One of my particular skills that isn't necessarily library related is being able to analyze systems and see where their weak points are. This can make me a bit of a pest around work ["well what happens if your plan doesn't work as expected... what is the Plan B?"] but I like to think that the stuff I design is more fault-tolerant than the average stuff. In the reference world, this is the difference between giving someone an answer versus a good answer versus "I don't know" as an answer. It can be tough to measure the cost of failure in this example, but it's not as tough in other situations. Susan Feldman talks about the high cost of not finding information.