I suppose the CRISPR people have developed some tools for designing controls for gRNAs?

If there is nothing helpful and you need to reinvent the wheel, BBTools contains a Bloom-Filter implementation, so it can separate reads that certainly do not match a genome reference (you however will need to adjust the parameters for the 2nts criterion). For 22-mers, it would be quite feasible to do the brute force approach of designing all possible sequences and running them through the filter, although that will most likely give you a few thousand possible sequences, which you could at least narrow down by the desired GC content. Lastly, you would need to check for hairpin-formation and maybe some other properties?