OpenAI Codex CLI v0.153.0 Changelog

This release is all about making the Codex command-line tool more reliable and flexible for developers. It matters most to programmers who use this tool daily to write code, especially those who like customizing their workflow with add-ons or keyboard shortcuts. A lot of the updates focus on making sure you don't lose your hard work if your internet connection hiccups, and giving you more control over how the AI remembers your conversation.

What's new

• Vim mode gets undo/redo: If you use Vim (a popular, keyboard-focused text editor), you can now press u to undo and Ctrl+R to redo. It perfectly saves everything you typed or pasted, even big chunks of text.
• Plugin manager: You can now use commands to list, install, and delete plugins (mini add-ons that give the tool extra features) from online marketplaces.
• Turn off auto-recaps: The tool usually gives you automatic summaries of what it has done. You can now turn these off in the settings (tui.auto_recap = false) if you find them annoying, but you can still ask for a summary manually by typing /recap.
• Better history viewing: Your history screen now shows the exact code changes (patches), the commands sent to background terminals, and individual completed commands, so you can see exactly what happened step-by-step.
• Usage warnings: If you are on a Plus or Team plan, you will now get a heads-up warning when you have used more than half of your allowed messages in a 5-hour window, so you don't run out unexpectedly.

Improvements

• Smarter context management (Experimental): There is a new, optional setting (features.context_management.experimental_mode) for certain ChatGPT Plus and Pro users. When turned on, it helps the AI manage its "memory" (token budget) better during long chats so it doesn't forget important details.
• New API settings: Developers can now use a new setting to ask the user questions in the background (request_user_input_async), and the system can now report which AI model and "reasoning effort" level is being used for a task.
• Cleaner settings: The setting to disable "paste burst" (which handles how fast pasted text is processed) has been moved under the [tui] section to keep things organized. The old way still works if you haven't updated your config file.

Fixes

• Reconnecting after drops: If your connection to the app server drops, the tool now reconnects automatically. It saves your unfinished drafts and chat history, and pauses anything it wasn't sure about so you can review it before it runs.
• Faster "Full Access" mode: When you let the tool run with "Full Access," it no longer forces simple "yes/no" confirmations through the Guardian (the safety checker). This speeds things up while keeping the important safety checks intact.
• Guardian memory fixed: The Guardian's review history now survives when the conversation gets too long and is compressed, when the app restarts, or when you branch off into a new chat.
• Plugin and Mac fixes: Remembered permissions for MCP (Model Context Protocol) tools are now tied to your specific account. Also, a bug that stopped certain plugins from starting correctly on Mac computers has been fixed.
• History compression fixes: The system now correctly compresses shared histories and can resume tasks even when that history is compressed. It also works correctly if you are using symlinks (shortcuts) for your session folders.

Breaking changes

No breaking changes in this release.