Codex Desktop: allow editing a terminal command inside the approval dialog

Feature request

Allow users to edit a proposed terminal command directly inside the Codex Desktop approval dialog before approving it.

Problem

Today, when Codex asks for permission to run a terminal command, the approval UI effectively gives the user two choices: approve the command exactly as proposed, or deny it.

If the command is almost correct but needs a small correction, the user has to:

1. Deny or dismiss the approval.
2. Return to the conversation.
3. Explain the correction to Codex.
4. Wait for Codex to regenerate the command.
5. Review a second approval request.

This adds friction to a workflow whose purpose is specifically to let the user inspect an action before execution.

We ran into this during ordinary development work when Codex proposed a PowerShell command that only needed a small syntax correction. The approval dialog displayed the command clearly, but there was no way to fix it in place.

Suggested UX

Add an editable command field or an Edit command action inside the approval dialog.

A possible flow:

• Deny
• Edit command
• Run modified command
• Reset to agent command

If the user changes the command, clearly mark the final command as Modified by you so there is no ambiguity about whether the model or the user authored what actually ran.

It would also be useful to let the user send a short correction/message back to Codex without dismissing the approval modal, e.g. "Yes, but add -ErrorAction SilentlyContinue."

Why this would help

This would make approvals both safer and faster:

• Users could correct an almost-right command instead of feeling pushed toward approving it unchanged.
• The exact command that will execute remains visible and reviewable.
• Small syntax/platform corrections would not require a full deny → chat → regenerate → reapprove loop.
• The approval surface becomes a true review step rather than a binary gate.

Related work

This is related to approval-editing requests such as #23227, but the request here is specifically about editing the actual proposed command in the Codex Desktop approval modal, not editing persistent sandbox/allow rules.

Start with the Codex Desktop approval dialog described in the issue and trace the entry point for proposed terminal commands. Define the editing, reset, modified-by-you, and optional correction-message flows, then verify that the final command remains visible for review before execution.