My Codex feature requests
OpenAI Developer Community
July 2, 2026
On the command-center idea, I would keep the manager read-only until a worker reaches a handoff point. Otherwise it becomes a second writer that can change scope while a task is still running.
The state I would want each worker to expose is small: current goal, last verified checkpoint, files or commands touched, blocker, and whether it is waiting for permission. Then the main chat can resume or reassign from that ledger, not from a loose chat summary.
Discussion in the ATmosphere