ValueOracle

I built ValueOracle after noticing a simple gap: AI agents can execute transactions, but they still can’t judge whether a price is actually fair. They can move fast, but they can also overpay fast.
ValueOracle adds a decision layer before money leaves a wallet. An agent sends purchase intent onchain, Chainlink CRE picks up the event, fetches external pricing and seller-quality signals, and returns a verifiable approve/reject decision that the contract enforces.
The result is practical: fewer overpriced buys, less exposure to manipulated listings, and better confidence in autonomous spending. I also added a confidential flow (commit-reveal + Confidential HTTP) so sensitive purchase details are not exposed publicly before execution.

The system has three main parts.

1. PurchaseGuard.sol on Sepolia receives purchase requests and waits for oracle fulfillment.
2. A Chainlink CRE workflow listens to PurchaseRequested, calls the decision API (standard or confidential endpoint), and returns the decision onchain.
3. A Node.js decision engine aggregates multiple mock marketplace prices plus seller trust/review data, computes a value score, and explains the verdict.
   For demoability, I included CLI commands for standard and private purchase flows, reveal, and review submission, plus CRE CLI simulation runs tied to real transaction events.