I would like the integrated terminal in a local Codex task to optionally show the commands and output produced by Codex itself.
Current behavior
While Codex is working, I can see its status and resulting file changes. However, opening the integrated terminal gives me a separate manual shell. It does not show the builds, tests, or diagnostic commands that Codex is currently running.
This makes it difficult to tell whether a long-running command is progressing, waiting, or repeatedly failing.
Requested behavior
Each local task could provide an Open agent terminal action showing:
-
the current command;
-
live stdout and stderr;
-
running, completed, failed, or interrupted status;
-
the exit code;
-
recent command sessions belonging to the task.
A read-only view would already be sufficient. Interactive input could remain optional and continue to use the existing permission and approval system.
The current manual terminal should remain available as a separate user-controlled shell.
Relationship to an existing request
A related discussion titled “How do I show the live Codex task-progress view in the unified ChatGPT desktop app?” asks for the general live progress view.
This request is narrower: it specifically asks for the real command session of the current local task to be visible through its integrated terminal.
Why this matters
This would make long-running Codex tasks easier to supervise, troubleshoot, and learn from. Users could see the commands already being executed on their own computer without requiring access to internal reasoning or additional permissions.
The goal is simply to connect the task’s visible terminal with the command activity already taking place for that task.