There's some confusion about terminology here.  When CONN does despiking, it's actually not going through and removing frames that display spiking behavior.  It utilizes a squashing function, so it pounds datapoints that are far from the median value down, or up, depending on the direction.  So it just pulls extreme points down.

The scrubbing covariate is derived from outside toolboxes, some of which are integrated into CONN, such as ART.  If you run ART, it will tag volumes that display aberrant behavior (in ART, you can exclude based on changes in global signal intensity, and scan-to-scan motion).  This toolbox outputs .mat files that can be used as first level scrubbing covariate, in other words the effect of those volumes gets regressed out in the nuisance regression.  If you have data input into CONN, you can have it go through and automatically generate these covariates by running ART through the preprocessing tab.

Best,
Jon