Claude Code now also accepts instructions in OpenAI’s Agents.md format

以下正文同步自 InfoWorld，版权归原站所有，已转换为易读排版。

One thing that made it difficult for developers to switch AI coding tools on a project is that Anthropic’s Claude Code didn’t look for instructions in the same place as other agents including OpenAI’s Codex — but now that’s changing.

Claude and Codex each accept instructions in markdown format, a plain-text way of giving AI coding agents project-specific behavioral instructions. Until now, Claude Code looked for project-specific instructions in a file named CLAUDE.md by default, while Codex and other agents use AGENTS.md, the format of which is an open source initiative governed by the Agentic AI Foundation, an initiative under the Linux Foundation.

But now, as Thariq Shihipar, a member of Anthropic’s technical staff, wrote in a post on X on Friday, “We’re adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md,”

That means developers using multiple coding tools alongside Claude Code can now use the same project instructions across those agents, rather than maintaining separate instruction files for Claude and for everything else.

Developers will no longer have to maintain the same or similar instructions in two files, nor to ensure that any change to a project’s coding conventions, build commands or other agent instructions are updated in two locations, a system that created additional maintenance work and left room for the instructions to fall out of sync. Instead, they can use AGENTS.md as a common instruction file across their coding tools, provided there is no CLAUDE.md file in the project.

AGENTS.md support could reduce operational burden

For Amit Kumar Jena, AI development head at IT consulting firm Kanerika, being able to use AGENTS.md with Claude Code could “remove some real busywork” for development teams working with multiple coding tools.

“Before, when Claude only read CLAUDE.md, any team running multiple coding agents on the same repository had to keep a AGENTS.md file in sync by hand, or fake it with a symlink or a one-line import pointing Claude at the other file. But that workaround could only hold up fine in a small project,” Jena said.

“Nested across a monorepo (which are often representative of large codebases) with instructions at multiple directory levels, it turns into real maintenance debt, the kind that gets missed under deadline pressure until the tools quietly drift out of sync,” Jena added.

The reduction in maintenance and operational work that Anthropic’s change enables could also make it easier for enterprises to govern how AI coding agents are used across development teams.

“For enterprises, a common instruction format can become a practical control point for coding conventions, security requirements, architectural guardrails, and compliance policies,” said Charlie Dai, principal correspondent at Forrester.

But not for every Claude Code deployment

However, a common instruction format does not necessarily make AI coding agents interchangeable, Jena cautioned: “CLAUDE.md includes Claude-specific features that AGENTS.md was not designed to support, so instructions written for one agent, such as Codex, may not behave in exactly the same way when Claude Code reads them.”

There is also a deployment-related limitation to Claude Code’s AGENTS.md support, Jena pointed out, quoting Claude Code’s documentation that reads: “Your session doesn’t fetch feature flags from Anthropic, for example because you use Amazon Bedrock or another third-party provider, or you disabled telemetry. “

The support for AGENTS.md depends on the tool fetching feature flags from Anthropic and if Claude Code cannot fetch those flags, AGENTS.md support is unavailable and Claude Code falls back to CLAUDE.md, Jena said.

If an enterprise uses Claude Code through environments such as Amazon Bedrock or Google Vertex AI, or if it disables telemetry, development teams may not be able to use AGENTS.md as the common instruction file even though the file itself is local and does not require a network connection to be read, Jena added.

Nevertheless, that limitation can be addressed to an extent.

Claude Code allows users to configure which files it reads, including AGENTS.md, CLAUDE.md, or both, according to Shihipar, who in his X post wrote that “users can change the behavior through the /config command.”

That should give teams a way to use AGENTS.md for instructions shared across coding tools while retaining CLAUDE.md for instructions specific to Claude Code.

正文由 FLUX 从来源站点 RSS 同步，内容未经改写；遇到排版缺失或需要图片、视频时请以原文为准。