Adversarial verification for AI agents: an independent skeptic attempts to refute a claim before the agent acts on it, returning a structured verdict (refuted / supported / inconclusive) with calibrated confidence, findings, and citations. Three tiers: quick reasoning check ($0.03), single skeptic with live web research ($0.20), three-skeptic panel with majority verdict ($0.75). USDC via x402 on Algorand, Base, and Solana. No accounts, no API keys.