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
Codex Cloud threads API: reading a local Codex thread fails with "unsupported placement format version 2"; cloud_threads_create returns invalid argument on a connected & authorized environment #50780
AppServerBackendRequestError: Client specified an invalid argument: post-cutover thread ID 01a0ffa8-c433-79a3-8a11-a1a771cc8903 has unsupported placement format version 2
error_code: UNKNOWN
No request ID was returned.
Last reproduced: 2026-10-04 03:17:06–03:17:14 UTC
The thread returns no content. Some other ChatGPT conversations can be read, so this is not a global read failure.
environmentId matches the corresponding list_environments output exactly.
No model, thinking, cwd, environmentConfigId, or fork_turns were passed.
Full error:
AppServerBackendRequestError: Something went wrong. Request ID: 2ba63be3-ce8e-4344-a42d-d8b51b27ad72
error_code: UNKNOWN
Time: 2026-10-03 06:59:53–07:00:17 UTC
A subsequent mcp__codex_apps__cloud_threads_list_threads with {"limit":10} returned threads=[], nextCursor=null. No new task ID and no DOT_CREATE_OK reply.
A previous create failure returned request ID: wfr_01a0ffbadfee79c58ca129c43a355f2a (2026-10-03 03:07 UTC).
Impact
Cannot set up the intended workflow (Codex delivers → dot hands off to a GPT command session → instructions return to Codex). Only a limited number of minimal tests were performed; no project files, task database, or security permissions were modified. Whether the read failure and the create failure share the same root cause is currently unknown.
Requests
Is reading / following local Codex threads like this supported by the cloud threads API?
Is unsupported placement format version 2 a known thread routing / format compatibility issue? Is the fix server-side, or does it require a specific client version?
Please look up the create-failure request IDs above and explain why a minimal task cannot be created even though the machine is online and authorized under the default model configuration.
Are there any verified workarounds, required versions, or an ETA for a fix?
Related public reproductions (not confirmed to share the same root cause)
Adding a macOS reproduction observed on 2026-10-05.
The Mac is reported as connected, attached, and authorized for tasks. Local task delegation worked on October 4, but the following operations fail on October 5:
Creating a local task with an explicit computer selector: AppServerBackendRequestError, error_code UNKNOWN. Request ID: ee1fcc8e-6bc0-427f-98f7-397c9393f153
Creating a local task using the default attached computer, without an explicit selector: the same error class. Request ID: d5bfd533-c6b4-41f1-8bf2-ba3ae5867650
Sending a follow-up to a previously working local child task also fails. Request ID: wfr_01a10c1e836d7d92b6c858b820289781
Each failed create was followed by a task-list check; no new task was found. Account-level task listing and reading an existing dot-created child task still work, as does cloud-browser work.
Separately, reading a known existing local Codex thread returns:
Client specified an invalid argument: post-cutover thread ID [REDACTED] has unsupported placement format version 1
That read error resembles the macOS report in #50077. Whether it shares a cause with the create/follow-up failures is unknown. This report does not establish a service-wide outage or a root cause.
The current client version/build has not been verified because local execution cannot start.
Could maintainers check the request IDs to identify where task admission fails, including whether the failure is on the client or service side, and advise on a supported recovery procedure or fixed client version?
Activity
github-actions commentedon Oct 4, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
taotao-520 commentedon Oct 5, 2026
Adding a macOS reproduction observed on 2026-10-05.
The Mac is reported as connected, attached, and authorized for tasks. Local task delegation worked on October 4, but the following operations fail on October 5:
Each failed create was followed by a task-list check; no new task was found. Account-level task listing and reading an existing dot-created child task still work, as does cloud-browser work.
Separately, reading a known existing local Codex thread returns:
Client specified an invalid argument: post-cutover thread ID [REDACTED] has unsupported placement format version 1
That read error resembles the macOS report in #50077. Whether it shares a cause with the create/follow-up failures is unknown. This report does not establish a service-wide outage or a root cause.
The current client version/build has not been verified because local execution cannot start.
Could maintainers check the request IDs to identify where task admission fails, including whether the failure is on the client or service side, and advise on a supported recovery procedure or fixed client version?