Claude Code vs AGENTS.md: What the Shopify Ban Means

Published August 25, 2026By ABD Legacy LLC
Claude Code AGENTS.md AI coding agents Shopify
Claude Code logo versus AGENTS.md standard — split-brain config files across AI coding agents, Shopify ban threat context

Does Claude Code read AGENTS.md? No. As of August 25, 2026, Claude Code reads CLAUDE.md, not AGENTS.md, per Anthropic's official documentation. AGENTS.md is the open, plain-Markdown "README for agents" standard used by more than 60,000 open-source projects and stewarded by the Linux Foundation's Agentic AI Foundation. The gap just exploded into a public fight: Shopify CEO Tobi Lütke said he is "thinking about banning Claude code at Shopify" until Anthropic reads AGENTS.md and .agents/skills. Here is what happened, why split-brain is real, and what it means for your stack.

What happened: Shopify's CEO put Claude Code on notice

On August 25, 2026, Lütke posted on X that he was considering banning Claude Code at Shopify:

I'm thinking about banning Claude code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.

Insisting on only reading CLAUDE.md sometimes leads to split brain problems when different team members use different tools. Just unnecessary.

The post drew hundreds of thousands of views within hours, and X's trending topic page framed it as "Shopify CEO Considers Banning Claude Code Over File Reading Issue." The New Stack's headline captured the awkward timing: "Shopify's CEO threatened to ban Claude Code. Anthropic had already closed the feature request."

Framing note: Lütke said he was "thinking about banning" Claude Code. There is no evidence an actual ban has been enacted at Shopify. This is a threat and a signal, not a policy change.

Does Claude Code read AGENTS.md? Not yet

The direct answer is no. Anthropic's official memory docs state it verbatim:

Claude Code reads CLAUDE.md, not AGENTS.md. If your repository already uses AGENTS.md for other coding agents, create a CLAUDE.md that imports it so both tools read the same instructions without duplicating them.

The docs list four ways to bridge the gap:

  1. @import in CLAUDE.md. Create a CLAUDE.md whose body is @AGENTS.md. Claude Code loads the imported file at session start, and you can still add Claude-specific instructions around it.
  2. Symlink. ln -s AGENTS.md CLAUDE.md works when you don't need Claude-specific content. Windows users should prefer the import, since symlinks require admin or Developer Mode.
  3. /init with CLAUDE_CODE_NEW_INIT=1. The init command also reads AGENTS.md, .devin/rules/, .windsurf/rules/, and .clinerules, and folds relevant parts into a generated CLAUDE.md.
  4. /import (v2.1.213+). Appends a one-time copy of AGENTS.md into the matching CLAUDE.md, and carries over MCP servers, commands, subagents, and skills.

One caution: some third-party guides claim Claude Code "reads AGENTS.md as a fallback" when no CLAUDE.md exists. That is not in the official documentation, and the Claude Code team's own reply contradicts it. Do not plan around it.

What is the AGENTS.md standard?

AGENTS.md is a plain Markdown file placed in a repository's root (or nested directories) that tells coding agents how to build, test, and contribute. The official site describes it as "a simple, open format for guiding coding agents, used by over 60k open-source projects" — a README for agents.

OpenAI shipped the format with Codex in August 2025, then handed it to the Agentic AI Foundation (AAIF), a Linux Foundation directed fund co-founded by OpenAI, Anthropic, and Block, which also stewards MCP and goose. Here is the irony at the center: Anthropic co-founded the foundation that stewards AGENTS.md, yet its own flagship coding tool did not read the format natively as of August 2026.

The supported-agents list on agents.md is long: Codex, Cursor, Gemini CLI, GitHub Copilot, Aider, Devin, Windsurf, Junie, Jules, Amp, and more. Claude Code is not on it.

AGENTS.md vs CLAUDE.md vs .agents/skills

File / conventionWhat it isWho reads itStatus
AGENTS.mdOpen, tool-neutral instructions for coding agents ("README for agents")Codex, Cursor, Gemini CLI, Copilot, Aider, Devin, Windsurf, and moreOpen standard, AAIF stewarded
CLAUDE.mdClaude Code's native memory / instructions fileClaude Code onlyVendor-specific
.agents/skillsCross-tool skills directory convention (Lütke's shorthand)Codex CLI and other toolsConvention; not part of the agents.md spec

Two notes. First, .agents/skills is Lütke's shorthand for the skills-directory convention that Codex CLI and others use; it is not documented on the agents.md spec site, and Claude Code's analog is .claude/skills. Second, the practical difference agencies care about: AGENTS.md is portable across tools, CLAUDE.md is not. If a repo standardizes on AGENTS.md, every agent that reads it works from the same instructions.

