Claude Code adds AGENTS.md fallback for repositories shared across coding agents

Why it matters

AGENTS.md is becoming shared infrastructure for coding agents. Claude Code's support reduces duplicated repository rules, while the underlying mod previews deeper control over Anthropic's harness.

Anthropic added AGENTS.md support to Claude Code on September 18th, letting the coding agent read a repository's shared instruction file when it cannot find a Claude-specific CLAUDE.md.

Thariq Shihipar (@trq212), an engineer on the Claude Code team, announced the change alongside version 2.1.277. Anthropic's release notes confirm that AGENTS.md becomes the fallback when a project has no CLAUDE.md. Users can change the behavior through the "Project instructions" setting in /config.

The fallback removes a minor but persistent tax for engineering teams that use several coding agents in the same repository. AGENTS.md provides build commands, testing requirements, code conventions and other standing instructions in standard Markdown. A team using that file with Codex, Cursor or Gemini CLI previously needed a separate CLAUDE.md, an import or a workaround to deliver the same guidance to Claude Code.

The existing Claude convention retains priority. Claude Code will continue loading CLAUDE.md when one is present, so the update does not silently replace instructions written specifically for Anthropic's agent. The fallback is also unavailable on Amazon Bedrock, Google Vertex AI and Microsoft Foundry at launch, according to the release notes.

One file for competing agents

AGENTS.md has moved beyond an OpenAI-specific convention. The format's documentation describes it as a schema-free Markdown file that can sit at a repository root or inside subdirectories, allowing more specific instructions to apply to individual packages in a larger codebase.

OpenAI released the convention in August 2025 and later contributed it to the Linux Foundation's Agentic AI Foundation. The foundation said in its December 2025 launch announcement that more than 60,000 open-source projects had adopted AGENTS.md, with support spanning tools including Codex, Cursor, Devin, Gemini CLI, GitHub Copilot and VS Code.

Claude Code's adoption gives repository maintainers a cleaner path to keeping agent behavior consistent across competing tools. A checked-in instruction file can tell each agent which package manager to use, which tests must pass, which generated files should remain untouched and which security constraints apply. Duplicating those rules across vendor-specific files creates an obvious failure mode: one copy changes while another does not.

Anthropic chose a compatibility-first default. Projects already organized around CLAUDE.md continue operating as before. Projects standardized on AGENTS.md can bring Claude Code into the same workflow without adding another file solely for Anthropic.

Anthropic exposes the mod underneath

The implementation also previews a broader Claude Code customization layer. Shihipar said AGENTS.md support was built as a native "mod," Anthropic's upcoming mechanism for changing the Claude Code harness, and published the mod's source and documentation.

The built-in mod documents four instruction modes. Claude Code can load only CLAUDE.md, use AGENTS.md as a fallback, load both formats, or limit the context to centrally managed instructions. The fallback mode is the default.

Those controls make the feature broader than a filename alias. Teams can preserve Claude-specific instructions where they need them, standardize entirely on AGENTS.md, or deliberately combine the two. The mod also places AGENTS.md content into the same project-instruction context and framing Claude Code already uses for CLAUDE.md, rather than injecting an unrelated prompt at session startup.

Anthropic has kept administrative control over that choice. The documentation says mod options can be stored in a user's settings, passed through --settings, or imposed through managed settings. A repository's own .claude/settings.json cannot set the mod options. That prevents code checked into a project from choosing an individual's or organization's instruction-loading policy.

Shihipar said developers will eventually be able to create custom versions of project-instruction handling through mods. Anthropic has not attached a general availability date to that system. The AGENTS.md implementation provides the first concrete view of its shape: configurable code that can alter how the harness builds context before the model starts working.

For engineering teams, the immediate change is smaller and useful. Claude Code can finally enter a repository already configured for other coding agents and read the instructions waiting there.