OpenAI Codex CLI v0.150.0 Changelog
What changed in Codex CLI v0.150.0, released 2026-08-26 — explained in plain English.
- The terminal interface is more polished, with auto-generated task names, clickable links, and new keyboard shortcuts.
- Several important bugs were squashed, including security fixes and annoying shutdown freezes.
Want a heads-up when OpenAI Codex CLI updates?
We email you the moment a new OpenAI Codex CLI version drops — plus plain-English release notes like these. No spam.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
This update is all about making the Codex command-line tool smarter and easier to live with in your terminal. It matters most to developers and power users who use Codex to manage multiple AI tasks at once, giving them better ways to connect tasks, grab the code they need, and keep their workspace organized without fighting the interface.
What's new
- Task mentions: You can now use
@mentions to reference other Codex tasks. You can ask the AI to read, create, or message other tasks right from your terminal. - Smarter copy command: The
/copycommand now gives you a menu (a "picker") so you can choose to copy a full AI response, just a specific block of code, or a quoted section. - Automatic task names: If you start a terminal task without naming it, Codex will automatically give it a descriptive title. If you use the
/renamecommand, it will suggest a title based on the conversation that you can easily edit. - Clickable links: Web links in Markdown format will now show up as neat, clickable labels in terminals that support it. In older terminals, the full web address will still be visible.
- New shortcuts: You can now set up keyboard shortcuts to quickly switch between permission modes (how much freedom the AI has to run commands). If you use Vim mode for typing, you can now press
.to repeat your last edit. - Interrupt hooks: A new "Interrupt" hook feature lets you run specific commands or tools (called MCP handlers) if an active AI task gets interrupted or stopped before it finishes.
Fixes
- Security rules fixed: Untrusted projects can no longer sneak in their own instructions via an
AGENTS.mdfile. Also, a bug where "deny-read" rules (rules that stop the AI from reading certain files) were accidentally turned off when permissions changed has been fixed. - Better password protection: Improved how the tool hides sensitive passwords and credentials in its diagnostic logs, making sure secrets stay secret.
- Remote connection fixes: Fixed issues with finding and using security tokens (bearer-tokens) for remote MCP servers, making sure required servers start up correctly without breaking older versions of the tool.
- Windows username bug: Fixed an issue where setting up a secure sandbox (an isolated space for running code) and launching aliases failed if your Windows username contained special characters (Unicode).
- Shutdown freeze solved: Fixed a bug where Unix systems would hang during shutdown because background processes were stubbornly holding onto the terminal's output.
- Amazon Bedrock compatibility: Fixed conversation compaction (summarizing long chats to save memory) and multi-agent compatibility for Amazon Bedrock AI models.
Breaking changes
No breaking changes in this release.
How to upgrade
Update to OpenAI Codex CLI v0.150.0 by running:
codex --upgrade
This is our plain-language summary. Read the complete, official notes on the OpenAI Codex CLI official changelog ↗.
