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
{{ message }}
Repository navigation
Commit 03f9cfd
Browse filesBrowse the repository at this point in the historyBrowse files
fix(server): Codex runs fail and recover instead of hanging when the app-server dies or stops answering
Every Codex thread shares one app-server process. When it stopped answering
or exited mid-turn, T3 never noticed: running turns stayed "running" forever,
delegated tasks never woke their parent, and new Codex threads hung at
"starting" until T3 restarted.
The adapter now fails its session when the app-server connection ends, and
when the app-server does not answer a once-a-minute thread/loaded/list probe
within a minute. The session manager then fails the session's runs and the
next turn starts a fresh app-server. Events Codex sent before the failure are
kept.
The protocol also ends the connection 2 s after the process exits, since a
command Codex started can keep its stdout open, and fails requests still
waiting for a reply when the client closes. A release no longer overwrites the
status of a replacement session that opened under the same id meanwhile.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
0 commit comments