you are right, RLHF penalties fix the root cause. Until then, these skills is the best runtime defense: hard integrity gates + citation anchors that block unverifiable output.

→ View original post on X — @alphasignalai