Codex should distinguish possible false-positive content handling from network reconnect errors

Codex version

26.727.51351

Platform

macOS 26.5

Problem

In some ordinary conversations, colloquial or ambiguous wording may be treated as sensitive even when the underlying request is benign. The assistant may then stop responding or the response stream may terminate.

The Codex UI can show only a generic message such as:

Reconnecting…
stream disconnected before completion
stream closed before response.completed

This makes it impossible for the user to determine whether the cause is:

• a local network interruption,
• a retryable server or transport failure,
• rate limiting,
• or possible content-safety handling / a false-positive sensitivity decision.

As a result, users may repeatedly troubleshoot their network when the problem is unrelated, or may incorrectly assume that a normal network failure was caused by their wording.

Expected behavior

Codex should clearly distinguish between:

1. Local network problem
2. Retryable server/transport failure
3. Rate limiting
4. Content-safety handling or refusal
5. Unknown error

If a response is stopped for content-safety reasons, the UI should provide a clear, neutral explanation instead of presenting it as a network reconnection. A non-retryable content decision should not trigger repeated reconnect attempts.

The client should also provide a privacy-preserving diagnostic code and a private feedback mechanism that does not require users to publish their conversation content.

Privacy note

This report intentionally contains no screenshot, conversation excerpt, prompt text, logs, request ID, account information, subscription information, device identifier, or hardware serial number.