What problem does it solve? Security boundaries, input validators, and prompt pipelines often ship with undiscovered bypasses because passive code review only looks for defects rather than actively attacking the control surface. This Skill tasks an independent verifier with generating exact adversarial inputs — payloads, malformed strings, injection prompts — designed to bypass, crash, or trigger misbehavior in a system you just built. ## Core Features & Use Cases - Adversarial Input Generation: Produces exact attack strings, payloads, and scenarios against a specified target with a defined threat model and success criterion (bypass, crash, or misbehavior). - Verifier Independence Contract: Enforces caller-verified reviewer lineage so the red-team pass comes from a model family distinct from both the primary and the artifact author. - Finding Verification Loop: Requires each claimed attack to be tested against the real system, categorized by attack class, scored by severity, and paired with concrete fix recommendations. - Use Case: After building a login rate-limiter, dispatch the verifier to generate inputs that desynchronize its per-IP counter via X-Forwarded-For spoofing, then test each claim locally before reporting. ## Quick Start Ask the agent to red-team the validation logic or prompt pipeline you just built and generate ten concrete adversarial inputs that bypass it.
