Codex CLI 0.169.0 shipped on 13 November 2026 and changes how codex exec works. In a folder without trust, project configuration and MCPMCPA standard that lets an AI program use your tools and data under clear permissions.Open the glossary servers stay off, also when you use --sandbox workspace-write. Scripts that need project MCP servers now need the new --trust flag or trust that you saved before.

This release came less than 5 hours after 0.168.0 on the same day. The 2 releases contain 469 change entries: 226 in 0.168.0 and 243 in 0.169.0, according to the official Codex release notes on GitHub. This article continues our review of Codex CLI 0.167.0 and the agent command center.

What is project trust in Codex CLI?

Project trust is a permission that lets Codex read the configuration and start the MCP servers of a project folder. A folder without this permission is untrusted. Codex ignores configuration files in that folder, because the files can come from a repository that someone else wrote.

The risk is real. A project configuration can register MCP servers, and an MCP server is a command that runs on your machine. If MCP is new to you, read our MCP (Model Context Protocol) guide first.

A row of closed teal lockers with one amber key that opens one locker, which shows trust given per project
Trust works per project folder. Only a folder with permission can load its configuration and MCP servers. Illustration: Rama Digital.

What changed in codex exec and codex review?

From 0.169.0, codex exec does not give trust automatically when the sandbox is set to workspace-write. This change came through PR #60041 in the openai/codex repository, merged on 7 November 2026.

ConditionBefore 0.169.0From 0.169.0
Untrusted folder with --sandbox workspace-writeTrust could become active without explicit permissionProject config and MCP stay off, Codex shows a warning
Untrusted folder with --trustFlag did not existTrust is saved for this run and later chats
--trust with --ephemeralFlag did not existTrust stays saved
--trust on a project marked untrustedFlag did not existRefused
--trust with --ignore-user-configFlag did not existRefused

The --trust flag is also available in codex review. The 0.169.0 help text says that this trust also applies to MCP commands outside the sandbox. Use this flag only in repositories whose content you know.

Who does this change affect?

The change affects most the teams that run codex exec from cron, scripts, or local CI in folders that nobody opened in an interactive session.

  • Scripts that make a new checkout or worktree on each run. A new folder has no trust.
  • Pipelines that use MCP servers from project configuration, for example a database MCP or a CMS MCP.
  • Headless servers, for example a VPS or a build machine, where people seldom open the Codex TUI.

Interactive TUI users in their own repositories will see almost no change. Trust for those repositories is usually saved from the first session.

Which new features matter in daily use?

The feature with the most daily effect is side chat, a side panel where you ask a different question without stopping the main task.

  • Side chat (0.168.0): turn on features.tui_workspace. In a terminal of 100 columns or more, the main chat and the side chat show next to each other. Drag the divider to change the width, or press leader z to zoom the active pane.
  • Side chat tabs (0.169.0): open more than one side chat as tabs. Switch tabs with the keyboard or the mouse, and open local side chats again after a Codex restart.
  • Collapse turn details (0.169.0): fold completed activity under each prompt. Answers stay visible, and you can still search tool output.
  • Search old tasks (0.169.0): search tasks in the agentAI agentAn AI program that performs work steps by itself, for example reading a message, drafting a reply, and recording the result.Open the glossary command center by name, prompt preview, or working directory.
  • Model providers (0.168.0 and 0.169.0): select the default provider for new chats, then add or edit custom providers from the terminal. Old chats keep their saved provider.
  • Diagnostic report (0.169.0): press Ctrl+S in the report editor to save a diagnostic ZIP locally, also when uploads are off.

Example scenario: a developer asks Codex to fix a failing test. While the task runs, the developer opens a side chat to ask about one error message. The main task continues, and the side question does not go into the context of the main task.

A teal train moves on the main track while one amber car turns onto a side track at a switch, which shows side chat
Side chat works like a side track: the side question runs on its own, and the main task continues. Illustration: Rama Digital.

This feature is opt-in. A terminal narrower than 100 columns does not show the panes next to each other.

Which problems have users reported?

In the 2 days after the release, users opened several issues that name 0.169.0 directly on GitHub openai/codex.

IssueSymptomWho is affected
#61568 (14 Nov)TUI startup uses 5 to 9 GB of RAM in the first 15 seconds. The reporter says 0.166.0 used less than 1 GB.Users with a large thread history, reported on WSL2
#61310 (13 Nov)Automatic review approved wsl.exe -u root, so the agent got root access without a password.WSL2 users with automatic approval review
#61485 (14 Nov)Skills from account plugins are missing in about 90% of new codex exec sessions.Users of account-level plugins
#61554 (14 Nov)The input area moves 1 row when the hint row disappears.Users with a custom status line

All 4 issues are open as of 15 November 2026. Users wrote these reports, and OpenAI has not confirmed them.

What did our check on the build server show?

We checked Codex CLI on our Linux build server on 15 November 2026, and the installed version was codex-cli 0.169.0.

The version manager on our server installed 0.169.0 automatically at shell login. Thus our scripts used the new trust rule with no manual upgrade step. The help text of codex exec --help and codex review --help both contain --trust, with a note that trust stays saved in --ephemeral mode. We did not test a model call in this check.

How do you upgrade safely?

A safe upgrade starts with a list of your codex exec scripts, not with the install command.

  1. Check the version with codex --version. Write down the old version for rollback.
  2. Find all scripts that call codex exec or codex review. Mark the scripts that need MCP or project configuration.
  3. Run each marked script 1 time after the upgrade. Read the warning that project configuration is off.
  4. For repositories that you know, save trust in an interactive session, or add --trust to the script on purpose.
  5. On WSL2, turn off automatic approval review for escalated commands until issue #61310 is closed.
  6. Monitor RAM at TUI startup. If RAM goes up by several GB, add your data to issue #61568 before you decide on a rollback.

What is the operational conclusion?

The upgrade to 0.169.0 is a good step for most users, because the trust rule closes a path where project configuration became active without permission. Before you upgrade an automation machine, check the codex exec scripts that use project MCP servers. Teams that want to use agents such as Codex in business processes with clear permission control can read about our AI business process automation service.

FAQ

Will my codex exec scripts break after the upgrade to 0.169.0?

The scripts continue to run. But project configuration and project MCP servers stay off in folders without trust, so the result can change if a script needs those MCP tools.

Is --trust safe in every repository?

No. The --trust flag saves permission for this run and later chats, including MCP commands outside the sandbox. Use it only in repositories whose content you know.

How do I turn on side chat in Codex CLI?

Turn on features.tui_workspace in the Codex configuration. Side chat shows next to the main chat in a terminal that is 100 columns wide or more.

Do I need to wait for the next release?

Wait if you use WSL2 with automatic approval review, or if TUI startup uses a lot of RAM. Other users can upgrade now after they check their codex exec scripts.