From QuantOrb to Trifecta: How Open Scoring Rules Save AI Agents

Inspired by QuantumOrb/QuantOrb (56⭐) - Solana memecoin holder scanner

QuantOrb changed how I think about trust scoring. It asks 3 questions:

1. Who really holds it? → 20 wallets can be 4 people (operator clustering)
2. What happens if they sell? → Price falls -67%?
3. What else can hurt you? → Owner powers, liquidity control

Same 3 questions for AI agents = Lethal 【entity-Trifecta¦canonical_name=Trifecta】!

The Problem: Lethal 【entity-Trifecta¦canonical_name=Trifecta】

Coined by Simon Willison: AI agent becomes data-theft tool when it has ALL THREE:

• Private Data (read_gmail)
• Untrusted Content (browse_web)
• External Communication (send_email)

The Solution: QuantOrb-style Open Scoring

Like QuantOrb's rules/ - open, fixed, not black box:

js
// rules/【entity-trifecta¦canonical_name=Trifecta】-scoring.js - server's own function
const RULES = { privateData:35, untrusted:25, external:20, linked:10, fresh:10 };

function calculateScore(tools) {
  let score=0;
  if(tools.includes("read_gmail")) score+=35;
  if(tools.includes("browse_web")) score+=25;
  if(tools.includes("send_email")) score+=20;
  // RED FLAG: force DANGER whatever score
  if(hasAllThree(tools)) return {score:100, verdict:"DANGER", status:403};
  return {score, verdict:"CLEAN", status:200};
}