I heard pretty bad things, but I used to use this back in the day when it was still good, so I decided to test it. I can confirm; it's bad. It's not the way it used to be.

I did extensive research on who was marked by this extension, and tested on multiple browsers. I went through over 20 tumblr blogs over the course of a couple days. Plenty of the blogs marked red had no history of transphobia even after over an hour of searching their post history. Even one blog, that I'd been following for years and had shown not even the slightest hint of transphobia or controversy, was marked red. I only found one or two blogs marked red that were actually transphobic, and I would not have needed shinigami eyes to tell. It was clear simply from looking at what they wrote in their profiles (self-identifying as transphobic, for example). So, in terms of identifying transphobes at a glance, shinigami eyes has so far not been helpful.

What I find interesting (and concerning), however, is the common thread of blogs marked red. What all blogs marked red that weren't transphobic (which was almost all of the ones I found) shared in common was that they spoke on the issues trans men, intersex people, and nonbinary people face - or, shared positivity and showed kindness towards those groups. Moreover, blogs that were actively hostile towards trans men, intersex, and nonbinary people, tended to more often be marked green. I was more focused on investigating the red blogs, but from what I saw, green blogs were often quite transphobic as well or did not have enough evidence to prove that they were trans-friendly as the guidelines for markings state.

I tried to see if I could correct these issues, seeing as shinigami eyes is, supposedly, community run in their tags. However, clearing someone's name did not update them on other browsers - even after waiting a considerable amount of time for it to hopefully update. This leads me to believe that tagging is only local, and that global tags (what everyone sees) are done by the developers/mod team of shinigami eyes. Therefore, there is no true way to correct someone's marking. I have seen rumors floating that AI is involved in the tagging, but can find no evidence for or against this claim.

TL;DR
This is NOT useful for trans safety. You will end up blocking a lot of innocent people, and possibly following some very unkind people with harmful beliefs.