adversarial-review

An adversarial code review deploys opposing-model reviewers to critically evaluate diffs from distinct lenses. The skill orchestrates parallel execution of these reviewers and compiles a synthesized verdict with findings and a lead judgment.

No, you cannot automatically apply changes after a multi-model review. The skill explicitly outputs a synthesized verdict and a set of findings without applying changes automatically, leaving the final modifications to the developer.

Multi-model diff analysis works by spawning reviewers on the opposing model, such as Claude spawning Codex, to challenge work. It enforces opposite-model execution to surface issues from multiple critical lenses and synthesizes a final verdict.

You should use an adversarial review for large diffs, planning sessions, or complex changes. It is designed to surface issues from multiple critical lenses that a single reviewer might miss when evaluating extensive or intricate code modifications.

No, setting up dependencies for the opposing model is not required. The skill manages multi-model orchestration internally, automatically handling the deployment of opposing models like Claude and Codex without external dependency configuration.