name: photo-grading description: Deterministic photo color grading / photo color correction execution Skill for AI photo agents. Use it when a caller has decided which color treatment to apply to a photo and needs it executed safely: the 3-way Color Grading wheels (Shadows / Midtones / Highlights + Global: hue, saturation, luminance per wheel; blending; balance), typed photo corrections (white balance temperature/tint, exposure, contrast, highlights/shadows/blacks/whites, saturation, sharpness), or 3D LUT application (PHOTO_LUT_APPLY). Executed as typed, deterministic operations through FFmpeg's photo filters (lut3d, eq, hue, unsharp), with output validation and provenance. Do NOT use it to decide which color treatment or look to apply, to judge a photo "cinematic", or to run arbitrary ffmpeg filters. WHITE_BALANCE is an alias of PHOTO_CORRECTION restricted to temperature/tint.
photo-grading
Machine interface: photo-grading run - --json with a photo-grading/request@1 document on
stdin; exactly one photo-grading/response@1 document on stdout. photo-grading skill --json
prints the contract; doctor --json the environment; plan - --json a dry run;
validate - --json a schema check.
Rules for a calling agent:
- Decide first: which operation, which look (the caller already did this). This skill has no defaults for creative parameters: wheels/parameters come from the request.
- Give explicit, typed parameters; each is independently optional, range-checked, and
"unchanged" when omitted.
PHOTO_GRADING:shadows_hue/shadows_saturation/shadows_luminance,midtones_*,highlights_*,global_*(hue 0-360, saturation 0-100, luminance -100..100),blending(0-100, default 50),balance(-100..100, default 0).PHOTO_CORRECTION:white_balance_temperature(2500-10000, default 6500),white_balance_tint(-100..100),exposure(-10..10),contrast(-100..100),highlights/shadows/blacks/whites(-100..100),saturation(-100..100),sharpness(0..100).PHOTO_LUT_APPLY:lut_path(required). - Never send commands, argv, filter strings or executable paths: the request is rejected.
- Keep outputs inside the workspace, never at the input path; set
overwrite: truedeliberately. This skill never mutates inputs. - Read
results[].status,outputs[].provenanceanderrorfrom the JSON; exit code 0 only whenokis true. - Measurements (signalstats means) are OBSERVED values; report what you saw.
Wheel semantics (PHOTO_GRADING): zones are BT.709 luma bands (shadows/midtones/highlights).
Each wheel with saturation > 0 rotates a pixel's hue by the wheel's hue and scales its
saturation by saturation/100 within its zone. blending mixes the tinted LUT back toward
identity; balance moves the zone boundaries. The classic cinematic look is teal shadows
- orange highlights.
See README.md for the schemas, operation table, error codes and provenance.