CONTENT SAFETY

Aegis

Content safety scanner for AI agent outputs.

57 detection rules across 5 threat categories. Zero dependencies. Runs in Node.js, browsers, and edge runtimes.

$npm install @authensor/aegisCopy
USAGE

Three lines to scan any content.

scanner.ts
import { scan } from '@authensor/aegis';

const result = scan(agentOutput);

if (result.threatLevel === 'critical') {
  // Block the output
  return { error: 'Content blocked by safety policy' };
}

// Safe to proceed
return result.clean ? agentOutput : sanitize(agentOutput);
DETECTORS

57 rules. Five categories.

Each detection rule is a battle-tested regex pattern. No ML models, no API calls, no latency surprises.

Prompt Injection

20 rules

Catches attempts to hijack agent behavior through crafted input. Covers direct injection, indirect injection via tool outputs, and multi-turn escalation patterns.

Instruction overrideRole manipulationDelimiter injectionEncoding attacksFew-shot poisoning

PII Detection

12 rules

Identifies personally identifiable information before it leaves your system. Supports US, EU, and international PII formats.

SSNEmail addressesPhone numbersCredit card numbersPassport numbersIP addresses

Credential Exposure

10 rules

Stops leaked credentials from reaching external services. Pattern-matched against real-world credential formats from major cloud providers.

API keysAWS secretsJWT tokensPrivate keysDatabase connection strings

Data Exfiltration

8 rules

Detects attempts to sneak data out through encoded channels, unusual URL patterns, and covert communication techniques.

Encoded payloadsDNS tunneling patternsUnusual URL parametersBase64-wrapped data

Code Safety

7 rules

Flags dangerous code patterns in agent-generated output. Catches shell injection, arbitrary file access, and unsafe eval usage.

eval() callsShell injectionFile system accessNetwork callsProcess spawning
BY THE NUMBERS

Built to be fast, small, and auditable.

0
npm dependencies
57
detection rules
5
threat categories
<1ms
typical scan time

Scan your first content in 30 seconds.