Claude Code Adds Support for OpenAI's AGENTS.md, Ending a Familiar Workaround

Anthropic has added native support for AGENTS.md to Claude Code, the company's command-line coding agent, letting the tool read instruction files written for OpenAI's format without any conversion step. "We're adding support for AGENTS.md to Claude Code," Claude Code engineer Thariq Shihipar said in a social media post announcing the change, according to The Register. "Starting today" the tool would pick up AGENTS.md files directly, he said, rather than requiring Claude's own CLAUDE.md format.

Before this update, developers maintaining repositories that served multiple AI coding tools had to duplicate instructions across separate files or engineer workarounds. According to KuCoin's writeup of the change, common fixes included creating symbolic links between CLAUDE.md and AGENTS.md, or manually inserting an @AGENTS.md reference inside CLAUDE.md so Claude would pull in the other file's contents. Both approaches worked but required developers to remember to set them up in every project, and both broke silently if a file got renamed or moved.

The new support is built on what KuCoin describes as Claude Code's existing "mods" mechanism, an extension system Anthropic has now used to make AGENTS.md support a built-in mod, with the underlying code open-sourced. Practically, that means Claude Code will automatically read the nearest AGENTS.md file in a project's directory tree, matching the behavior described on the AGENTS.md project site, where a file placed inside a specific package takes precedence over one higher up.

AGENTS.md itself did not originate with a single company. Its own documentation credits "collaborative efforts across the AI software development ecosystem, including OpenAI Codex, Amp, Jules from Google, Cursor," and other tools, and the spec is now stewarded by the Agentic AI Foundation under the Linux Foundation, a neutral home that predates Anthropic's decision to adopt it. That governance structure is part of why the reaction, per KuCoin's account of the social-media response, ran toward relief rather than rivalry, with comments reading "Finally unified!" following the announcement.

The move mirrors an earlier reversal in the other direction. According to a deploy-focused writeup from bex.co, Anthropic published its Agent Skills specification, a folder of Markdown paired with a small YAML header called SKILL.md, on December 18, 2025, and OpenAI shipped support for the identical spec in ChatGPT and Codex CLI within 48 hours rather than building a competing format. By March 2026, the same account says, roughly 30 agent clients had adopted Agent Skills, including Microsoft Copilot, Cursor, JetBrains, Google's Gemini CLI, and Block's Goose.

None of this guarantees the two specs stay aligned. The bex.co analysis flags the obvious risk directly: two vendors shipping support for the same file format can drift apart if one side patches its interpretation without coordinating with the other, leaving teams that built deploy runbooks around a shared assumption exposed to silent breakage. For now, though, the fix is narrower and more immediate: a Claude Code project with an AGENTS.md file already in place needs no symlink, no manual import line in CLAUDE.md, and no separate maintenance step to have Anthropic's agent read it correctly.