Appending prompts with Esc should not pause active goal execution

Open 💬 0 comments Opened Jul 6, 2026 by Wizmann

What variant of Codex are you using?

CLI

What feature would you like to see?

Summary

When Codex is pursuing a goal, users can type additional prompts and press Esc to append those prompts to the AI. However, doing this currently pauses the active goal.

From a UX perspective, this feels surprising: appending prompts seems like it should provide extra context or guidance while Codex continues working, not interrupt the goal entirely.

Reproduction Steps

  1. Start Codex on a goal.
  2. While Codex is pursuing the goal, type additional prompts or guidance.
  3. Press Esc to append the prompts to the AI.
  4. Observe that the active goal becomes paused.

Actual Behavior

The goal is paused after the prompts are appended to the AI.

Expected Behavior

The prompts should be appended as additional context or instructions, and Codex should continue pursuing the active goal.

Codex may briefly acknowledge the new prompt, but the goal should not pause unless the user explicitly requests a pause, stop, cancel, or goal replacement.

Example

Initial goal:

Refactor the authentication flow and add tests.

Prompt typed while Codex is working:

Also handle expired sessions in the token refresh logic.

After pressing Esc, Codex should incorporate that instruction and continue the refactor.

Instead, the goal is paused.

Impact

This makes it harder to steer Codex during longer-running tasks. Users often think of additional requirements while a goal is already in progress, and appending those prompts should feel like guiding the agent mid-task rather than interrupting it.

The current behavior can also make users hesitant to provide useful mid-task context, because doing so may unexpectedly stop progress.

Suggested UX

Treat prompts appended with Esc during goal execution as non-blocking goal injections by default.

Only pause the goal if:

  • The user explicitly asks to pause, stop, or cancel.
  • The new prompt clearly replaces the current goal.
  • The prompt creates a conflict that requires clarification before continuing.

A lightweight acknowledgment would be enough, for example:

Got it — I’ll incorporate that and continue.

Additional information

_No response_

View original on GitHub ↗