@anticlaude/engine

Core scanning engine for AntiClaude.

Features

• Attack Runner — 64 bundled payloads with adaptive variant strategies
• Detection Engine — Rule-based + global pattern detection with confidence scoring
• LLM Judge — Optional semantic detection via OpenAI/Anthropic APIs
• MCP Scanner — Audit MCP server configurations for security issues
• Skill Auditor — Static analysis of tool/skill definitions (6 dimensions)
• Reporter — JSON, Markdown, and HTML report generation

Usage

import { runScan } from '@anticlaude/engine'

const report = await runScan({
  endpoint: 'https://your-agent.com/api/chat',
  payloadCount: 10,
})

console.log(`Score: ${report.score}/100`)

See the main repository for full documentation.