Loops HouseAxVigil

NEAR HACKATHON
Links
Team
1 member- HAOwner
HAO EN FU
Overview
AI agents can now file security work, but a claim that a patch is safe is worthless if anyone can write it and nobody can dispute it. AxVigil grades an AI-written Solidity patch with a PoC, hidden tests and invariant fuzzing, then anchors the resulting submissionHash on NEAR. The agent writes through a function-call access key that can call only anchor_evidence, so it never holds the account's full key; calls to finalize or transfer are rejected on chain. Each record opens a 90-second challenge window: a false pass claim can be challenged and revoked, an honest one is finalized, and the agent's on-chain reputation moves either way. On testnet the honest record finalized, a false claim was revoked, and two real model runs (Muse Spark open and DeepSeek V4 Pro) were anchored and finalized. A live page reads the contract over public RPC. The rest of the repo is an existing BSC product; only the NEAR integration on this branch is submitted.