OpenAI has published a Codex plugin for Claude Code on GitHub, letting developers run code reviews, adversarial challenges, and background task delegation through Codex without leaving the Claude Code environment they already work in.
The AI coding tools market has been defined by rivalry. OpenAI's Codex and Anthropic's Claude Code compete for the same developers, the same workflows, and increasingly the same enterprise contracts. Which makes it genuinely surprising that OpenAI has shipped an official plugin that brings Codex directly into Claude Code, letting users leverage both tools from a single interface. Whether you read that as pragmatic interoperability or a quiet confidence move depends on your vantage point. Either way, the plugin is real, it is live on OpenAI's GitHub, and it changes the practical calculus for developers who have been forced to choose between the two.
The plugin surfaces through a set of slash commands inside Claude Code. The most straightforward is /codex:review, which triggers a standard read-only code review from Codex on whatever you are working on. That alone covers the most common reason a Claude Code user might want to reach for Codex: a second opinion on code quality, logic, or structure from a model trained differently and likely to surface different observations. Two AI reviewers with distinct training backgrounds catching different categories of issues is a meaningful practical benefit, not a redundancy.
The more interesting command is /codex:adversarial-review. This is not a standard code review. It is a steerable challenge review, meaning Codex is prompted to actively look for weaknesses, edge cases, security considerations, and logical failures rather than simply assessing code quality in a balanced way. For developers shipping to production, having a model that is specifically trying to break your code rather than evaluate it neutrally is a different and often more useful kind of feedback. The adversarial framing is a deliberate design choice that reflects how serious code review actually works when the stakes are high.
The background job commands round out the practical utility. /codex:rescue, /codex:status, /codex:result, and /codex:cancel allow developers to delegate work to Codex and manage those tasks asynchronously without interrupting the primary Claude Code session. This matters for workflows involving longer-running tasks where waiting for a synchronous response would break concentration. Delegating a refactoring job or a test generation task to Codex in the background while continuing to work in Claude Code is the kind of workflow improvement that sounds minor until you have been doing it for a week and realize how much friction it removes.
The requirement to have both a Claude Code setup and a ChatGPT subscription means the plugin is aimed squarely at professional developers who are already paying for multiple AI tools, which is increasingly common. The cost of running two AI subscriptions has dropped significantly as pricing competition has compressed rates across the market. For anyone already holding both subscriptions, the plugin effectively unlocks a capability they were not using simply because the tools lived in separate interfaces.
What This Signals About the AI Developer Tool Market
OpenAI publishing this plugin from its own GitHub account carries a different weight than a third-party integration. This is not a community hack or an unofficial workaround. It is a deliberate product decision to build a bridge into a competitor's environment. The most straightforward reading is that OpenAI is confident enough in Codex's capabilities that exposing it inside Claude Code represents an opportunity rather than a risk. Developers who try Codex through the plugin and find it useful for specific tasks may migrate more of their workflow toward it over time. The plugin is, in effect, a distribution channel that reaches Claude Code's user base directly.
There is also a broader market signal here about where the developer tools space is heading. The era of any single AI coding tool capturing an entire workflow is giving way to a more modular reality where developers assemble the combination of models and tools that best fits their specific needs. Plugins, integrations, and cross-platform commands are the infrastructure of that modular world. OpenAI building one for Claude Code rather than waiting for Anthropic to build one for Codex suggests the company understands that distribution in this environment requires meeting developers where they already are, not waiting for them to come to you.
For Claude Code users who have been curious about Codex but reluctant to switch contexts to evaluate it, the plugin removes that friction entirely. Try the review commands on your next pull request, run an adversarial review on something you are about to ship, and see whether the output is useful enough to make the subscription worth keeping. That is a much lower-friction evaluation than standing up a separate workflow, and it is exactly the kind of trial that converts curious developers into regular users.
Also read: Unsloth's custom kernels make LLM fine-tuning viable on consumer GPUs • GPT Image 2 disinformation arrives within days of the model's launch • A student just beat vector databases on AI memory with structured storage
