On March 26, OpenAI added plugins to Codex, its AI model for coding, giving developers a simpler way to package, share, and reuse workflows across projects and teams. For teams already using Codex for AI coding, the update reduces the setup work that is usually repeated from one machine to the next.
The plugin launch is only part of the picture. Codex already includes a wider set of features that make it more useful than a basic code assistant, and developers may be missing its importance.
Here are 10 Codex features developers should actually pay attention to.
1. Turn team setup into a reusable install
The clearest benefit of plugins is reuse. OpenAI says plugins can package skills, app integrations, and MCP server configurations for shareable workflows across the Codex app, CLI, and IDE extensions.
That gives teams a cleaner starting point. Instead of handing every new developer a long setup document and hoping they configure the same tools the same way, they can install one shared workflow and start from the same base.
2. Let Codex read terminal output while it works
Codex can read what’s happening in your terminal while you’re working.
That means it can check things like a running dev server or see why a build failed without you having to explain it.
You don’t need to copy logs or paste error messages into the chat anymore. Codex can follow what’s already happening in your terminal and use that to keep working.
3. Run parallel work on the same repo with worktrees
Codex supports worktrees, which let it handle multiple tasks in the same project without those tasks colliding. Each task gets its own isolated copy of the code while staying tied to the same repository.
That is useful when you want to try two different fixes, split maintenance work from feature work, or compare approaches without constantly juggling branches and local state.
4. Schedule recurring maintenance with automations
Codex also supports automations for recurring background tasks. These jobs can run in the background, add findings to the inbox, and close themselves when there is nothing to report. In Git repositories, they can run in the local project or on a dedicated worktree.
That opens the door for recurring engineering work that usually slips through the cracks, like scanning recent commits for likely bugs, summarising activity, drafting release notes, or checking flaky failures on a schedule.
Damilare Odedina
5. Connect Codex to outside tools with MCP
The Model Context Protocol gives Codex access to tools and external context. In Codex, that includes third-party documentation and developer tools, with MCP servers supported in both the CLI and the IDE extension.
That gives Codex a wider working environment. It is no longer limited to your prompt and your local code. It can work with more of the systems and references that shape real software teams.
6. Switch models depending on the task
Codex supports both GPT-5.4 and GPT-5.4 mini. OpenAI recommends GPT-5.4 mini for lighter coding tasks and subagents, while GPT-5.4 is better suited to more complex planning, coordination, and final review.
Not every coding task needs the same balance of speed and depth. A quick scan or a lightweight fix may be better with the smaller model. A harder architecture decision may need a stronger one.
7. Keep your setup aligned across app and VS Code
OpenAI says key settings are now synced between the Codex app and the VS Code extension, and there is now a settings entry point directly in the extension.
This solves a small but frustrating problem. When a coding tool behaves differently depending on where you open it, trust drops fast. Shared settings make Codex feel more consistent across clients.
8. Run Codex inside GitHub workflows
OpenAI publishes Codex GitHub Action. Teams can use it to run Codex in CI/CD jobs, apply patches, post reviews, gate changes on Codex-driven checks, and handle repeatable tasks such as code review, release prep, or migrations from a workflow file.
This moves Codex beyond the editor. It becomes part of the workflow around the code, not just a tool a developer opens during a session.
9. Steer Codex before it finishes the task
Codex supports mid-turn steering, which lets users send a message while it is still working in order to redirect the task.
That cuts down wasted time. When Codex is heading toward the wrong interpretation or approach, you can step in before it spends another ten minutes finishing work you already know you will not keep.
10. Turn repeated prompts into reusable skills
OpenAI says skills extend Codex with task-specific capabilities. A skill can package instructions, resources, and optional scripts so Codex can follow a workflow more reliably.
This is where the tool starts to build memory at the workflow level. Once a process becomes predictable, like incident triage, release preparation, checklist-based review, or a recurring maintenance routine, you can turn it into something reusable instead of explaining it from scratch every time.
Damilare Odedina
