Codex enters infinite processing loop when GitHub repository is deleted (404 not handled as terminal state)

[alexshescp]

What issue are you seeing?

When Codex is asked to expand or analyze a GitHub repository that has been deleted, the request does not fail gracefully. Instead, Codex enters an infinite or unbounded processing loop and remains in a "processing" state for an extended period of time (observed >1 hour).

This behavior is reproducible and appears to affect both Codex Web (via ChatGPT UI) and Codex CLI, suggesting a shared backend issue rather than a UI-specific problem.

What steps can reproduce the bug?

1. Create a new repository on GitHub.
2. Connect the repository to Codex (e.g., via ChatGPT Codex Web UI or Codex CLI).
3. Delete the repository from GitHub.
4. Submit a Codex request to expand or analyze code from the deleted repository using its original GitHub URL.
5. Observe the processing behavior.

What is the expected behavior?

• The request does not terminate.
• Codex remains in a processing state indefinitely.
• No explicit error is returned to the user.
• The task continues running for more than 1 hour.

Additional information

Expected Behavior

• GitHub 404 / 410 responses for deleted repositories should be treated as terminal errors.
• The task should fail fast with a clear and actionable error message.
• Retry attempts should be bounded and respect a global execution timeout.
• The user should be able to recover without restarting the session.

Impact

• Poor user experience due to lack of feedback and stuck UI state.
• Unnecessary frontend and backend resource consumption.
• Potential infrastructure load caused by unbounded retry behavior.

Environment

• Codex Web (ChatGPT-hosted, browser-based)
• Codex CLI
• GitHub repository URL input

Additional Notes

• The issue reproduces consistently across different Codex interfaces.
• This suggests missing handling of terminal states for permanently unavailable repositories.

Suggested Fix (Optional)

• Treat GitHub 404 and 410 responses as non-retriable errors.
• Add an upper bound on retry attempts with exponential backoff.
• Enforce a hard timeout for repository resolution and analysis stages.