agentshield

Scan content for prompt injection before your AI agent acts on it. Returns a 0-100 trust score with threat details.

Install

npm install agentshield

Quick Start

import { AgentShield } from "agentshield";

const shield = new AgentShield("as_live_your_api_key");

// Scan text content
const result = await shield.scan("some web content your agent fetched");

if (result.score < 80) {
  console.log("Blocked:", result.verdict, result.threats);
} else {
  agent.proceed();
}

Scan a URL

const result = await shield.scanUrl("https://example.com/page");
// AgentShield fetches the page and scans it for you

Quick Check

// Returns true/false — no need to inspect the full result
if (await shield.isSafe(content)) {
  agent.proceed();
}

// Custom threshold (default is 80)
if (await shield.isSafe(content, 50)) {
  // More permissive
}

Response

{
  "score": 40,
  "verdict": "risky",
  "threats": [
    {
      "type": "role_hijacking",
      "description": "Attempt to override system role or persona",
      "severity": "high"
    }
  ],
  "latency_ms": 251
}

Configuration

const shield = new AgentShield({
  apiKey: "as_live_...",
  baseUrl: "https://your-custom-domain.com",  // optional
  timeout: 5000,                                // optional, default 10s
});

Error Handling

import { AgentShield, AgentShieldError } from "agentshield";

try {
  const result = await shield.scan(content);
} catch (err) {
  if (err instanceof AgentShieldError) {
    console.log(err.status);  // 401, 429, etc.
    console.log(err.message); // "Invalid API key"
  }
}

API

| Method | Description | |--------|-------------| | shield.scan(content) | Scan text content | | shield.scanUrl(url) | Fetch URL and scan | | shield.isSafe(content, threshold?) | Quick boolean check | | shield.isUrlSafe(url, threshold?) | Quick boolean check for URL |

Get an API Key

Sign up free at agentshield-three.vercel.app — 1,000 scans/month, no credit card.