What problem does it solve? Settled proposals, plans, and conclusions often carry hidden assumptions, unexamined incentives, and untested failure modes that friendly review misses. This Skill applies a disciplined adversarial review that steelmans the artifact first, then challenges it across relevant lanes while keeping speculation, demonstrated defects, and value disagreements clearly separated. ## Core Features & Use Cases - Steelman-first adversarial review: Builds the strongest fair case for the artifact before criticizing it, so findings target real weaknesses rather than strawmen. - Classified finding register: Assigns every finding exactly one class (demonstrated-defect, plausible-risk, speculative-case, or value-disagreement) with evidence IDs, severity rationale, and closure criteria. - Coverage map and counterargument analysis: Discloses which adversarial lanes were tested, tests the strongest counterargument, and states what evidence would reverse each conclusion. - Use Case: Before approving a security-sensitive architecture proposal, run an adversarial review with a threat frame and restricted sensitivity to surface plausible risks, misuse scenarios, and reversal conditions without exposing exploit detail. ## Quick Start Ask the AI to red-team the attached proposal with a skeptical-reader frame at standard depth and standard sensitivity.
