I've killed my PICKIT2 (I know, not same, but very similar analog design I believe) with the logic analyser tool. I had a MCU powered from it, was probing some SPI signals with it, and didn't trigger at all. I pulled what I believed to be my trigger signal from the board and started poking around. I had it set to trigger when it sees a '0' (CS line). So I thought: does it work at all? Let's connect it to ground.
It just so happens 1 data signal and power are located next to each other. So I ofcourse picked the wrong one. Why? Because I didn't use the right color of wiring from the PICKIT2 connector on my breadboard.
Because the PICKIT2 software was 'BUSY' and polling for a trigger, the software protection for overvoltage, current limiting etc. didn't work. After holding the wire for 2 seconds some magic smoke came out.
I burned a N/P-MOSFET SOT23-6 chip for switching the VDD target power and a schottky diode. I replaced them and it still works today, although the voltage output calibration is incorrect because I couldn't be bothered to place a similarly specced diode in there. Recalibrating doesn't help because of a higher voltage drop under load.
I am uncertain whether PICKIT3 has this bug too, but it is certainly a possibility (and fault from MCH) that there is no protection on the pickit rails if the software is so called 'busy'.
If I connect to a self-powered target, I force the software to never power the target. I don't want the software to figure out anymore whether it needs to supply power or not - I want to be consent of my own test setup.