Repository navigation
Commands fail for ancient versions of codex #393
Description
Activity
I faced the same issue. For me for some reason homebrew was not updating to the latest version. I had to reinstall use npm to get the latest version.
npm i -g @openai/codexReacted by Ala BEN SLAMAReacted by Ala BEN SLAMAitsjeremyjohnson commented
on Apr 17, 2026 More actionsI can confirm I am still hitting the same user-facing error string, but not in the original “ancient Codex CLI” setup from this issue.
On my machine this is happening with a much newer stack:
- T3 Code (Alpha)
0.0.20 - Codex CLI
0.121.0 - macOS
26.3(25D125) - Codex auth state:
authenticated - Account type: ChatGPT
Pro
Relevant local evidence:
- T3’s cached provider health shows Codex as installed, authenticated, and ready:
version: 0.121.0status: readyauth.type: pro
- Current startup logs also show
codex model/listandcodex account/readsucceeding normally. - So at least in my case, this does not look limited to “Codex is too old and cannot speak the current app-server protocol.”
What makes this feel related to the broader stuck-thread lifecycle bugs:
- After these failures, T3’s local
state.sqlitecontains orphaned pending turns withturn_id = NULL, for example:- thread
fe471aed-9aa0-403b-9022-2859be284ac2requested at2026-03-22T03:54:58.615Z - thread
2f883475-7954-43d5-b596-602e232e276brequested at2026-03-07T18:41:15.471Z - thread
08094728-79ae-4624-a91b-3d6a9b235b62requested at2026-03-07T15:58:17.095Z - thread
75b2ae66-2360-48db-b407-edf34f2e512drequested at2026-03-07T02:36:10.760Z
- thread
- Those rows look similar to the orphaned pending-turn / stuck-session shape discussed in:
#313#1048
So my read is:
#393is a real issue, but the title/root cause is probably too narrow now.- The same banner (
Received invalid JSON from codex app-server.) can still show up on newer Codex versions. - In at least some cases, the downstream effect appears to be bad persisted turn/session state rather than just a one-time startup incompatibility.
I did not find the exact invalid-JSON line preserved in the current summary logs (
server.log/server-child.log), which is also part of the problem: the visible banner survives in the UI longer than the actionable raw transport detail survives in logs.It would help a lot if T3 logged the offending raw app-server line / parse failure cause alongside the thread id and provider session id whenever this banner is shown.
If useful, I can open a separate issue specifically for the newer-version + orphaned-pending-turn variant and link it back here.
Reacted by Sasha Duke- T3 Code (Alpha)
Yup can repro this error msg consistently on latest versions of everything, it's the most common family of issues ive been encountering in various places
This report is tied to Codex v0.36 and the provider/update path has changed substantially. Please confirm whether it still reproduces with a currently supported Codex CLI and current T3 build, including both versions and the exact command/error.\n\n
- addedneeds more infoInitial triage showed no bug. Awaiting more infoInitial triage showed no bug. Awaiting more info
on Jul 20, 2026 Note
🤖 GPT-5.6 Sol in Codex responding on behalf of Theo
Closing because the report does not have the current details needed to investigate. The report targets an unsupported Codex release. A July 20 request for a current reproduction received no reply.
A new issue can be opened with the current application version, reproduction steps, and relevant logs if the problem still occurs.
- added 8 commits that reference this issue
on Aug 26, 2026 8 remaining items
- added 15 commits that reference this issue
on Aug 26, 2026
I got this error when I tried to run a command with codex
v0.36. I recommend first checking the current version of codex and showing an upgrade pane if it's below a minimum version.application logs