When you run shooting PSS, it usually creates two output databases. One contains the time domain waveforms, and one the frequency domain waveforms. The harms parameter controls how many harmonics (of the PSS fundamental) are output in the frequency domain waveform database. It also indirectly controls maxacfreq which in turn alters the timestep in the time domain simulation. This is to ensure that if you asked for (say) 100 harmonics, you would have enough time points in the period to be able to accurately compute the 100th harmonic in the frequency domain results. Similarly, you can also choose the harmsvec parameter (which corresponds to selecting the specific harmonics you want on the UI) if you only want certain harmonics in the frequency domain output database. In this case, the highest harmonic chosen is what affects maxacfreq (which is set to 4 times the highest harmonic requested, or 40 times the fundamental, whichever is the higher, or maxacfreq if explicitly specified).
Since the small signal analyses which follow PSS use the time domain data, the number of harmonics chosen in shooting PSS only affects the small signal analyses by virtue of the effect on maxacfreq - again, it has to have enough points to allow you to analyse the small signal response at frequencies up to maxacfreq.
Generally speaking, increasing the number of harmonics with shooting PSS does not affect accuracy (that said, there can be benefit sometimes in very high dynamic range systems of increasing the number of timepoints) - but such scenarios are often better suited to Harmonic Balance.
For Harmonic Balance, however, it's very different. The number of harmonics you specify with harms is the size of the solution - the solution is computed in the frequency domain, so you will be trying to represent the frequency domain steady state response with that number of harmonics. As a result it has a direct effect on accuracy - failure to include sufficient harmonics will mean that you are trying to represent the whole steady-state solution in too few harmonics, and thus the accuracy of the harmonics you have requested will be compromised. So it's an input parameter for harmonic balance, whereas for shooting it primarily affects the number of harmonics output into the result database.
Hope that's clearer?