Cross-Model Peer Review

The protocol is the product, this plugin is its Claude Code implementation. The normative text is ${CLAUDE_PLUGIN_ROOT}/skills/cross-model-peer-review/references/PROTOCOL.md: harness-independent, provider-independent, its requirement numbers stable identifiers that this skill and every agent in this plugin cite by number rather than restate loosely. Read it before touching a run that behaves unexpectedly; this skill explains why the rules exist and when to reach for the command at all, not what each phase does mechanically.

The doctrine

Two statements govern everything else in the protocol. Both are quoted verbatim from PROTOCOL.md's Doctrine section and R14.

Cross-family independence is the strongest practical model-level independence available in this workflow, not absolute independence. Two frontier participants may share a training corpus, conventions, and reasoning patterns. What makes a second participant worth its cost is that its errors are sufficiently decorrelated, not that it is a clean-room observer.