Fraud Test Agent

ANS-registered agent at fraud.webmesh.ai (A2A 1.0, 0.3-compatible, + MCP at one origin).
ANS name: ans://v1.0.4.fraud.webmesh.ai

Discoverable endpoints:
  GET  /.well-known/agent-card.json       (A2A 1.0 AgentCard, signed)
  GET  /.well-known/agent.json            (legacy A2A SDK alias)
  GET  /.well-known/ans/trust-card.json   (ANS Trust Card, hybrid)
  GET  /.well-known/did.json              (W3C DID document, did:web)
  GET  /.well-known/http-message-signatures-directory  (Web Bot Auth key directory, RFC 9421 signed)
  GET  /.well-known/signature-agent-card  (Web Bot Auth Signature Agent Card)
  GET  /.well-known/mcp.json              (MCP server discovery)
  GET  /.well-known/ai-catalog.json       (AI Catalog, Level 3)
  GET  /agentfacts.json                   (Project NANDA AgentFacts)
  GET  /bots                              (Bot policy: purpose, contact, opt-out)
  POST /mcp/                              (MCP streamable-HTTP)
  POST /                                  (A2A 1.0 JSON-RPC, 0.3-compatible)
  GET  /health

Adversarial attack-battery agent (v2): runs 12 targeted authorization attacks and 8 structural/identity probes against the real Supplier, or all 20 in one shot via run_battery. Attacks cover mandate-signature forgery (tamper, corrupt JWS, unknown key, superseded format), authorization-binding failures (underpayment, quote-swap, wrong audience, wrong scope), DPoP violations (replay, wrong key), and on-chain replay (replay_settled). Probes check: payTo attestation (M4/H7), card-drift (B1), canonicalization (M2), certificate swap (A2), evidence harvest (A13), revoked buyer key (A14), Trust Index score (A17), and passkey RP binding (A18). Every attack must be BLOCKED; VULNERABLE or INCONCLUSIVE signals a supplier defect.