Optimizes image reading by caching LLM-generated content as XMP metadata inside image files. Use BEFORE reading any image file (.png, .jpg, .jpeg, .webp, .gif, .bmp). Reduces token usage by ~90% on repeated image reads. Triggers on: image, screenshot, picture, photo, .png, .jpg, .jpeg, read image, view image, look at image.
2026-03-31