What problem does it solve? Single-model workflows risk shipping unreviewed code, analysis, or brain pages because the same model that produced the work also judges it. This skill adds a quality gate where a different AI model reviews the work product against the originating skill's Contract before finalizing. ## Core Features & Use Cases - Cross-model review with grading: Sends the work product plus the originating skill's Contract to a different model, which returns a pass/fail verdict with cited findings and an agreement percentage. - Refusal routing: If one review model refuses, the skill silently switches to the next model in the chain defined in conventions/cross-modal.yaml, escalating to the user only if all models refuse. - Codex code-review handoff: For diff review, recommends the gstack /codex skill and frames Codex output alongside Claude's analysis with an overlap report; an adversarial mode probes security-sensitive changes for injection vectors, auth bypasses, and privilege escalation. - Use Case: After refactoring an auth flow across 8 files, invoke the skill to get an independent adversarial review rating each finding CRITICAL/HIGH/MEDIUM/LOW before merging, while keeping final approval with the user. ## Quick Start Ask the agent to get a second opinion on this diff with cross-modal review before committing.
