Key takeaways

  • The /plan slash command turns on plan mode in the Codex CLI, the IDE extension and the ChatGPT desktop app.
  • OpenAI's docs describe /plan and Shift+Tab as toggles, so you exit plan mode by switching it off again, not with /exit, which closes the CLI.
  • Codex stays in plan mode until the mode changes, and a request to implement made in plan mode is treated as a request to plan the implementation.

Codex plan mode is turned on with the /plan slash command in all three Codex surfaces: the Codex CLI, the IDE extension and the ChatGPT desktop app. To leave plan mode, you switch the mode off again with the same toggle (/plan, or Shift+Tab where the docs mention it), not by asking Codex to start coding. OpenAI's best practices guide puts it in one line: "Toggle with /plan or Shift+Tab."

That second part trips people up. Codex stays in plan mode until the mode is switched, and typing "go ahead and implement it" while still in plan mode only gets you a plan for the implementation. The details below come from OpenAI's Codex docs, changelog and the open-source openai/codex repository, as of November 16, 2026.

What Codex plan mode does

Plan mode makes Codex think before it edits. The best practices guide calls it "the easiest and most effective option" for complex or vague tasks, because it "lets Codex gather context, ask clarifying questions, and build a stronger plan before implementation."

The plan mode instructions in the openai/codex repository spell out what that means in practice:

  • Codex may read and search files, inspect configs and run tests or builds that only write to caches or build folders.
  • It may not edit or write files, run formatters that rewrite files, or apply patches and migrations.
  • It explores the repository first, then asks you questions about what you want, then finalizes a detailed plan in a <proposed_plan> block.

The same file says Codex is in plan mode "until a developer message explicitly ends it," and that a request to execute made in plan mode is treated as a request to plan the execution. In other words, the model won't leave plan mode on its own because you asked nicely. You have to switch the mode.

Plan mode is also separate from the checklist you sometimes see while Codex works. That checklist comes from a tool called update_plan, which tracks progress and, per the repository, "does not enter or exit Plan Mode."

How to enter plan mode in the Codex CLI

The Codex CLI slash command docs describe it in three steps:

  1. In an interactive Codex session, type /plan and press Enter. This switches the active chat into plan mode.

  2. Optionally, add your request on the same line, so Codex starts planning right away:

    /plan Propose a migration plan for this service
    
  3. You can paste content or attach images along with the inline request.

Codex then enters plan mode and treats your inline text as the first planning request. One catch: while Codex is already working on a turn, /plan is temporarily unavailable. Wait for the turn to finish, or interrupt it, before switching.

The best practices guide also lists Shift+Tab as a way to toggle plan mode. The CLI shows a hint for it: the changelog for Codex CLI 0.160.0, released October 1, 2026, added "the Plan mode cycling hint in the fullscreen status line."

How to exit plan mode in the Codex CLI

The CLI docs don't list a separate "exit plan mode" command. What they do document:

  • Shift+Tab toggles plan mode, according to the best practices guide. Press it again to switch back.
  • /plan is the command that switches into plan mode. In the IDE extension and the desktop app the docs describe /plan as a toggle; the CLI page describes it only as the way in.
  • Don't use /exit, /quit or Ctrl+C to get out of plan mode. The CLI docs say /quit and /exit close the whole CLI session, and Ctrl+C closes it too.

A few CLI changes affect how plan mode sticks around:

  • Codex CLI 0.156.0 (September 22, 2026) restores plan mode when you resume a session. If you resume an old chat and Codex refuses to edit, check whether it came back in plan mode.
  • Codex CLI 0.165.0 (October 27, 2026) stopped queued agent messages from restarting work after an interrupt or a switch to plan mode.

Plan mode is a different setting from your approval mode, which decides whether Codex asks before it runs commands or edits files. If you want Codex to work without asking at each step once the plan is done, see how to enable auto mode in Codex.

How to use plan mode in the Codex IDE extension

In the VS Code extension, /plan is listed in the IDE slash commands as "Toggle plan mode for multi-step planning."