The split-brain problem, explained

"Split brain" is what happens when one source of truth divides into per-tool copies. Team members use different coding agents, each reading only its own config file. Project context drifts: a build command, a test convention, or a security rule lives in AGENTS.md for the Codex users and in CLAUDE.md for the Claude Code users, and nothing guarantees the two stay in sync.

Lütke's follow-up post made the pain concrete:

But agents and claude files are recursively applied through the tree. With thousands of developed [developers] in a mono repo, it just does happen that one directory is missing one of the two files and this means that a subset of devs work with lobotomy.

We fix this with automation but it's a stupid complexity tax that shouldn't have to be paid.

That is the argument in two lines: a missing file means a subset of developers run their agent with incomplete context, and keeping every directory in sync across thousands of developers is a tax the toolmaker should not have imposed. This is not about whether Claude Code is good. It is about whether the config layer plays well with the rest of the ecosystem.

Anthropic's reply: "more hackable" Claude Code

The Claude Code team responded roughly three hours after the original post. Thariq Shihipar, a member of the technical staff on the team, wrote:

We are working on making Claude Code more hackable, which will include being able to easily use Agents.MD or make other system prompt modifications. I'll share more here when it's ready to roll out.

He also explained the product reasoning and gave the immediate workaround:

Our reasoning for this was that we don't think that model families are interchangeable and the system prompt can have a big impact on performance.

In the immediate term, you can always @Agents.MD from your Claude.MD.

Translation: native AGENTS.md support is on the roadmap, not shipped. Anthropic's position is that different models want differently formatted system prompts, so a single cross-tool file cannot capture what each model needs. That is a defensible engineering position. It is also, from a customer's seat, the exact reason the ecosystem standard exists.

The context: the feature request for AGENTS.md support (GitHub issue #6235, opened August 2025) had 377 comments and more than 6,400 reactions, and was closed as completed on August 17, 2026, eight days before Lütke's tweet, even though the docs still described bridges rather than native reading.

What agencies should do

For agency owners, this fight is not entertainment. Your choice of coding agent stack is a cost and a lock-in decision, and AGENTS.md interoperability is now a first-order selection criterion.

The pattern is familiar from every infrastructure fight of the last decade: whoever controls the format controls the switching costs. AGENTS.md is the industry's attempt to keep that power with the user. Agencies that standardize on it early will not have to re-platform when the next tool war starts.

Frequently asked questions

Does Claude Code read AGENTS.md?

No. As of August 25, 2026, Claude Code reads CLAUDE.md, not AGENTS.md. The official workaround is a CLAUDE.md containing @AGENTS.md, or a symlink (ln -s AGENTS.md CLAUDE.md).

What is the AGENTS.md standard?

An open, plain-Markdown file in a repo's root that tells coding agents how to build, test, and contribute. It is a "README for agents," used by more than 60,000 open-source projects and stewarded by the Linux Foundation's Agentic AI Foundation.

What is the difference between AGENTS.md and CLAUDE.md?

AGENTS.md is the open, cross-tool standard read by Codex, Cursor, Gemini CLI, Copilot, and others. CLAUDE.md is Claude Code's native, vendor-specific file. Claude Code does not natively read AGENTS.md.

Why doesn't Claude Code read AGENTS.md?

Anthropic says different model families need differently formatted system prompts for performance. The feature request is one of the most-reacted-to on its tracker, and the team said on August 25, 2026, it is working on making Claude Code "more hackable," including AGENTS.md support.

Why is Shopify banning Claude Code?

Shopify CEO Tobi Lütke posted that he is "thinking about banning Claude code at Shopify" until Anthropic reads AGENTS.md and .agents/skills, citing split-brain problems when different team members use different tools. As of August 25, 2026, it is a threat, not an enacted policy.

What is the split-brain problem in AI coding?

When team members use different coding agents that each read only their own config file, project context drifts across tools: one source of truth splits into per-tool copies that can fall out of sync.

How do I make Claude Code read AGENTS.md?

Create a CLAUDE.md containing @AGENTS.md (supports Claude-specific additions), or symlink it with ln -s AGENTS.md CLAUDE.md. On Windows, use the import; symlinks need admin or Developer Mode.

Choosing your agency's AI stack? Work through the decision framework in our guide to choosing an AI automation agency, and price the tooling with our AI coding agent pricing guide.

Choose your AI automation agency →

Sources

Accuracy note: Facts verified 2026-08-25 against the sources above (source pack t_89d4e5b0, citation coverage 89%). Quotes preserved verbatim, including Lütke's "Claude code" capitalization and the team's "Agents.MD"/"Claude.MD" spellings. Tweet metrics are time-of-fetch snapshots. The Shopify ban is a threat, not enacted policy, as of publish date.