AI keeps turning every tiny decision into an ADR

I've been using this skill for a long time, and I've ended up with 70 ADRs in one repo. We already have ADR-FORMAT.md to keep the ADR format/content in check, but the AI keeps treating anything that isn't a hard rule as just a suggestion—or treating suggestions as if they're mandatory. Once there are more than ~20 ADRs, especially after grill-with-docs, it tends to turn every decision that comes up into an ADR. It'll even do it for super detailed stuff. I've also noticed it leans more toward writing RFCs. This is just what I've seen in my repo. I can't think of a good fix, other than after every chat telling it, "Hey, you shouldn't write this as an ADR because blah blah."

After I noticed the ADR bloat, I did take steps to force-cull subpar ADRs. But is there a better way to maintain ADRs? ADRs are important docs, so their quality shouldn't depend only on humans. I think there should be a corresponding skill/review process to maintain and ensure ADR quality. And not just ADRs—I think CONTEXT.md should also be included in the review scope.