Allow skills and prompts to show the structured question dialog outside Plan mode

Open 💬 0 comments Opened Aug 6, 2026 by matbishop

What variant of Codex are you using?

Desktop App on macOS

What feature would you like to see?

I would like skills and prompts to be able to show the same structured question dialog that Codex uses in Plan mode.

This is the dialog that presents selectable options, a recommended choice, short descriptions, and a free-form response option.

Currently, this dialog appears to be available only when I manually enable Plan mode before submitting my prompt. If I invoke a skill while the task is in Default mode, the skill cannot request this UI and must instead print the choices as ordinary chat text.

I am not necessarily asking for the skill or prompt to switch the entire task into Plan mode. I would prefer the structured question dialog to be available independently of Plan mode.

Example use case

I have a draft-commit skill that reviews changes and presents an approval decision such as:

  • Approve all
  • Commit only
  • Revise

When the task is already in Plan mode, Codex can present these choices using the structured dialog.

When the same skill is invoked in Default mode, it can only print the choices in the conversation and ask me to type a response. The skill cannot enable the dialog itself, and I need to remember to select Plan mode before invoking it.

Expected behavior

A skill or explicit prompt should be able to request structured user input during a Default-mode task.

For example:

  1. A skill reaches a decision or approval point.
  2. It requests structured user input.
  3. Codex displays the existing question dialog with selectable options.
  4. I select an option or provide a free-form answer.
  5. The current task continues in the same mode.

Displaying the dialog should not change permissions, enable additional tools, or automatically switch the task into Plan mode.

Why this would be useful

This UI is clearer and safer for:

  • Approval-gated workflows
  • Yes/no confirmations
  • Selecting between several scopes
  • Reviewing proposed external actions
  • Skills that must pause for an explicit user decision

It would also make reusable skills behave consistently regardless of whether the user remembered to enable Plan mode before invoking them.

Additional information

_No response_

View original on GitHub ↗