IMPORTANT multiagents issue fail after a while.
What version of Codex CLI is running?
codex-cli 0.114.0
What subscription do you have?
Pro
Which model were you using?
gpt-5.4 high
What platform is your computer?
Linux 6.8.0-101-generic x86_64 x86_64
What terminal emulator and version are you using (if applicable)?
Tabby
What issue are you seeing?
Through this chat '019cdfa0-7d65-7590-90da-7808c4859fe2', we can see that after a while, even if they are automatically closed by the main agent, the multi-agents are no longer available. They will always fail when attempts are made to spawn them in the environment, and I am forced to restart the task from scratch, losing important tokens and context windows that I will need (since I now have to restart the analysis to make them understand what needs to be modified).
Furthermore, I have a maximum of 12 as max_threads in the config.toml file. Initially, I had 10, but I started to notice this “bug” and decided to increase it. As max_depth, I have 2.
I also think that there is no API that the agent can use to view all the multi-agents present.
What steps can reproduce the bug?
Uploaded thread: 019cdfa0-7d65-7590-90da-7808c4859fe2
User-Id: user-grEQqq7dokHb56wxlYbBeUWO
What is the expected behavior?
I expect that multi-agents can always be spawned, especially if they are closed.
Additional information
_No response_
4 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
<img width="1861" height="72" alt="Image" src="https://github.com/user-attachments/assets/27ca9f80-d555-41e2-aba9-fd12a46ba745" />
losing a lot of usage for this bug.. and im doing bad updates in my codebase..
I've taken to reminding it to close out any old agents before continuing, it eventually remembers to do this but when it doesn't I have to exit codex and resume the session to get them back. This has gotten slightly better over time but does still happen
The main agent also has a bad habit of rushing agents to completion or closing them out when it is hung on permissions or long running tests/tool usages. Asking it not to do this helps but not always. This has been the primary cause of sub-agents making messes for me
This is a known issue of depth > 1
Will work on a fix