REAL

Holds up: one AGENTS.md really can feed both Codex and Claude Code, because Codex reads it natively and Claude Code v2.1.277+ reads it when no CLAUDE.md exists, and the six-section rule set is sound practice

Facebook · Oct 2026

The claim'One AGENTS.md file can give Claude Code and Codex the same working rules (six sections, 42 lines).' Codex reads AGENTS.md natively, and recent Claude Code versions read it when the project has no CLAUDE.md. Sections: plan before touching code (PLAN.md), smallest change that works, split work across subagents, own the bug, verify before saying done, write down every correction under Lessons. 'If you keep a CLAUDE.md, check that it points Claude Code to AGENTS.md.' Credits: @Voxyz_ai.

Both technical claims check out against the vendors' own docs. OpenAI's Codex docs say Codex builds an instruction chain from AGENTS.md files on every run: a global one in ~/.codex, then each folder from the Git root down to where you are, capped at 32 KiB by default. Anthropic's Claude Code memory docs say that from v2.1.277 Claude Code reads AGENTS.md as your project instructions when there is no CLAUDE.md, .claude/CLAUDE.md or CLAUDE.local.md in the working directory or above it. If a CLAUDE.md exists, Claude Code reads only that, unless it imports AGENTS.md with an @AGENTS.md line or you change the Project instructions setting in /config. So the post's last tip, make your CLAUDE.md point at AGENTS.md, is exactly what the docs recommend. One trap the post misses: adding a personal CLAUDE.local.md silently switches AGENTS.md off for you. The six sections are good, mainstream agent hygiene: plan first, small diffs, backups before deletes, reproduce bugs, run the checks, log corrections. They are instructions, not enforcement; Anthropic says to use hooks for anything that must never happen. Nothing is sold here; it is a save-and-follow post that credits its source, and the wording matches the style of @Voxyz_ai, an X account that posts Claude Code and Codex rule files, though the exact original thread was not located.

What holds up

Anthropic docs (code.claude.com/docs/en/memory, read Oct 8, 2026): 'Claude reads AGENTS.md only when you have no CLAUDE.md in your working directory or above it'; requires v2.1.277 or later; CLAUDE.local.md counts as a CLAUDE.md; the setting claude-md-and-agents-md loads both.
OpenAI Codex docs (learn.chatgpt.com/docs/agent-configuration/agents-md): Codex reads ~/.codex/AGENTS.md plus project AGENTS.md files from the Git root down, supports AGENTS.override.md, and stops adding files past project_doc_max_bytes (32 KiB default).
Post text read in full: six sections, about 42 lines, covering PLAN.md, minimal diffs, subagent roles, bug reproduction, verification, and a Lessons log; ends 'your review still decides what ships.'
@Voxyz_ai exists on X ('I bridge visual AI and coding tools') and publishes CLAUDE.md and AGENTS.md rule blocks; the specific 42-line thread was not found, so attribution is taken from the post.

What doesn’t

A CLAUDE.local.md in the repo turns off AGENTS.md reading by default; set Project instructions to claude-md-and-agents-md or import it.
Rule files are context, not guardrails: an agent can still ignore 'back up before deleting'; use a PreToolUse hook or Codex sandboxing for hard limits.
Codex-only files such as AGENTS.override.md and AGENTS.local.md are not read by Claude Code.

The catch

The one-file setup works, but only if you know which file each tool is actually loading. Run /context in Claude Code to confirm AGENTS.md is listed under Memory files, and keep the shared file short so it fits Codex's size cap and stays readable.

How to actually do it

Put the shared rules in AGENTS.md at the repo root; keep it well under 32 KiB.
If the repo also needs a CLAUDE.md, make its first line @AGENTS.md so Claude Code imports it, then add only Claude-specific notes below.
Open a Claude Code session and run /context to confirm AGENTS.md loaded; in Codex, ask it to summarize its current instructions to confirm the same.
Move any rule that must never be broken (no deletes, no pushes) into a Claude Code PreToolUse hook or Codex approval settings instead of prose.

Verified against Anthropic's Claude Code memory docs and OpenAI's Codex AGENTS.md docs on Oct 8, 2026. The post is accurate and its advice is sound; the only gaps are the CLAUDE.local.md trap and the difference between written rules and enforced ones.

Confidence: High
Posted by: a member of a Claude AI community Facebook group, small account, reposting a framework credited to @Voxyz_ai on X; post sent to Buddy Mon Oct 5, 2026, ~10:10 PM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.