CodeGuardian iNFT

Problem: iNFTs can be just metadata. Judges, buyers, and builders need proof that an agent actually has embedded intelligence, persistent memory, autonomous behavior, and replayable evidence.

Our solution: CodeGuardian is a minted 0G Galileo iNFT code-review agent.

It has encrypted intelligence evidence, evolving memory across three certified runs, compute/critic records, replayable traces, a dynamic policy upgrade, and a public Proof-of-Intelligence certificate.

AgentProof verifies all of it through a live explorer, passport, API, badge, SDK, and CLI. FakeAgent is included as a negative control to show the verifier rejects metadata-only agents.

The project is a pnpm TypeScript monorepo. The hosted product is a Next.js explorer on Vercel with pages for the CodeGuardian Agent Console, verifier, Passport, certificate, replay views, public API, and badge endpoint. The agent runtime generates deterministic allowlisted code-review runs, memory updates, critic/self-review records, policy upgrade evidence, and canonical trace roots. The SDK handles manifest schemas, canonical SHA-256 hashing, AES-256-GCM demo encryption helpers, proof recording, adapters, and verification logic. Solidity contracts implement an ERC-7857-style demo iNFT plus a Proof-of-Intelligence registry on 0G Galileo testnet. 0G Chain anchors the live iNFT, registry, passport, and certificate records; 0G Storage and 0G Compute are wired through adapter-shaped evidence and currently shown with honest live/hybrid labels. The most hackathon-practical part is the deterministic fallback path: the demo stays reliable while preserving the same proof model and source labels judges would see from live service-backed evidence.