Frequently asked questions
Native messaging is the pipe. Coordinator gives the agents one goal, clear lanes, and rules for when another task actually needs to hear something.
Yes. Codex provides native inter-task messaging. Coordinator uses it sparingly for bounded assignments, results, dependencies, collision notices, and other handoffs where another task actually needs to act. You stop being the person copying those messages between windows.
Messaging alone does not decide who owns what, whether two agents are doing the same work, or how several tasks get back to one result. Coordinator gives them one shared goal, clear ownership boundaries, and rules for when to communicate.
Use it when one project genuinely benefits from multiple Codex tasks working at the same time and you would otherwise relay updates or untangle overlapping work yourself. For one job or a small edit, stay with one Codex task.
No. It starts with one task, reuses a useful related task when possible, and opens another only when a complete job truly benefits from its own task.
It starts with one task and allows up to five active tasks by default, including the task that owns the goal. If more would genuinely help, it asks before opening them. You can approve a larger count for that goal or change the project’s normal ceiling.
Yes. Coordinator uses native task visibility plus its small local ownership records to show each participating task’s job, planned files, dependencies, and current status when you ask.
No. That would just replace human noise with machine noise. Routine progress stays with the task. Cross-task messages are reserved for a real assignment, result, dependency, collision, release, or other handoff where a specific recipient needs to act.
Yes. Mission Control is an optional read-only local view of one enabled project’s active tasks and conflicts. It starts only when you ask, refreshes only when you press the button, and has no task controls.
No. On supported non-Ultra intelligence levels, you can request subagents directly or ask for them through project and skill instructions. Ultra adds proactive delegation, so you do not have to remember to ask each time. Read OpenAI’s subagent guide.
No. For everyday solo development, separate tasks can work in one checkout with clear jobs and visible file areas. Keep using worktrees when you specifically need hard isolation.
It makes planned paths visible as warnings. Tasks can re-read shared files and pause only an actually conflicting change or exclusive action, instead of blocking a whole task because two paths overlap.
No. It does not poll every task or run a permanent watcher. Each assigned task returns when it finishes. If you request unattended supervision, Coordinator can use one temporary Codex follow-up until the goal reaches its next decision.
No. Task conversations stay in Codex. Coordinator keeps only the small amount of local information needed to show what each task is doing.
No. Your tasks can keep using the same project, local tools, branches, commits, and review process.
Its job remains visible, so you can return to it or hand the work to another task without copying the whole conversation.
Run the included read-only check. If the installed package is broken or outdated, update or reinstall it. The check does not scan projects, watch tasks, or repair files.
OpenAI Codex with plugin support, Git, and Python 3.10 or newer. Codex Coordinator itself is free and open source.
The developer page covers how it stores project information, how inter-task communication is bounded, how it handles Git, its limits, tests, and design history.