Skip to content

ChatGPT–Codex coordination is asymmetric: native send fails after successful discovery/read #62860

Open

Description

@brando90

What version of the Codex App are you using?

ChatGPT desktop app's Codex surface on macOS: 26.1111.71704, build 29030, bundle identifier com.openai.codex (read from the installed application's Info.plist). The separately installed CLI reports codex-cli 0.171.0; this report concerns the desktop/chat integration, not a reproduced CLI defect.

What subscription do you have?

The exact subscription tier for both participating surfaces was not independently verified in this reproduction.

What platform is your computer?

macOS 27.2.1, Apple silicon (arm64). uname -mprs: Darwin 27.2.0 arm64 arm.

What issue are you seeing?

I explicitly asked my Codex chat and an existing ChatGPT web chat to communicate and agree on which agent would own a shared document-production task. The integration was asymmetric and the native send failed:

  • From Codex, list_threads with source: "chatgpt" found the correct existing ChatGPT conversation, and read_thread returned its actual messages.
  • send_message_to_thread targeting that exact discovered conversation failed on two attempts with: Steering requires an active compatible turn.
  • The recipient was surfaced as active during the inspection, but the error did not explain what made it incompatible or offer a supported queue/delivery path.
  • In the reverse direction, the ChatGPT conversation reported that it had no direct tool for contacting the Codex chat and that its attempted local session-queue fallback could not open the Codex session database. That reverse-direction detail is the other agent's report, not a separately instrumented backend diagnosis.
  • A background browser message to the exact ChatGPT conversation did succeed. The user-authorized coordination therefore required UI automation instead of the native messaging interface.

This is frustrating: I should not have to act as a copy/paste message bus between my own OpenAI agents, or have one agent drive a browser merely to send a message that I explicitly authorized.

What steps can reproduce the bug?

Observed on 11-20-2026 at approximately 3:20–3:27 p.m. America/Los_Angeles (PST).

  1. Have an existing Codex desktop chat and an existing ChatGPT web chat, both accessible to the user.
  2. Explicitly authorize the chats to coordinate about a shared task.
  3. From Codex, list ChatGPT history and read the discovered target conversation.
  4. Call send_message_to_thread with the exact discovered threadId and an ordinary coordination prompt, without model or reasoning overrides.
  5. Observe Steering requires an active compatible turn; retry once after re-reading target state and observe the same error.
  6. Compare with sending the coordination message through that conversation's normal web composer, which succeeded in this case.

Private conversation IDs, document links, email addresses, and screenshots containing unrelated chat titles are intentionally omitted from this public report. I can supply a minimal private diagnostic record through OpenAI Support. No account/workspace identity equivalence is assumed merely because both chats are user-owned.

What is the expected behavior?

Please make user-authorized ChatGPT ↔ Codex coordination work natively in both directions:

  • Discover and read selected user-owned conversations when account/workspace permissions allow it.
  • Send a message to an existing chat without requiring the user to relay it manually.
  • Queue a message or clearly explain whether an active, idle, completed, or incompatible target supports delivery.
  • Return a delivery identifier/status and make the recipient's response accessible to the sender.
  • Expose supported capabilities consistently, or give an actionable explanation of host/account/workspace restrictions.

This is a request for an explicit, permission-controlled bridge, not unrestricted access to other users' chats or a bypass of workspace boundaries. If native send intentionally does not support this target, its tool contract should make that limitation clear before the attempted send.

Additional information

The closest reports I found are #59085 (same error in a ChatGPT composer path), #40852 (send tooling availability), and #60808 (history visibility after a successful send). This report concerns a callable native Codex-to-ChatGPT send that failed after successful discovery/read, plus the missing reciprocal coordination path.

Please route this to the ChatGPT/Codex integration team and track the bidirectional coordination request alongside the send failure. Prepared and submitted with Codex at my explicit request.

Activity

  1. added
    bugSomething isn't working
    appIssues related to the Codex desktop app
    app-serverIssues involving app server protocol or interfaces
    tool-callsIssues related to tool calling
    on Nov 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    appIssues related to the Codex desktop appapp-serverIssues involving app server protocol or interfacesbugSomething isn't workingtool-callsIssues related to tool calling

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions