adversarial-thinking

Stress-test architecture decisions by generating three divergent candidates and evaluating them through isolated critique and revision cycles. This pipeline selects a winner and runner-up via pairwise comparison to ensure robust outcomes.

Adversarial thinking uses multi-agent critique where isolated attacker and defender agents evaluate divergent approaches. It enforces phase-based sub-agent isolation and Phase 2.5 checks to guide evaluation using enriched requirements with anti-requirements.

Use divergent candidate generation to produce multiple approaches, then run them through isolated critique cycles. Pairwise comparison of the revised candidates identifies the strongest solution, minimizing the risk of flawed strategic decisions.

Yes, multi-agent critique supports risk assessment in high-stakes domains by isolating attacker and defender dialogues to evaluate candidates. It outputs a rigorously stress-tested winner and runner-up for complex strategy planning.

Isolated critique cycles require enriched requirements with anti-requirements to guide evaluation effectively. Without providing three divergent approaches initially, the pairwise comparison cannot adequately select a winner to minimize decision risk.