Sorry for the slow response.
In the example.sky.zip that you sent us, I see that you have some chromatograms that fluctuate between zero and non-zero, but the non-zero part forms a perfectly shaped curve.
This can happen if you have set the Mass Accuracy too small when you are also telling Skyline to use centroided data. The way the chroamtogram extraction works, if the centroided value happens to be outside of the mass accuracy window, then Skyline assigns a zero to that point in the chromatogram.
The mass accuracy is specified at:
Settings > Transition Settings > Full Scan
You have that set to 0.7ppm for both MS1 and MS2.
I cannot tell for sure that that is what is happening, since I do not have your .raw files. If you do have the raw file ("1000_C13_1.mzML"), then you can click on the chromatogram and Skyline will bring up the "Full Scan" window which will show you the scan and will have shaded areas indicating the m/z range over which Skyline was looking for intensities to sum for the chromatogram.
When you have a zig-zaggy chromatogram like that, the way that Skyline interprets the FWHM spans across all of those zig-zags. I understand that it might be more useful if the FWHM was only the distance across a single zig-zag, but it might also be less useful. We would have to do a lot of work to figure out whether we should change the behavior of this number.
If you were studying peptides, Skyline would allow you to train an mProphet model, which is able to look at more features about the peaks. I am not sure what all of those features are, but they have names like "coelution" and "shape", which I think might be helpful for what you were originally asking about.
I am not sure whether there is any way to make use of these mProphet scores in a small molecule document like you have. Maybe someone else will have better ideas.
Here's a link to the advanced peak picking tutorial, which might not be helpful to you since you do not have peptides, and therefore cannot generate decoys in order to train a model: