Cross-Model Review — Two Models, One Reconciled Verdict

A single model reviews a diff with a single set of blind spots. Running a second model from a different vendor, with no knowledge of the first review, surfaces findings the first pass structurally could not reach — and independently re-rates severity the first pass got wrong.

The value is not "two reviews." It's that the overlap is earned: when both models independently land on the same finding, that is corroboration. When only one does, that is a lead to verify — never a finding to publish.

Output: {SKILL_OUTPUT_DIR}/cross-model-review/ — see ../OUTPUT.md

When NOT to use this

Skip it and just review normally when: