How to measure the direction of current harmonics?

If the generation side of the point-of-common-coupling (PCC) is essentially an islanded setup consisting of a few specific generators, then the amount of harmonics resulting from the generation "source" can be readily identified in a mathematical fashion from the relative impedance on the generation side of the PCC compared to the combined impedance on the load side of the PCC. (The source of such harmonics is related to both the rotor geometry and the stator winding geometry of the rotating supply, as well as any power electronics or other switched supplies that are on the "provision" side of the equation.)

If the generation side of the PCC is comprised of a large number of generation facilities and/or a complicated network of power supplies (such as a utility), the ability to mathematically arrive at a value is much harder, verging on the impossible because of the sheer number of variables involved.

To physically determine the "source" component of the harmonics (i.e. the facility side of a main power transformer), the ENTIRE facility load must be removed and the harmonic content measured. From a practical perspective, this is not a realistic alternative except in the absolute worst case.

You CAN shut down portions of the load and verify how the harmonic content is affected. Care must be taken to limit the effect of changing load impedances and other loading conditions as a result of the test, so that some form of "superposition" can be used to sum up all the minor culprits into one lump sum. Alternatively, precise measurements can be taken throughout the facility distribution system to determine the "harmonic signature" at all relevant points. Then, armed with a bigger picture of the harmonic distribution, it is possible to work backwards to the PCC and determine if what is measured at that point is the sum of the "load" side ... or if there's another as-yet-to-be-determined "source" for specific harmonics in the observed signature. By process of elimination - if there's still more than can be accounted for from a load side perspective, then it has to be coming from the utility. However, if you want to suggest that the utility is the issue, you'll need to have a virtually iron-clad set of data and calculations to back up your conclusions.