Troy Sobotka (one of the original AgX authors) has commented on Reddit:
Depictions of “caucasian skin” commonly require a reasonable rotation of the R primary toward G.
The default tuning that ended up in Blender and is often repeated elsewhere does not rotate the R nearly far enough.
And, hence, the tuning that ended up in darktable? Interesting… Perhaps the summer release of dt should ship with a preset to compensate for this (if it doesn’t already). 
EDIT: Oops, I see now you’re already on it.
I’m no good at colour and creative edits, so I’d like the community to come up with suggestions.
I would want to see some samples of this need for rotation. I didn’t see an obvious problem comparing skin tone in AgX to Sigmoid and Base Curve. I definitely preferred AgX skin tone over fimlic V5 or V7.
I sometimes apply this color tweak in color zones to warm skin tone closer to the nice out of camera JPG color for my R7. Some images benefit from a double application of this module.
I haven’t much used AgX in a while, due to the need to get through a large backlog. Sigmoid is very fast, sometimes needing no attention at all.
I am usually photographing the many shades of Indian skin under mixed but warmish LED lights. My subjects would have a touch of jaundice and I would have to move both red and green away from yellow. And also, preserve hue at either 50% or 100%, picture dependent.
When I last looked, the latest set of defaults and presets had gone a long way towards fixing this.
(I think I might have another go at AgX now my backlog is a pastlog)
Thanks for the feedback. The Blender-based primaries have remained unchanged (the default is still Blender-based).
Here is my take as a a user who is still learning about color.
AgX can be used for the technical aspects of tone mapping, including hue rotation in the highlights, avoiding the notorious six, as has been extensively discussed elsewhere and which is also mentioned in the documentation.
AgX can also be used for more-artistic rendering of the image, particularly in the “look” section, but also in the reverse rotation section.
Skin tones, at least in my photos, are usually targeted near middle grey. Near that tone level, I would expect that technical color rotation to be low. Of course, there are often some highlights on the skin.
My use of AgX to-date is mostly on the technical tone-mapping of things in my mental model, and I frequently adjust contrast (around 3 to 3.2 if memory serves), and also toe and shoulder in some cases. I usually leave primaries and look sliders alone.
For good skin tone my primary focus is first on white balance. I find this to be easy in natural light, but I often struggle indoors with led-lighting. Sometimes indoors I get one person looking “correct” but another person in the photo looks off.
For split-toning and general color adjustments I prefer to use a combination of rgb primaries and color balance rgb. With the great control of contrast in AgX I’m using color balance rgb less than before, mostly just to tweak saturation a little. And for overall color rotation and purity, using rgb primaries allows me to easily toggle the effect on and off while I decide how well I like the edit.
I am just starting to read about color science from the old days at Kodak, and it has been interesting to learn that Kodak (apparently) tuned color response of their films to fit with personal preferences of people, rather than technical accuracy. I’m on the hunt for good references (free stuff; I know there are books).
I think AgX is a fantastic and amazing module, and I feel very fortunate to have this in darktable as-is. Thank you, István!
From my extensive experience with color film stock from the analog days this is true for every manufacturer. As a generalization Kodak color negative film targeted warm tones, while fuji film targeted blues and greens and tended to have higher contrast making it popular with some landscape photographers.
As a processor the most important thing I had to deal with was the cross curves that films had. For example with fuji film the highlights would come out magenta so an overexposed image needed more green added in the printing. However, an under exposed image would come out too green and so magenta would have to be added. At least this is one problem that doesn’t raise its ugly head in the digital world as much.
Also last night I spent some time experimenting with the primaries adjustments trying to get my RAW file to better match my cameras nicer out of camera JPG color. I found the attenuation settings before tone mapping most helpful for this and set red rotation to zero. I also did some tweaking with after tone mapping settings. I saved this as a preset and will continue to fine tune it over time. I am lucky that my Canon R7 gives a nice color in its JPG which I can use as an aim point. This may not be the case for all cameras.
Cool, I will pay attention when I get back to printing. Photo lab prints for me. 
