OpenCode plugin that lets text-only models work with user-provided and tool-returned images by sending each image to a vision-capable model first, then replacing the image with a text description.