Anthropic Adopts OpenAI's AGENTS.md Standard for Claude Code Interoperability

Anthropic has announced that its AI coding tool Claude Code will support the AGENTS.md instruction format originally developed by rival OpenAI, marking a key step toward interoperability among AI programming assistants.

What Is AGENTS.md?

AGENTS.md is a Markdown file placed in a repository root that serves as a project-level README for AI agents. It documents build procedures, test workflows, coding standards, dependency installation, and other conventions so that AI coding assistants can follow project-specific rules when generating code.

Previous Fragmentation

Before this update, Claude Code only recognized its proprietary CLAUDE.md configuration file. Developers using both OpenAI Codex and Claude Code had to maintain two nearly identical instruction files, creating significant duplication of effort.

Implementation Details

The feature lands in Claude Code 2.1.277 . When a project directory lacks CLAUDE.md, the tool automatically falls back to reading AGENTS.md. The compatibility layer is built on Claude Code's internal mod extension mechanism, and the relevant source code has been open-sourced. Teams can enable or disable the fallback via the interactive /config command.

Industry Adoption and Governance

As of December 2025, over 60,000 open-source repositories on GitHub already use AGENTS.md. OpenAI has donated the specification to the Agentic AI Foundation , where it is managed by an industry consortium rather than remaining a private OpenAI standard. Notably, Anthropic is a founding member of that foundation.

OpenAI Codex's lead publicly congratulated the move, calling cross-vendor compatibility the right direction for the industry.

What This Means for Developers

Eliminates duplicate maintenance: A single AGENTS.md file can be consumed by OpenAI Codex, Claude Code, and any other tool that supports the specification.

Backward compatibility preserved: CLAUDE.md retains highest priority; AGENTS.md is only used when CLAUDE.md is absent.

Controllable opt-in: Teams that prefer not to use the fallback can disable it via /config.

Conclusion

The rapid evolution of AI coding agents has led each vendor to introduce its own project instruction format, forcing developers to write multiple configurations. With Claude Code adopting AGENTS.md — already supported by Cursor, Aider, JetBrains Junie, Gemini‑CLI, and others — a single repository-level AI instruction file that works across tools is becoming a practical reality.