Queue, Steer, (new) Stash
What variant of Codex are you using?
App
What feature would you like to see?
Related to #2617 and #2966, but I think this is a slightly different workflow.
Right now, when Codex is working, follow-up messages mostly fall into two categories:
- Queue the message and let Codex process it after the current task finishes.
- Send/steer immediately when the message is urgent enough to interrupt or pivot the current task.
I would like to propose a third category: a session-level “stash” for follow-up prompts.
A stashed prompt would appear near the current queued messages, but it would not be submitted automatically. It would just stay there until the user explicitly sends it, edits it, merges it with another prompt, or deletes it.
The use case is: while Codex is working, I often think of follow-up questions, alternative directions, cleanup tasks, or “ask this next” ideas. They are relevant to this exact chat/session, but I do not necessarily want Codex to run them as soon as the current task finishes. First I want to see whether the current task completed correctly, whether more feedback is needed, and only then decide which follow-ups should actually be sent.
Today the workaround is to keep these ideas in the input box, move them to a separate scratch file, or put them into a general TODO list. None of those fit very well. These are not durable project TODOs, and they may not belong to the whole feature or repository. They are temporary, session-scoped prompts that belong next to the conversation.
Possible UX:
- queued prompt: automatically sent after the current task;
- urgent/steer prompt: sent immediately;
- stashed prompt: saved in the session, but only sent manually.
This would make the running-session workflow smoother without turning every idea into an automatic instruction for the model.
Additional information
_No response_
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I think this is related to #23085, but not a duplicate.
#23085 is mainly about autosaving the current unsent composer draft and preserving side-chat state across restarts/crashes. That solves the “don’t lose my draft” problem.
This request is about a separate workflow: keeping multiple session-scoped follow-up prompts in a manual stash while Codex is running. These prompts should not be automatically submitted after the current task finishes, unlike queued messages. They are temporary “maybe ask this next” items that belong to the current chat/session, and the user should explicitly choose when to send, edit, merge, or delete each one.
So autosaved drafts would protect the current composer state, while a prompt stash would manage several future prompts without turning them into an execution queue.