Codex Image

Use this skill when the user asks for a technical infographic, explainer image, visual one-pager, codebase map, architecture visual, concept visual, or content pack from a URL, pasted brief, local file, or local repo.

Canonical Runtime Contract

Canonical runtime guidance lives in this file because local installation copies .agents/skills/codex-image into $HOME/.agents/skills/codex-image. Repo-root templates are development references that mirror this contract, but installed users must not need repo-root files for correct behavior.

This skill assumes the active Codex environment has Codex built-in gpt-image-2 image generation. If image generation is unavailable and the user did not ask for prompt-only mode, stop and say: codex-image requires Codex built-in gpt-image-2 image generation in the active session.

Do not call the OpenAI API directly. Do not ask for API keys. Do not automate browser login or account state.

Modes

Default mode is one-shot. Research the source, choose a lane, write the brief, write the prompt, generate one image, inspect it, and save artifacts without asking questions unless the input is missing, inaccessible, unsafe, or impossible to interpret.

Art-director mode is selected when the user says art-director, art director, ask me first, or asks to choose style before generation. Ask only these questions, then proceed: