There's nothing automated in the software for that kind of logic, unfortunately. You would probably need to create a CSV of that kind of a variable (0=bad, 1=good, or something similar) and import it in to SMRT to use as a filter.
However, you should consider if this is the best approach. If you were using HB, and had a large number/proportion of these "bad" respondents, you probably would want to remove them prior to utility estimation with a filter back in SSI Web.
As a counter-argument, though, it's possible that someone could choose the same answer several times in a row in the conjoint. Since there's no correlation between the spot a concept shows up on the screen and the levels it contains, it might be that the same concept has the best combination of levels for a person several screens in a row.
In general I would recommend an approach that uses a few things, such as straightlining, time to complete the survey/conjoint portion, and probably the fit statistic that gets calculated to identify "bad" respondents, rather than solely relying on one metric alone.