- Those environments are listed as unsupported for now, so keep a one-line CLAUDE.md with '@AGENTS.md'.
- Keep skills under .claude or copy them there.
Try It
# Run in a terminal with Claude Code (temp folder with only AGENTS.md)
mkdir /tmp/agents-test && cd /tmp/agents-test
echo '모든 답변 끝에 "AGENTS-OK"를 붙여라' > AGENTS.md
claude --version
# After starting claude, type:
안녕, 프로젝트 지침에 따라 인사해 줘
# Next: change Project instructions in /config and compare the resultsTerminology
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.