  1. In the Codex composer, type /.
  2. Pick /plan from the list, or keep typing to filter it.
  3. Press Enter. Plan mode is now on.

To exit, run /plan again the same way. Because it is a toggle, the second run turns plan mode off. The IDE docs don't mention Shift+Tab for the extension, so use the slash command there.

How to use plan mode in the Codex app

The Codex app now lives inside the ChatGPT desktop app: you choose Codex when you start a chat, then work in its composer. The desktop app's slash command reference lists /plan with the same description as the IDE: "Toggle plan mode for multi-step planning."

  • To enter: type /plan in the composer and press Enter. The prompting guide recommends this for multi-step tasks "when you want Codex to investigate and propose an approach before editing." The app's Add menu also has a Plan mode entry, shown on the features page as "Turn plan mode on."
  • To exit: run /plan again to toggle it off.

Two app updates in the changelog help here. Codex app 26.210 (February 10, 2026) added "clearer guidance for entering plan mode when you type plan in the composer." Codex app 26.317 (March 18, 2026) added notifications for plan mode questions, so you notice when Codex is waiting on your answer.

Plan mode in each Codex surface

Surface Enter plan mode Exit plan mode Documented shortcut
Codex CLI /plan, optionally with a prompt Toggle with Shift+Tab Shift+Tab (best practices guide)
IDE extension /plan from the composer Run /plan again None documented
ChatGPT desktop app (Codex) /plan, or Plan mode in the Add menu Run /plan again None documented

Plan first, then set a goal

OpenAI's docs pair plan mode with Goal mode, which keeps Codex working toward one objective across a long task. The desktop app's slash command page says: "To define the goal with ChatGPT first, start with /plan, then set the refined goal with /goal." The goal docs give the same advice for the CLI and the IDE extension when the outcome is still unclear.

So a typical flow for a bigger change is: /plan with your request, answer Codex's questions, read the proposed plan, switch plan mode off, then either tell Codex to implement it or start a /goal built from the plan.

Set a different reasoning effort for plan mode

Plan mode can think harder than your normal chats. The configuration reference documents a plan_mode_reasoning_effort key for ~/.codex/config.toml, described as a "Plan-mode-specific reasoning override." It has to be a level the selected model supports. When it isn't set, plan mode uses its built-in preset default. The sample configuration shows it like this:

plan_mode_reasoning_effort = "high"

Long planning sessions add up, so keep an eye on your limits. To see how much you have left, read how to check token usage in Codex.

Frequently asked questions

What is the Codex plan mode shortcut?

OpenAI's best practices guide says plan mode can be toggled with /plan or Shift+Tab. The Codex CLI shows a plan mode cycling hint in its fullscreen status line since version 0.160.0.

How do I exit plan mode in Codex?

Toggle it off with the same control you used to turn it on: Shift+Tab in the CLI, or /plan again in the IDE extension and the desktop app. Don't use /exit or /quit, which close the whole CLI session.

Does Codex plan mode work in VS Code?

Yes. The Codex IDE extension lists /plan in its slash commands as a toggle for multi-step planning. Type / in the Codex composer, pick /plan and press Enter.

Can I use a higher reasoning effort in Codex plan mode?

Yes. Set plan_mode_reasoning_effort in ~/.codex/config.toml to a level your model supports, such as "high". When it is unset, plan mode uses its built-in preset default.

Sources

  1. Slash commands in Codex CLIOpenAI
  2. Best practicesOpenAI
  3. Codex IDE extension slash commandsOpenAI
  4. Slash commands (ChatGPT desktop app)OpenAI
  5. PromptingOpenAI
  6. FeaturesOpenAI
  7. Configuration referenceOpenAI
  8. Sample configurationOpenAI
  9. ChangelogOpenAI
  10. Plan mode collaboration template (plan.md)OpenAI on GitHub

Add ClaudeGPT Si as a preferred source on Google