But my comments of cross curves applies to film only. I am also not sure how current film stock works as that is old technology to me which I have left behind. I have a friend who still shoots film because he likes nostalgia.
Ok, thank you.
But my overall perspective is unchanged; the camera records (approximate) photometric data, but that is not what we generally see as observers. Tone mappers apply a logarithmic curve, and at high luminance there are issues with color, especially mapping to pure white. So advanced tone mappers also modify color in order to compensate. AgX does this very well.
I think there should be technical correction of colors for editing (on monitors), for perceptual “correction” which is not technical, and for printing/archiving. I would prefer editing that is technical first and artistic second, with clear segregation in the editing pipeline.
Or in simple terms:
My 2 cents’ as a general user. 
I think the defaults seem to be fine with caucasian skin tones, but I am not trying to make perfectly accurate photos. I’m trying to make beautiful/meaningful/fun photos.
I often add additional rotation using the RGB primaries module (red towards orange, blue towards cyan, green towards yellow) to accentuate the film-like color shifting.
As long as we’re spending 2 cents on opinions, hands- down the best way to get accurate color reproduction is to measure the colors of concern with a spectrophotometer, put the XYZ values in a spectral reference file and use that to “train” a camera profile. Absent that for skin tones, there are reference datasets for such, I have a dcamprof- digestible version of the Lippmam 2000 skin tone dataset that should do the trick.
Messing with hues using sliders is just guessing based on personal recall of the scene.
Sorry for sounding grumpy, just feel the need to put this in the context of measurement…
If you hover on the patch in the colorpicker that gives you the color name…ethnic skin tones are there as a defined range of values… you can sample your skin and adjust to try and hit it and see… often they overlap and you have to lighten the skin to separate between the data provided for a few groups…I forget where the values come from but the reference is in the code I just couldn’t find the right module spot in a quick search…
Edit:
Ah its here with avg LAB values and a stdev for a range of values around them…
src/common/color_vocabulary.c
I am less interested in technically correct which varies between people anyway. Color blindness being an extreme example. I am more interested in the artistic choice or what might be called ‘pretty’. This is not too dissimilar to the picture styles camera manufacturers develop for their camera. A landscape style will look different to a portrait style. I prefer to go as directly to this look as possible. Just my 2 cents and not trying to discount your view.
Same for me. But I use the stand alone primaries module because I like the tint hue slider to add some kodak warm to the photo.
Indeed. When I worked in publishing/printing, it was impressed on me that one cannot authoritatively colour correct without the original in front of you, in the same light.
But in practice, we cannot do much better than “just guessing based on personal recall of the scene.”
I can’t call the musicians to my house at 3.00am. Landscape photographers? hahaha.
oh! A brain cell just tipped out a memory from about 60 years ago. A photoshoot with my baby cousin. The studio did keep small clippings from my shirt and her dress!
As to colour standards for “ethnic” skin, Here, at least, I don’t believe it can be done. This country has every shade and more.
I do agree with “make it correct; make it pretty.” I would add, at least make it realistic. I’ve heard tales of top pros, top magazines, even for cover pics, actually not succeeding at that!
Above all I want my musician subjects to like their photos. I don’t do it for me: I do it for them. (It didn’t start that way: it started as just wanting half-decent pics as a diary thing.)
Even if it is only family, same thing applies. The only time it doesn’t is when a human subject is anonymous. And even then, I don’t want to make them ugly!
This is also what Aurélien Pierre advocates. First, the scene is white balanced according to the illuminant (neutral colours), then colour-graded. He demonstrates this with the following image paid (“Photo licensed under Creative Commons by Andreas Schneider”, processing and demonstration by Aurélien Pierre)
The process is valid for darktable, as well:
Taking the suggestion from this topic, I am increasingly interested in using AgX presets just like film stock.
That is to say, I would be happy to just apply a preset developed by knowledgeable people without feeling the urge to tweak anything. AgX is versatile enough to support this.
This is not to say that all those options are useless, it’s just that for 99% of my images, I am not interested in customizing the tonemapper if there is a look I like.
Having a multitude of such presets would be nice.
But I wonder if these presets would hold true for files from different makes and models of cameras? Just asking as I am unsure.
