img-gen

name	img-gen
description	Generate raster images with the OpenAI Image API through Zditor Native Agent's http_request tool. Use when the user asks to create or generate an image, illustration, diagram, cover, poster, concept art, or other visual asset, including requests that call this skill img_gen or img-gen.

Image Generation

Generate images according to the user's requirements. Preserve the requested subject, style, text, language, aspect ratio, colors, and composition. Do not add an unrequested visual style or invent important content.

Setup

Install this directory into the Zditor App's global skills directory so Zditor Native Agent can discover it. When installing from this repository, use scripts/install-skills.sh --target zditor. Before first use, create .env beside SKILL.md from the included template:

cp .env.example .env

Open .env and set API_KEY to an OpenAI API key. Keep IMAGE_API_URL unchanged unless the official service endpoint changes. Never commit .env or paste the API key into chat.

Generate

Call http_request once per image. Replace the prompt and change model options only when the user requests different values. Copy the exact final prompt and model from body.value into image_metadata; Zditor uses this metadata to display and copy the prompt with the generated image.

{
  "method": "POST",
  "url_skill_env": {
    "skill": "img-gen",
    "key": "IMAGE_API_URL"
  },
  "header_skill_env": {
    "Authorization": {
      "skill": "img-gen",
      "key": "API_KEY",
      "prefix": "Bearer "
    }
  },
  "body": {
    "type": "json",
    "value": {
      "model": "gpt-image-2",
      "prompt": "<user requirements>",
      "n": 1,
      "size": "3840x2160",
      "quality": "high"
    }
  },
  "save_to_session": "auto",
  "base64_json_path": "/data/0/b64_json",
  "image_metadata": {
    "prompt": "<same final prompt as body.value.prompt>",
    "model": "gpt-image-2"
  },
  "timeout_ms": 300000,
  "max_response_bytes": 67108864,
  "display_saved_image": true
}

Confirm success only when the tool returns ok: true. Inspect the displayed image for obvious failures, but do not claim its exact pixel dimensions were verified. For multiple images, repeat the request with n: 1 so every result is saved and displayed separately.

The configured endpoint uses official OpenAI HTTPS. Never read, print, or place API_KEY directly in tool arguments or chat. If it is not configured, ask the user to complete the setup above.

Store Images

• Keep images in the current session with save_to_session: "auto" by default. Zditor removes session images when the session is deleted.
• Use the returned saved_to value to identify the session image. Do not expose or hardcode its app-data absolute path.
• When an image must be referenced by a workspace document, call materialize_session_image with saved_to as source and the workspace-relative document path as document_path. Use its returned documentReference in the document.
• Use save_to only when the user explicitly requests an immediate workspace output path.
• Session images are limited to 20 MiB. If an image exceeds the limit, retry with a smaller size or lower quality.

Do not retry authentication or invalid-request errors. Retry 429 or 5xx responses at most twice, and do not automatically retry a timeout because the server may already have generated the image.