Clarify $imagegen quality controls and built-in vs CLI behavior

What version of Codex CLI is running?

codex-cli 0.147.0

What subscription do you have?

Pro

Which model were you using?

Not applicable; this concerns the $imagegen skill and its image-generation backend.

What platform is your computer?

macOS v26

What issue are you seeing?

While using the $imagegen skill, it is unclear whether the built-in image_gen path uses gpt-image-2, whether a user can control output quality, and which parameter name applies. The skill mentions gpt-image-2 and quality in its CLI fallback documentation, but it does not clearly state that those controls are unavailable in the built-in tool. A user asking about outputQuality cannot tell whether it is supported, ignored, or should be translated to quality.

What steps can reproduce the bug?

See https://learn.chatgpt.com/docs/image-generation?surface=cli

1. Invoke $imagegen through the default built-in path.
2. Ask whether outputQuality (or quality) can be set for the presumed gpt-image-2 backend.
3. Read the skill's built-in and CLI-fallback sections. They distinguish the modes, but leave the built-in model identity and parameter forwarding ambiguous.

Feedback thread ID: 01a00184-e2ba-7630-8a99-2553d9a1b2a1

What is the expected behavior?

The documentation should explicitly state: (a) whether the built-in tool is backed by gpt-image-2 or is implementation-dependent; (b) whether built-in calls expose any quality control; (c) that the documented API/CLI field is quality, including allowed values, if applicable, and whether outputQuality is unsupported or an alias; and (d) that selecting gpt-image-2 and setting --quality requires the explicit CLI/API fallback, if that remains the intended workflow. A minimal example and a clear distinction between prompt wording such as "high quality" and an API quality parameter would remove the ambiguity.

Additional information

This is a documentation-clarity request rather than a runtime error. The current docs mention fallback-only execution controls, but the boundary is easy to miss. Feedback thread ID: 01a00184-e2ba-7630-8a99-2553d9a1b2a1.