image-editlisted

# image-edit Not every image job is a generation. When the pixels already exist, a photo, an earlier render, a folder the user handed over, the honest, free move is to edit them, not spend credits making new ones. This skill is Loki's local workbench for exactly that: deterministic operations run through `zanmai/system/scripts/image-edit.py`, no model, no backend, no cost. Loki still directs it with judgment (which operation serves the piece, what target size, which look), but the pixels are moved by code, not guessed by a model. Reach for it before the generation gate whenever adapting existing imagery would do, the media skill's own rule ("prefer adapting given or existing imagery to generating fresh") lands here. ## The operations Run `image-edit.py detect` first on an unfamiliar host, it reports which tiers this machine can do. Then: **Core, always available once Pillow is provisioned.** `convert` (format from the output extension, WebP included), `resize` (exact, a scale factor, or fit-within keeping aspect), `rotate`, `crop`, `grayscale`, `composite` (overlay one image on another with position, scale, opacity), `optimize` (re-save smaller). `batch` applies convert / resize / rotate / grayscale / optimize to every image in a folder, the path for "prepare this whole set for the web" or "shrink these for the deck". **Detected, heavier, host-dependent.** `grade --lut <file.cube>` bakes one colour look consistently across a shot or a whole batch; `grade --match <refere