Built-in image_gen returns opaque ServerError without request ID or diagnostics

Codex version

• Codex Desktop: 26.602.40724 (build 3593)
• macOS: 26.1 (25B78)
• Architecture: Apple Silicon / arm64
• Plan: Plus

Description

The built-in image_gen tool failed twice consecutively while editing an existing PNG design mockup.

Both calls returned only:

ServerError
Encountered exception: <class 'oai_errors.errors.ServerError'>

No HTTP status, error code, request ID, retry guidance, or underlying cause was provided.

An earlier image_gen request in the same Codex session completed successfully, so the tool was available and functional before these failures.

Steps to reproduce

1. Open Codex Desktop.
2. Load an existing PNG into the conversation.
3. Ask Codex to edit the image while preserving most of the original layout.
4. Invoke the built-in image_gen tool with detailed image-editing instructions.
5. Retry the same operation once after the first failure.

Expected behavior

Either:

• The image edit succeeds, or
• Codex returns an actionable error containing:
• error category/code;
• request ID;
• whether the failure is retryable;
• whether it was caused by rate limiting, input validation, safety checks, or an internal service error.

Actual behavior

Both attempts immediately returned:

ServerError
Encountered exception: <class 'oai_errors.errors.ServerError'>

No image was generated and no additional diagnostic information was exposed.

Additional context

• A simpler image-generation request and an earlier complex UI mockup generation succeeded in the same session.
• No 429 or explicit rate-limit error was returned.
• The session telemetry did not indicate rate_limit_reached_type.
• OpenAI Status showed the service as operational when checked.
• The failure occurred during an edit/refinement request for an existing generated PNG.
• Retrying once produced the identical opaque error.

Why this matters

Without a request ID or error classification, users cannot determine whether they should:

• retry later;
• simplify the prompt;
• restart Codex;
• wait for a rate-limit reset;
• report a backend incident.

Please expose structured diagnostic information for built-in image_gen failures, even when the underlying error is an internal server error.