GPT Image Skill

Generate or edit one or more raster images, save results under the current workspace, and show them in chat.

Keep the subscription boundary

• Never call the OpenAI Images API or fall back to a separately billed route.
• Never use, request, inspect, print, or forward OPENAI_API_KEY.
• Never use codex login --with-api-key or read ~/.codex/auth.json.
• Require explicit ChatGPT auth before the bridge generates an image.
• Stop when API-key auth is detected. Replace authentication only after a separate explicit user choice.

Choose the route

1. Use a host-native OpenAI/Codex image_gen tool directly only when it is backed by the user's included ChatGPT/Codex usage. Do not install Node.js or start nested Codex in that case.
2. Otherwise use scripts/gpt_image.mjs. Claude Code, Google Antigravity, and generic local agents need Node.js 22+, Codex CLI, and Sign in with ChatGPT in the same OS environment.
3. In Antigravity, do not substitute its built-in generate_image tool. That is a different provider route and does not satisfy this skill's GPT Image through ChatGPT-subscription contract. Run the Codex CLI bridge instead.

Read image-workflows.md for references, conversation attachments, edits, follow-up revisions, or multiple images. Read subscription-runtime.md for authentication or architecture. Read platform-setup.md only when setup is missing or uncertain.