SorryCode Image2

Use this skill when the user wants to create image assets through the unified SorryCode Images API.

Default Path

Clarify whether the user wants to generate a new image or edit an existing one.
Check whether SORRYCODE_API_KEY exists before writing or running any request.
If the key is missing, stop and help the user set it up. Do not invent a key and do not continue with a fake request.
Use gpt-image-2 by default. If the user asks for Gemini, Nano Banana, or a Gemini image model, use gemini-3-pro-image-preview unless they specify another model.
Use 1024x1024 by default. If the user asks for aspect ratio, high resolution, 2K / 4K, or size, load references/size-guide.md before choosing the parameter.
Use the bundled Node script scripts/sorrycode-image2.mjs for actual requests. Do not hand-write inline JSON for shell one-offs.
For first-run Gemini image checks, use --no-stream so the request is plain OpenAI-compatible JSON with response_format: b64_json.
For slow OpenAI image requests or large images, use the default streaming mode with stream: true and partial_images: 2.
For editing, require a local input image path, then run the script with --mode edit --image <path>.
Save outputs under outputs/images/<short-slug>/ unless the user asks for another folder. The script writes prompt.txt, request.json, headers.txt, response.json, summary.json, and events.ndjson when streaming.