Codex Imagegen

Codex ImageGen

Overview

Use this skill when the repo contract asks for actual generated or edited image assets and the billing/auth path must be Codex subscription usage. This is not an OpenAI API integration skill.

Auth And Billing Boundary

• Prefer Codex authenticated with ChatGPT/Codex subscription usage.
• Do not create or run scripts that call the OpenAI Images API or Responses API.
• Do not read .env, .env.*, local credential files, API keys, tokens, or

ignored provider profiles to prove access.

• Do not use OPENAI_API_KEY, openai images.generate, responses.create, or

direct HTTP calls for image generation unless the user explicitly says to use API billing instead of subscription usage.

• If the current Codex runtime exposes a native image generation tool, use that

tool directly.

• If the current runtime has no native image generation tool, fail closed:

write a prompt/handoff artifact and state that no image was generated.

Input Contract

Read the active SPEC.md first. Also read only task-relevant public assets, taste.md, taste.vision, visualization packages, or design docs. Extract an asset contract for every requested image:

• asset id and purpose
• target path and format, such as public/hero.webp or assets/logo.png
• dimensions, aspect ratio, background, and transparency needs
• style, brand, visual references, and forbidden elements
• exact text that must appear, or no embedded text
• required variants
• acceptance checks

If SPEC.md asks for images but lacks output paths or acceptance checks, update or request a spec clarification before generating.

Generation Workflow

1. Create an asset manifest in the workflow or task output, listing every image

to generate or edit.

1. Build a precise prompt for each asset. Include product context, subject,

composition, lighting, style, dimensions, text constraints, and disallowed artifacts.

1. Use the native Codex image generation capability when available. Favor high

quality for product-facing assets unless the spec explicitly says draft, preview, or low cost.

1. Save each generated file to the exact path required by SPEC.md. Create

parent directories as needed.

1. For edits, preserve the original source asset unless the spec explicitly

allows overwrite. Write edited variants to new paths.

1. Record the revised prompt or final prompt, generated artifact path, and

generation status.

1. Inspect the output before closeout. At minimum, check file existence, file

type, approximate dimensions, visible subject, text legibility when relevant, and obvious prompt violations.

Fail-Closed Output

When image generation is unavailable, write a handoff file beside the requested asset or under the workflow artifact, for example:

# Image Generation Handoff: hero

- Status: blocked-no-native-codex-image-tool
- Intended output: public/hero.webp
- Billing mode requested: Codex subscription, not OpenAI API
- Prompt: ...
- Acceptance checks: ...

Do not claim the image exists. Do not substitute stock images, SVG placeholders, CSS gradients, or API-generated images unless the spec explicitly accepts that fallback.

Closeout Requirements

Before reporting success:

• every requested generated image has a real artifact path
• each path exists in the workspace or is an attached image artifact that the

runtime can map to the workspace

• prompts and revised prompts are recorded
• any ungenerated image is marked blocked, not done
• no API keys or secret files were read

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: waitdeadai
• Source: waitdeadai/minmaxing
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.