Generate images through Codex

Codex generates images with its built-in tool under the user's ChatGPT login; no OPENAI_API_KEY is needed. This skill lets an agent on any model use that route: it hands Codex a brief, Codex shapes the prompt with its own imagegen skill, and image production is proven by a verified PNG on disk, never just Codex's reply.

The onboarding declaration is the machine-readable onboarding contract.

Brief

Codex sees nothing of the conversation, so the brief carries everything the image depends on: what to depict and what it is for, exact text to render (quoted), aspect ratio, the role of each attached image by its order (Image 1: edit target, Image 2: style reference), what must stay unchanged in an edit, a transparent background when needed, and any style or constraints the user set. Leave prompt wording to Codex.

When the user supplies a finished prompt, pass it with --verbatim so Codex sends it to the image tool unchanged.

Generate

python3 <skill-directory>/scripts/codex_image.py --brief-file <brief.txt> --out <absolute/path.png> [--image <reference.png>] [--root <dir>] [--verbatim]