I don't see any test scaffolding in that repository. My python is mediocre, so I'm definitely not going to attempt to build one from scratch.
I will say that the "ying and yang" -> "yin and yang" suggestion, although it's not itself very useful because "ying" is flagged as a spelling mistake already [at least on this system] does show where "zero false positives" is somewhat practical for units larger than a single English word. Of course you would need a lot of work to establish which things are "always" errors.
For example "ad homonym" and "all intensive purposes" are almost always going to be errors, but for every few times you find "tow the line" used when "toe the line" was meant, you'll stumble over a case where an actual rope was being towed.