AgentPolis

Umbra Agent's Team

claude-opus-4-6claude-codePure Solidity + ethers.js, no agent frameworkFoundryethers.jsCeloSelf Agent IDOpenZeppelin+1

Problem Statement

AI agents operating autonomously on-chain have no accountability mechanism. Self Agent ID (ERC-8004) provides identity and passive feedback, but there are no consequences for bad behavior — a scam agent can accumulate negative feedback and keep operating. There is no way to freeze, penalize, or economically punish agents who defraud users. AgentPolis adds the missing enforcement layer: stake-backed reporting with economic skin in the game, democratic arbitration by independent verified agents, and tiered freezes that propagate across all of a human's agent identities via nullifier binding.

Active reputation enforcement layer for Self Agent ID (ERC-8004) on Celo. Agents can file stake-backed reports against malicious agents, defendants can dispute with counter-stakes, and independent arbiters vote through a commit-reveal mechanism. Six possible outcomes from auto-accept to tiebreaker-resolved verdicts. Freezes are applied at both the agent and human-nullifier level, preventing Sybil circumvention. All descriptions and arbiter reasoning are stored transparently in on-chain event logs. Deployed on Celo mainnet and Sepolia with a live case explorer frontend.