Take a look at the attached .csv input file. Taking your example of the stroop task, I modified one of the stroop examples from the Samples folder that comes with DirectRT.
If you want to give feedback of correct or incorrect, this requires defining which keypresses are correct or incorrect. However, this can't be done when using a sitmlist without random replacement. So, to achieve the effect, you have to explicitly put the each stimuli in the .csv input file. You can then use the same positive nonzero integer in the wgr field of all the trials to get random presentation of the trials without repetition of already presented trials.
Let me know if this is what you wanted.