Well...this is fun. Copy and paste... or else dejavu!
I think you are talking about Hartmut Bornneman's script (in the "Render" section correct?).
I just recently highlighted this in a tutorial I made on deconvolution.
Not much to it. You just select the range of brightnesses of the stars you want the script to sample. You can also adjust a sensitivity that affects the star detection. It scans all the stars in the image and measures those stars that meet your criteria. It then it gives you a report of how many stars and the metrics. Finally you press the C R E A T E button (geez, better get this right!) to create the synthetic PSF. As Rob mentioned, use this in the deconvolution process.
This script is something of a automation of DynamicPSF.
-adam
P.S. The more advanced thing is to see how closely the chosen Moffat shape fits the average PSF with the graph- you can select other shapes to see if the fit improves. In many cases, this is a second order effect.