Codex in the terminal gains tabs for side chats, and running codex exec in an unfamiliar folder now has to be explicitly allowed
OpenAI released two versions of its Codex CLI on a single day, with a new workspace and stricter handling of untrusted folders.
OpenAI released versions 0.168.0 and 0.169.0 of the Codex CLI on 13 November, the official changelog shows. Version 0.168.0 introduces an opt-in terminal workspace in which a side conversation appears next to the main chat, with resizable panes. In 0.169.0, several side chats can sit side by side there as tabs, and locally saved side chats reopen after a restart. Developers can also add their own model providers from the terminal, with code-based sign-in for those working over SSH or without a local browser.
The most important change for anyone using Codex in scripts or CI is in codex exec. In folders not marked as trusted, the --sandbox workspace-write option no longer enables project configuration and MCP servers. That now requires the --trust flag, also for codex review. A weekly roundup of coding agents on DEV Community highlighted that change as the main point of attention in the release, alongside a warning that the features.multi_agent setting is deprecated in favour of agents.enabled.
Smaller additions include a search function for older tasks in the agent overview, an option to collapse completed steps under each prompt and the ability to save a diagnostic ZIP file locally with Ctrl+S, even when uploads are disabled.