@sint/gate-policy-gateway

Single enforcement choke point for AI agent actions. Every tool call, robot command, and actuator movement flows through the Policy Gateway.

Install

npm install @sint/gate-policy-gateway

Usage

import { PolicyGateway } from "@pshkv/gate-policy-gateway";
import { Tier } from "@pshkv/core";

const gateway = new PolicyGateway({
  tokenResolver: async (tokenId) => lookupToken(tokenId),
  rules: [
    { match: { actions: ["read"] }, tier: Tier.T0_OBSERVE, approve: "auto" },
    { match: { actions: ["write"] }, tier: Tier.T1_PREPARE, approve: "auto" },
    { match: { actions: ["execute"] }, tier: Tier.T3_COMMIT, approve: "human" },
  ],
});

// Intercept an action
const decision = await gateway.intercept({
  tokenId: "token-123",
  resource: "file:///workspace/output.txt",
  action: "write",
});

console.log(decision.action);  // "allow" | "deny" | "escalate"
console.log(decision.tier);    // Tier.T1_PREPARE

Features

• 5-tier approval model — T0 (observe) through T4 (critical)
• Tier assignment — automatic classification based on action + resource
• Physical constraints — velocity, force, geofence enforcement
• Forbidden combos — detect dangerous action sequences
• Circuit breaker — automatic lockout on repeated violations
• Per-server policy — maxTier ceiling, requireApproval override
• Goal hijack detection — identify manipulation attempts
• Supply chain verification — validate tool provenance
• Memory integrity — detect context tampering

Part of SINT Protocol

📖 Full documentation · 🚀 Getting Started