Anthropic Adds AGENTS.md Fallback to Claude Code, Aligning with OpenAI's Agent Instruction Spec
Anthropic shipped Claude Code version 2.1.277 on September 18 with one ecosystem-significant addition: in any project that has no CLAUDE.md, Claude Code now reads AGENTS.md instead.
The behavior is configurable. Developers can review or change the active instruction file under “Project instructions” in /config. Pinning a specific file takes precedence over the automatic fallback.
One caveat: the feature is not yet available on Bedrock, Vertex, or Foundry. Managed API deployments will follow separately.
Why it matters
AGENTS.md is the instruction file format OpenAI’s Codex and other tools in that ecosystem use to pass project-level configuration to agents. Until now, teams running both Claude Code and OpenAI-adjacent tooling on the same codebase had to maintain two separate instruction files — CLAUDE.md for Anthropic, AGENTS.md for OpenAI.
With this change, a single AGENTS.md works for both. Teams that have already standardized on AGENTS.md can add Claude Code to their stack without touching their existing configuration. Claude Code falls back gracefully: CLAUDE.md takes priority if it exists; AGENTS.md is the next read.
The broader signal
Anthropic adopting a spec originated at OpenAI is unusual. The company has historically maintained its own instruction format and its own tooling conventions. Accepting AGENTS.md as a valid project instruction source signals that the two ecosystems are converging around shared conventions, at least at the developer tooling layer.
For enterprises managing agent fleets across multiple providers, reducing the instruction-file surface area from N formats to one is a concrete operational improvement — fewer files to audit, fewer drift vectors.
The feature ships immediately in Claude Code 2.1.277. Developers on Bedrock, Vertex, or Foundry will need to wait for a separate rollout.