request_user_input glitches out when checking /agents

What version of Codex CLI is running?

0.104.0

What subscription do you have?

Pro

Which model were you using?

Codex 5.3 with Spark Subagents

What platform is your computer?

Linux

What terminal emulator and version are you using (if applicable)?

_No response_

What issue are you seeing?

I am having Codex glitch out and ask me the same sets of questions it asked during plan mode, except it asks the same questions, mid implementation, during the implementation phase of a plan (eg in execution mode, literally halfway through the plan).

Whenever I switch to /agents to check on a subagent's work, when I flip back to the main orchestration agent, it pauses and asks me the same questions I already answered in the planning phase (even though I already answered all of them, and switch out of plan mode).

This requires me to either answer them all again, or to ESC out and tell it to resume.

What steps can reproduce the bug?

1. Enter plan mode
2. Generate plan (ensure it uses request_user_input tool)
3. At the end of the plan, ESC out
4. Ask it to save plan to file
5. Tell it to implement plan using subagents exclusively
6. When it spawns a subagent, use /agents and switch to the subagent thread
7. Switch back to main agent

When you switch back, you should immediately be presented with literally the same questions. In fact, if it asked you 15 questions in the planning phase, it'll ask all 15 again, consecutively, 3 at a time.

While I have your attention PLEASE consider the following 2 changes:

1. Add an option to Plan mode to Save plan to file. It's better to manage state of a long horizon task in a md file. I dont want to have to ESC out, and ask it to save. It would be great if it was just an option.
2. When the orchestrator closes a spawned subagent, it is no longer accessible via /agents. This is a massive problem for auditing and logging. How am I supposed to triage what went wrong if I can no longer access the thread? Please change it back. I should be able to see EVERY single subagent that was spawned in a session, whether its still open or not. Please :)

What is the expected behavior?

_No response_

Additional information

You all rock.