You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Allows long builds or server runs to not block and allows the agent to inspect logs of the long running process or once it completes without blocking other tasks
This is much needed just ran into this when I asked it to start my API, just sat there watching the time in `Working 10s ..." go up without being able to do anything else 😅.
I'd love something like this so I could do something like
"ok do a test install & watch the log", then go & do stuff & have my interactive codex session see & respond to exceptions in the log. The mechanical "see error->copy-error/point-agent-at-error->ask for fix" loop gets monotonous & motivation-sucking.
I can easily setup an exec loop—but then I lose the interaction & ability to steer.
I built roughly this as an agent-callable tool and have been running it. It is a feature-gated monitor tool with three actions: start (give it a shell command to run as a background watcher, like a dev server or a long build), list (see what is currently being watched), and stop. When the watched command prints output, that line wakes the session, so the agent can launch a server, keep working, and react the moment the log shows it came up or crashed, without sitting and blocking on it. It is live in my own Codex (CLI and Desktop) with tests behind the flag. Reference branch: rust-v0.142.0...yaanfpv:codex:monitor-tool, full design in #29922, and #22003 is the broader hub for this capability. Glad to turn it into a PR if there is appetite.
Activity
Frulfump commentedon Aug 9, 2025
This is much needed just ran into this when I asked it to start my API, just sat there watching the time in `Working 10s ..." go up without being able to do anything else 😅.
aehlke commentedon Aug 10, 2025
More context on use cases: https://x.com/steipete/status/1954594876743389444
etraut-openai commentedon Nov 27, 2025
Closing in favor of #3663
etraut-openai commentedon Dec 18, 2025
I previously closed this as a duplicate, but I'm going to reopen because this has received more upvotes than the other one.
lastobelus commentedon Mar 27, 2026
I'd love something like this so I could do something like
"ok do a test install & watch the log", then go & do stuff & have my interactive codex session see & respond to exceptions in the log. The mechanical "see error->copy-error/point-agent-at-error->ask for fix" loop gets monotonous & motivation-sucking.
I can easily setup an exec loop—but then I lose the interaction & ability to steer.
mrlzla commentedon Jun 19, 2026
Still no monitor? Token usage during long running tasks constant validation is severe.
yaanfpv commentedon Jun 24, 2026
I built roughly this as an agent-callable tool and have been running it. It is a feature-gated
monitortool with three actions: start (give it a shell command to run as a background watcher, like a dev server or a long build), list (see what is currently being watched), and stop. When the watched command prints output, that line wakes the session, so the agent can launch a server, keep working, and react the moment the log shows it came up or crashed, without sitting and blocking on it. It is live in my own Codex (CLI and Desktop) with tests behind the flag. Reference branch: rust-v0.142.0...yaanfpv:codex:monitor-tool, full design in #29922, and #22003 is the broader hub for this capability. Glad to turn it into a PR if there is appetite.jtsang4 commentedon Aug 11, 2026
This feature is very useful. Codex is a year behind Claude Code and still hasn't supported it. I hope it can be supported.
s-banach commentedon Aug 14, 2026
This is one of the most basic features you'd build into version 0.1.0 of a harness, I can't understand how this issue is still open...