adversarial-review

Adversarial code review catches hidden flaws by challenging work with independent reviewers that think from different critical angles. It applies multi-lens critique through skeptic, architect, and minimalist perspectives to surface correctness, design, and complexity issues before shipping.

To review code changes, you spawn independent reviewers through codex exec or claude -p and synthesize their findings. This opposing-model orchestration prevents self-confirmation bias and produces a remediation-focused verdict without modifying your files.

Stress-testing implementation plans for hidden flaws is best done by running independent adversarial reviewers through skeptic, architect, and minimalist perspectives. This multi-lens critique surfaces correctness, design, and complexity issues before shipping.

No, adversarial review does not modify your code during the synthesis process. It strictly produces a single verdict and remediation-focused findings from the reviewers without making any changes to your working files.

You should use an opposing-model review when you have large diffs, implementation phases, or planning outputs that need rigorous critical scrutiny. It catches real flaws by challenging the work with independent reviewers on the opposite model.