AgentGuard VerifyPay is a policy-controlled escrow layer for autonomous AI agents on Arbitrum. It lets a Buyer Agent hire a Seller Agent within explicit budget, asset, risk, and expiry rules. Quotes that violate policy are BLOCKED, invalid results are FROZEN, and payment is released only after independent verification with publicly verifiable evidence.
AgentGuard VerifyPay is a policy-controlled execution and escrow layer for autonomous AI agents on Arbitrum. A Buyer Agent can hire a specialized Seller Agent under explicit budget, asset, risk, and expiry rules. The workflow checks every stage: quote, task creation, submission, verification, and settlement. Over-budget or unauthorized quotes are marked BLOCKED, invalid or unverifiable work is FROZEN, and payment is released only after verification. Each execution produces a policy hash, evidence hash, agent identities, and publicly verifiable Arbitrum Sepolia transactions. The project demonstrates a real Treasury Agent-to-YieldScout task and a judge-friendly Marketplace for inspecting execution evidence.
https://x.com/0xCaptain888/status/2092514296383357374
During the buildathon, we adapted AgentGuard VerifyPay for Arbitrum and completed a working end-to-end testnet implementation. We deployed and verified the PolicyEscrowV3 contract on Arbitrum Sepolia, configured Buyer and Seller Agent identities, and executed a real Buyer-to-Seller task between treasury-agent and yield-scout. The task progressed through task creation, result submission, independent verification, and payment release, with public transaction hashes and evidence hashes. We also implemented BLOCKED and FROZEN demonstrations, added a network-specific Marketplace view, and published documentation, judge instructions, security notes, and reproducible verification commands. The current release is a testnet demo and is not presented as a mainnet production system.
Bootstrapped / No external funding
