OpenAI Codex Plugin for Claude Code — Setup, Commands, and Who Benefits

Two frontier coding agents. One terminal. OpenAI built the bridge months ago — X just noticed.

On July 12, 2026, Shubham Saboo (@Saboo_Shubham_) posted that OpenAI dropped a Codex plugin for Claude Code — use Codex inside Claude for code reviews or delegated tasks. The tweet hit 63k+ views and linked openai/codex-plugin-cc.

Replies were immediate: "4 months ago… stop clickbaiting." "Fable 5 + gpt-5.5 medium is pretty good." "Two competing products are now teammates — who benefits?"

explainx.ai's read: not a launch, but the cleanest official pattern for the Codex vs Claude Code split developers already run — plan in Claude, execute or review in Codex — without bash duct tape.

TL;DR — what people are asking

Not "just dropped" — timeline vs hype

explainx.ai standard: treat viral reposts as discovery events, not ship dates. The plugin matters because it is official, maintained, and documented — not because July 12 was day zero.

What you get — commands at a glance

From the official README:

Background pattern (recommended for heavy reviews):

/codex:review --background
/codex:status
/codex:result

Natural language also works: "Ask Codex to redesign the database connection to be more resilient."

Install — copy-paste sequence

In Claude Code:

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setup

Requirements:

• ChatGPT subscription (incl. Free) or OpenAI API key
• Node.js ≥ 18.18
• Global Codex CLI: npm install -g @openai/codex
• Auth: !codex login if /codex:setup fails

After install you should see slash commands plus the codex:codex-rescue subagent in /agents.

Config inheritance: plugin reads ~/.codex/config.toml and trusted project .codex/config.toml — same models and reasoning effort as standalone Codex. Example project override:

model = "gpt-5.4-mini"
model_reasoning_effort = "high"

Typical workflows

1. Ship gate — dual review

# Claude implements with Fable 5
/codex:review --base main
/codex:adversarial-review --base main challenge retry and cache invalidation assumptions

Matches Gokul Rajaram's split: Claude for planning, Codex for review and execution.

2. Rescue lane — cheap second implementer

/codex:rescue --model gpt-5.4-mini --effort medium investigate the flaky integration test
/codex:rescue --background fix the failing test with the smallest safe patch

spark maps to gpt-5.3-codex-spark for fast passes per README.

3. Session transfer — leave Claude, continue in Codex App/TUI

/codex:transfer
# → prints codex resume <session-id>

Uses Codex's external-agent session importer — same rules as importing Claude history in the Codex desktop app.

4. Theo-style routing in CLAUDE.md

explainx.ai already documents this in what is CLAUDE.md:

• Fable 5 / Opus — taste, plans, user-facing design
• gpt-5.5 via plugin — bulk implementation, migrations, data work
• /codex:review — non-destructive quality pass
• /codex:adversarial-review — skeptical design review with custom focus text

No custom bash wrappers — native slash commands inherit Codex config.

One-way bridge — why Codex→Claude is hard

@iret77 (Jul 12 reply):

"you can simply ask Claude to call codex, grok and gemini as long as they are installed and authenticated… But when I try to call claude from codex, the tool call gets blocked hard."

explainx.ai read:

Anthropic and OpenAI both protect their harness — the plugin is distribution for Codex usage inside Anthropic's IDE, not neutral orchestration. ChatGPT Work vs Codex already showed OpenAI merging surfaces; this is the cross-vendor version of "stay in one UI, burn two quotas."

Who benefits — "two competitors as teammates"

@AgentOrToy asked who wins when Fable and Codex share a session.

CodeBucks reply: "Fable 5 + gpt 5.5 medium is pretty good combination" — cheap Codex passes under Fable planning.

Risk: limit whiplash on both sides if you chain background rescues + review gates during promo weeks.

Review gate — read the warning

/codex:setup --enable-review-gate

When enabled, a Stop hook runs targeted Codex review on Claude's response. Issues block stop until fixed — can loop Claude/Codex until quotas empty.

OpenAI README:

"The review gate can create a long-running Claude/Codex loop and may drain usage limits quickly."

Use only when actively monitoring — not overnight on Fable July 19 cliff anxiety runs.

vs rolling your own CLI calls

For enterprise, pair plugin use with DCG destructive-command guards on whichever agent touches prod.

FAQ from GitHub — quick answers

Same Codex account? Yes — local CLI auth shared.

Separate runtime? No — same machine, same checkout, same config.toml.

API key / custom base URL? Yes — set openai_base_url in Codex config.

Resume in Codex app? codex resume <session-id> from /codex:result.

Summary

openai/codex-plugin-cc is OpenAI's official Claude Code marketplace plugin — not new in July 2026, but the right integration if you already run Claude Code + Codex as a dual-agent stack. Install via marketplace commands, use /codex:review and /codex:rescue for read-only and implementation delegation, /codex:transfer to hop into Codex natively. Claude→Codex works; Codex→Claude mostly doesn't. Fable 5 + gpt-5.5 medium is the community sweet spot. Ignore "just dropped" hype — do enable the plugin if you pay for both harnesses anyway.

Related on explainx.ai

• Codex vs Claude Code — developer verdict
• ChatGPT Work vs Codex — harness split same week
• CLAUDE.md routing — Fable + codex-plugin-cc mechanics
• Codex slash commands reference
• AI coding plugins ecosystem — OpenAI + Anthropic
• GPT-5.6 Sol in Claude Code via claudex
• Fable advisor + Sonnet executor patterns
• Loop engineering — when to stay single-agent

Sources: openai/codex-plugin-cc · @Saboo_Shubham_ July 12, 2026 · Codex v1.0.6 release

Plugin commands and model aliases follow the GitHub README as of July 13, 2026. Verify /plugin marketplace availability in your Claude Code build before production rollout.