Anthropic's Claude Code Adds AGENTS.md Support

Anthropic released version 2.1.277 of Claude Code on September 18, 2026, adding native support for the AGENTS.md instruction file format. The update allows Claude Code to automatically read AGENTS.md files when a project lacks the tool’s proprietary CLAUDE.md format, eliminating a compatibility friction point for developers using multiple AI coding assistants. This marks Claude Code’s adoption of a cross-tool standard already used by OpenAI Codex, GitHub Copilot, Gemini CLI, and other AI development tools.

Cross-Platform Compatibility Arrives in Claude Code

Version 2.1.277, shipped on September 18, enables Claude Code to automatically read AGENTS.md files whenever a project lacks CLAUDE.md, removing a friction point for developers using multiple AI coding agents. Before this release, developers working with both Claude Code and other AI assistants had to maintain separate instruction files or use workarounds such as symlinks and import statements.

AGENTS.md, originally from OpenAI’s Codex, outlines project structure and dependencies. The standardized format has been adopted across the AI coding tool ecosystem, allowing developers to write a single set of project instructions that multiple assistants can understand. In a project with no CLAUDE.md, Claude Code now reads AGENTS.md instead.

The agents-md mod is now open-sourced, enabling developers to build custom versions. This gives teams flexibility to tailor how their AI coding tools interpret project instructions while maintaining compatibility with the broader standard.

Configuration and Implementation Details

Claude Code now reads AGENTS.md directly in version 2.1.277, whereas previously developers had to import it from a CLAUDE.md file. The feature can be configured through the “Project instructions” setting in the /config menu, though the update is not yet available on Bedrock, Vertex or Foundry.

The implementation includes specific conditions where AGENTS.md support may not be available. Sessions that don’t fetch feature flags from Anthropic, such as those running on Amazon Bedrock or other third-party providers, will continue to require CLAUDE.md files. The same limitation applies to sessions with telemetry disabled or in the first session after installing or upgrading the tool.

Projects Feature Adds Cloud-Based Workflow Automation

Alongside the AGENTS.md update, Anthropic has rolled out a Projects feature for Claude Code. Anthropic rebuilt Claude Projects into a coordinator that delegates work to cloud threads and sub-agents, with shared memory and RAG retrieval. This architecture allows a single conversation to spawn parallel cloud sessions that continue running after the user disconnects.

The Projects capability transforms Claude Code from a single-session tool into a multi-threaded system where work can be distributed across different cloud instances. Each thread maintains access to shared context and retrieval-augmented generation capabilities, enabling more complex development workflows that persist beyond individual user sessions.

Key Facts

Version 2.1.277 shipped on September 18, 2026, adding AGENTS.md fallback support
AGENTS.md is a cross-tool standard used by OpenAI Codex, GitHub Copilot, Gemini CLI, and other AI coding assistants
Open-source mod: The agents-md implementation is now available for developers to customize
Not yet on cloud providers: Bedrock, Vertex, and Foundry deployments don’t yet support the feature
Projects feature: New coordinator system delegates work to cloud threads with shared memory

Industry Context

The AGENTS.md standard has gained traction as development teams increasingly use multiple AI coding assistants for different tasks. Maintaining separate instruction files for each tool created redundant work and increased the risk of inconsistent behavior across different AI agents working on the same codebase.

Claude Code 2.1.277 adds native AGENTS.md fallback support, aligning with the cross-tool standard used in over 60,000 repositories. The widespread adoption of the format across the developer community created pressure on tool vendors to support a unified approach to project instructions.

The update reduces integration costs for teams already using the format, particularly those running heterogeneous AI development environments where different team members or different projects may prefer different coding assistants.