Codex CLI shipped version 0.147.0 on August 7, 2026, and it is a bigger release than the version number suggests. Three separate improvements landed at once: a new way to organize your growing pile of conversations, a middle ground between manual approvals and full autonomy, and a smarter way to bring your Cursor setup along when you switch tools. If you have been putting off this codex cli update, here is what changed and why it matters, and how it stacks up against Claude Code in the ongoing agentic CLI race.
Key Takeaways
- Codex CLI 0.147.0 (shipped August 7, 2026) adds persistent, manually ordered sections so heavy users can group threads by project instead of scrolling one chronological list.
- A new --approve-for-me flag cuts approval fatigue by skipping manual prompts on routine, low-risk actions, while an automated review layer still checks every action behind the scenes.
- The older --full-auto flag was removed in favor of explicit --sandbox workspace-write configuration, positioning --approve-for-me as a middle ground between constant approvals and full autonomy.
- The /import command now pulls in Cursor-managed skills and gained duplicate-free sync, turning it from a one-time migration tool into an ongoing way to keep tools like Cursor and Codex aligned.
- This is the third consecutive release to expand thread management, following pinning and forking in 0.146.0 and paginated history in 0.145.0, reflecting how running many concurrent agent sessions has become the norm.
Learn this hands-on
Go from idea to a live product with real users. Build and ship your first SaaS with Claude Code in a live cohort, with auth, payments and deployment done by the end. Join the Ship Your First SaaS with Claude Code cohort.
Persistent Sections and Faster Transcript Browsing
The first problem this release tackles is one every heavy Codex user runs into eventually: too many threads, no good way to organize them. Before 0.147.0, Codex CLI showed one continuously scrolling, chronological list of conversations. If you were juggling several projects at once, finding the thread you wanted meant hunting through an interleaved timeline sorted by recency, not by what you were actually working on.
Codex CLI 0.147.0 introduces persistent, manually ordered sections. You define named groupings yourself and arrange them in whatever order makes sense to you, rather than relying on Codex to sort things automatically. If you run three parallel projects, you can now keep each project's threads together in its own section instead of scrolling past unrelated work to find the thread you need.
This is not an isolated feature. It is the third consecutive release to expand Codex CLI's thread management system. Version 0.145.0 added paginated thread history with search and persisted names. Version 0.146.0, released July 29, added thread pinning, forking, and the ability to switch between side conversations without losing your place in the main one. Sections are the next layer on top of that: pinning surfaces a handful of important threads, forking branches off a single conversation, and sections let you impose structure across your entire thread list.
Alongside sections, 0.147.0 also adds incremental loading for long transcripts. Previously, opening a long, tool-heavy session meant waiting for the entire history to load before you could scroll. Now transcripts load progressively as you scroll, cutting wait times and memory overhead on your longest sessions. It pairs naturally with the paginated thread history from 0.145.0, extending the same performance-minded approach from the thread list itself down into individual conversations.
Both changes target the same underlying shift: as Codex has added longer-running agent capabilities, multi-agent orchestration, and remote execution, running many concurrent sessions has become the norm rather than the exception. Simple chronological lists stop scaling once you are past a handful of threads, and this release is Codex's answer to that. It is worth comparing this approach to how Claude Code handles parallel agentic sessions, where thread isolation and session management have been a differentiator against competing CLIs for a while.
The New --approve-for-me Flag
The second change addresses a much more universal complaint about agentic coding CLIs: approval fatigue. The more autonomous a Codex session becomes, the more often you get pulled back in just to approve a routine command or file edit.
Codex CLI 0.147.0 introduces a new --approve-for-me flag that lets Codex proceed through routine actions without stopping for a manual yes or no on each one. Critically, this is not a blanket bypass of safety checks. Actions still pass through an automated review layer behind the scenes, what changes is that you are no longer interrupted for every low-risk decision.
That distinction matters because this flag ships in the same release that removed the older --full-auto flag entirely, in favor of the more explicit --sandbox workspace-write configuration. Where a pure skip-everything mode trades away oversight altogether, --approve-for-me is positioned as a middle ground: it removes the friction of constant prompts while keeping an automated review pass in the loop.
For anyone running long or multi-step Codex sessions, this is the kind of change that compounds. Fewer interruptions during routine work means fewer context switches, which matters more the longer and more autonomous your sessions get. If you have been weighing how far to push autonomous execution in your own workflow, it is worth reading about what actually changed when Claude Code's auto mode shipped, since the same trade-off between speed and oversight applies here.
Cursor Skill Import With Duplicate-Free Sync
The third piece of this release is aimed at developers running a mixed toolchain, Cursor for some work, Codex or Claude Code for other work, and tired of manually rebuilding their setup every time they switch.
Codex's /import command already handled Claude Code migration and, since version 0.145.0 in July 2026, Cursor configuration (MCP servers, plugins, sessions, and project-scoped memories). With 0.147.0, /import can now also pull in Cursor-managed skills directly. Skills are one of the more time-consuming assets to rebuild by hand, so making them portable lowers the switching cost for anyone evaluating Codex against other coding agents. This same problem, packaging and sharing reusable skills across a team rather than rebuilding them per tool, is exactly what Claude Code's plugin system was built to solve.
The second half of this change is arguably more useful day to day: duplicate-free sync. Previously, re-running /import against a Claude Code or Cursor conversation you had already imported could create a repeat copy in Codex's thread list. Now, re-running the command updates the existing conversation in place instead of duplicating it. That turns /import from a one-time migration tool into something closer to an ongoing sync mechanism between the tools you use side by side.
Getting Codex CLI 0.147.0
All three changes ship in the same release and are available now:
npm install -g @openai/codex@0.147.0
None of the three changes depend on each other. You can start using the new thread sections without touching --approve-for-me, and vice versa. But together they point at the same direction Codex has been heading over its last several releases: fewer interruptions, better organization for the sessions you keep running, and less friction moving between the AI coding tools already in your stack. If you run Codex CLI daily, this is a release worth pulling down.
If you are building out your own command-line AI workflow from scratch, mastering the terminal and Git fundamentals first will make every one of these agentic CLIs, Codex included, far easier to drive. And if Claude Code is the tool you are standardizing on for daily agentic work, our How to Master Claude Code series walks through shipping code faster and building your own AI agents on top of it.
Full release notes are available on GitHub.
