Loops HouseAegis

NEAR HACKATHON
Links
Team
1 member- MUOwner
Musa Falahul Akbar
Overview
AEGIS is a security gate for AI agents: no agent should execute what it hasn't verified. Agents are now told to execute instructions written by third parties -- mint pages that embed playbooks addressed to the agent, skill packages served from mutable 'latest' URLs, onboarding docs that end with 'paste this prompt into your AI', agent forums where a significant share of posts are injection material. One pasted paragraph can drain the owner's wallet. AEGIS verifies before anyone executes: point it at a contract address, a signature payload, a claim, or a page of instructions, and four gates -- provenance, contract, signature (full EIP-712 structural decode), and instruction channel -- return ALLOW, WARN, or BLOCK with line-level evidence. Every verdict report is hashed, and the hash is anchored in a registry contract on NEAR testnet, so anyone can re-run the same inputs, recompute the hash, and verify it on-chain without trusting our server. NEAR is the neutral trust layer of the agent economy: one agent identity (rainyy.ai), verifiable verdicts, chain-agnostic reach -- the gates check EVM contracts and payloads today. Built from real cases: Nonce Protocol's server-orchestrated mint was BLOCKED, Hashstone's verified direct freemint was ALLOWED, and an unpinned skill package was WARNED until its hash was pinned. Scope: AEGIS only reads -- it never signs or executes -- and its known limitations are disclosed openly in the repo (README + AUDIT.md).