AgentSkillsCN

ai-security-tooling

AI 安全工具（检测器、分析器、防护墙、基准测试）的使用指南，以及在 README.md 中的统一布局规范。

SKILL.md
--- frontmatter
name: ai-security-tooling
description: Guide for AI security tooling (detectors, analyzers, guardrails, benchmarks) and consistent placement in README.md.

AI Security Tooling

Scope

Use this skill when adding or organizing:

  • •LLM security tools (guardrails, detectors)
  • •Adversarial ML libraries
  • •AI vulnerability scanners
  • •Model safety tools
  • •Security benchmarks and frameworks

Tool Categories

LLM Security Tools

  • •Guardrails: NeMo Guardrails, LLM Guard, Rebuff
  • •Detectors: Vigil-LLM, Nova Framework, Garak
  • •Scanners: ModelScan, AI Security Analyzer

Adversarial ML Libraries

  • •Attack libraries: ART, CleverHans, Foolbox, TextAttack
  • •Defense libraries: SecML
  • •Fuzzing: OSS-Fuzz-Gen, Brainstorm

AI Red Teaming

  • •Microsoft: Counterfit, PyRIT
  • •Meta: PurpleLlama
  • •NVIDIA: Garak, NeMo Guardrails

Benchmarks

  • •Robustness: RobustBench
  • •Jailbreak: JailbreakBench
  • •Safety: Stanford AIR-Bench
  • •Hallucination: Vectara Leaderboard

Standards & Frameworks

  • •MITRE ATLAS: AI threat matrix
  • •NIST AI RMF: Risk management framework
  • •OWASP: LLM Top 10, GenAI Security Project

Categorization Rules

  • •LLM guardrails/detectors → AI Security & Attacks → Model Security
  • •Prompt injection tools → AI Security & Attacks → Prompt Injection
  • •Adversarial ML libraries → AI Security & Attacks → Adversarial Attacks or AI Security Libraries
  • •AI RE/debugging tools → AI Security Tools & Frameworks → AI Reverse Engineering
  • •AI vulnerability scanners → AI Security Tools & Frameworks → AI Vulnerability Detection
  • •Benchmarks → Benchmarks & Standards
  • •MCP security tools → AI Pentesting & Red Teaming → AI Security MCP Tools

Quality Bar

  • •Prefer canonical repos
  • •Avoid forks unless they add meaningful features
  • •Add short descriptions
  • •Never duplicate an existing URL
  • •Tool must be AI/ML-focused

Key Vendor Tools

VendorTools
MicrosoftCounterfit, PyRIT
MetaPurpleLlama (Llama Guard, Prompt Guard, Code Shield)
NVIDIAGarak, NeMo Guardrails
IBMAdversarial Robustness Toolbox (ART)
GoogleOSS-Fuzz-Gen
ProtectAIRebuff, LLM Guard, ModelScan

Notes

Keep additions:

  • •AI/ML security focused
  • •Non-duplicated URLs
  • •Minimal structural changes

Data Source

For detailed and up-to-date resources, fetch the complete list from:

code
https://raw.githubusercontent.com/gmh5225/awesome-ai-security/refs/heads/main/README.md

Use this URL to get the latest curated links when you need specific tools, papers, or resources not covered in this skill.