SentinelCRE

SentinelCRE is a proactive risk monitoring protocol that evaluates every autonomous AI agent action through three independent defense layers before it executes on-chain. 
Layer 1 enforces on-chain compliance (value limits, contract whitelists, rate limits, mint caps, Proof of Reserves via Data Feeds). 
Layer 2 runs a 7-dimension behavioral risk engine that learns per-agent baselines and catches subtle attacks like sequential probing, slow drift, and off-hours exploitation. 
Layer 3 requires dual-AI consensus (Claude + GPT-4) inside a TEE via ConfidentialHTTPClient — both models must independently approve, and agents can never see the evaluation criteria.
 The problem: $3.4B+ stolen from DeFi exploits, and AI agents are now autonomously discovering vulnerabilities for $1.22 each (Anthropic, 2025). 
Current solutions are reactive — kill switches fire after the damage. SentinelCRE blocks threats before execution, with every verdict recorded immutably on-chain.

CRE Workflow (SDK v1.0.9) using 8 primitives: ConfidentialHTTPClient for TEE-backed AI evaluation, HTTPClient with ConsensusAggregationByFields for DON-level BFT consensus, EVMClient (callContract, writeReport, filterLogs, headerByNumber, logTrigger), CronCapability, and HTTPCapability across 3 trigger types. 
Smart contracts in Solidity 0.8.24 (Foundry, OpenZeppelin v5.5.0): SentinelGuardian.sol (AccessControl + Pausable, verdict processing, circuit breakers, challenge appeals) and PolicyLib.sol (7 compliance checks + Chainlink Data Feeds for Proof of Reserves). 90 tests across 5 suites. 
Dashboard built with Next.js 15, React 19, Tailwind CSS 4, deployed on Vercel. All contracts deployed on Tenderly Virtual TestNet (Sepolia fork) with live transaction monitoring, Simulation API for what-if scenarios, and a full audit trail.