English | 简体中文 | Español
Introduction
- EN The Ultimate Open-Source Security Arsenal for Hackers, Enterprises, and AI Agents
- 中文 面向极客、企业与 AI 智能体的全域开源网络安全工具矩阵
- ES El arsenal definitivo de seguridad de código abierto para hackers, empresas y agentes de IA
| Sponsor |
About |
 |
维塔斯AI聚合 是企业级 AI 大模型 API 聚合服务：开发者只需一个 Key，即可统一调用 GPT、Claude、DeepSeek、Gemini 等 200+ 模型（文本/生图/视频），完全兼容 OpenAI 协议，无需海外账号与信用卡 https://www.aiwts.com/ |
Contents
🤖 Security For AI & AI For Security
AI Red Teaming Frameworks & Agents
Mature autonomous agents, plus foundational frameworks to build your own.
- https://github.com/confident-ai/deepteam - Open-source red teaming framework for LLM systems — “penetration testing, but for LLMs”: 50+ ready-to-use vulnerabilities with local LLM-as-a-Judge scoring, simulating jailbreaks, prompt injection and multi-turn exploitation across AI agents, RAG pipelines and chatbots, plus guardrails for production
- https://github.com/vxcontrol/pentagi - Fully autonomous multi-agent pentesting system in Go — runs whole engagements inside an isolated Docker sandbox with 20+ built-in tools (nmap, metasploit, sqlmap), long-term memory, optional Neo4j knowledge graph, and a web console for human-in-the-loop monitoring
- https://github.com/oritera/Cairn - A general-purpose state-space search engine validated on autonomous penetration testing — no predefined roles or workflows, purely goal-driven pathfinding
- https://github.com/KeygraphHQ/shannon - Autonomous white-box AI pentester for web apps & APIs — analyzes source code, identifies attack vectors, and executes real exploits pre-production
- https://github.com/SanMuzZzZz/LuaN1aoAgent - LuaN1ao (鸾鸟) - A next-generation Autonomous Penetration Testing Agent powered by LLMs, integrating P-E-R Agent Collaboration with Causal Graph Reasoning to simulate security expert thinking patterns
- https://github.com/Ed1s0nZ/CyberStrikeAI - An AI-native security testing platform in Go, integrating 100+ tools, intelligent orchestration, role-based testing, skills system, full lifecycle management, and a built-in lightweight C2 framework for authorized engagements
- https://github.com/ASCIT31/Dark-Moon - Autonomous AI pentesting engine performing continuous offensive security across web, cloud, AD and Kubernetes. Uses agentic reasoning, real exploit execution and attack path analysis to deliver proof-based vulnerabilities.
- https://github.com/utkusen/promptmap - Automated prompt injection scanner for LLM applications — dual-LLM architecture (target + controller) testing 50+ attack rules covering jailbreaking, bias, and prompt stealing via white-box or black-box assessment
LLM-Powered Vulnerability Scanners
Scanners where the LLM does the hunting — beyond signature matching.
- https://github.com/vercel-labs/deepsec - An agent-powered vulnerability scanner for on-demand review of all code in existing large-scale repos, surfacing hard-to-find issues that have been lurking in applications for a long time
Security Auditing for AI Agents & Apps
Security scanners aimed at the AI systems themselves: LLMs, agents and the applications built on them.
- https://github.com/leondz/garak - LLM vulnerability scanner for hallucination, data leakage, promp injection, misinformation, toxicity generation, jailbreaks, and many other weaknesses
- https://github.com/affaan-m/agentshield - Security auditor for AI agent configurations — scans Claude Code setups for hardcoded secrets, permission misconfigs, hook injection, MCP server risks and prompt injection vectors; 268 rules across 15 modules, scored 0-100, with auto-fix, CI gating and compliance mapping (SOC 2 / PCI DSS / ISO 27001)
AI Infrastructure Attack Surface Discovery
Discovering and risk-scoring internet-exposed AI infrastructure — model servers, MCP endpoints and inference services.
- https://github.com/BishopFox/aimap - Discovers exposed AI services — fingerprints MCP servers, Ollama endpoints and other AI infrastructure via Shodan + Nuclei, runs WebSocket-based security tests and outputs risk scores
- https://github.com/cisco-ai-defense/mcp-scanner - Security scanner for Model Context Protocol (MCP) servers — combines YARA rules, LLM-based analysis and Cisco AI Defense detections to flag malicious MCP tools, tool poisoning, prompt injection and other threats across the MCP ecosystem
AI Agent Runtime Controls
Sandboxes, policy engines and approval gates that bound what agents can do.
- https://github.com/NVIDIA/OpenShell - Sandboxed runtime for autonomous AI agents — declarative YAML policies enforce boundaries across filesystem, network, process and provider access, block data exfiltration and unauthorized egress, and inject credentials only into policy-admitted endpoints; runs Claude Code, Codex, OpenCode and Copilot CLI in isolated containers on Docker, Podman, MicroVM or Kubernetes
- https://github.com/microsoft/agent-governance-toolkit - Microsoft’s governance toolkit for autonomous AI agents — a policy engine authorizing tools and actions, zero-trust agent identity with short-lived scoped credentials, OS-level execution sandboxing, plus reliability engineering (circuit breakers, budget caps, audit trails); ships reference policy packs mapped to all 10 OWASP Agentic Top 10 categories
- https://github.com/agentkitai/agentgate - Approval workflow engine for AI agents — define policies to auto-approve safe actions, auto-deny dangerous ones, and route the rest to a human via dashboard, Slack, Discord, or email
- https://github.com/kubernetes-sigs/agent-sandbox - Kubernetes-native sandbox for AI agent workloads — the Sandbox CRD manages isolated, stateful, singleton pods with stable identity and persistent storage, supports gVisor/Kata secure runtimes, deep hibernation and auto-resume, plus SandboxTemplate/SandboxClaim/SandboxWarmPool extensions and Go/Python SDKs
Security Auditing for Agent Skills
Scan agent skill packages for malicious patterns before you install them.
- https://github.com/NVIDIA/SkillSpector - Security scanner for AI agent skills — detect vulnerabilities, malicious patterns, and security risks before installing agent skills (Claude Code, Codex CLI, Gemini CLI, etc.)
- https://github.com/cisco-ai-defense/skill-scanner - A best-effort security scanner for AI Agent Skills — detects prompt injection, data exfiltration, and malicious code patterns via pattern-based detection (YAML+YARA), LLM-as-a-judge, and behavioral dataflow analysis. Supports OpenAI Codex Skills, Cursor Agent Skills, and Claude Code commands.
Skill packs that teach coding agents to find and fix vulnerabilities end to end.
Unified tooling that equips agents for professional security work.
- https://github.com/h5i-dev/h5i - Agent-native web security workspace — hands AI agents a unified toolkit (browser automation, HTTP traffic capture / tamper / replay, recon crawling) to red-team web apps inside a configurable sandbox, replays confirmed attack flows in CI/CD as security regression checks, and formally verifies Rust application logic against Lean 4
AI Security Benchmarks & Evaluation
Benchmarks and leaderboards that quantify the security of AI systems and AI-generated code.
- https://github.com/Tencent/AICGSecEval - A.S.E — repository-level benchmark for the security of AI-generated code by Tencent’s Wukong team: tasks built from real GitHub projects and CVE patches, project-context extraction to simulate real AI-assisted development, and hybrid static + dynamic scoring (SAST rules plus test cases and vulnerability PoCs) across 29 CWE types in C/C++, Java, Python, PHP and JavaScript; evaluates both raw LLMs and agentic coding tools (e.g. Claude Code) with a public leaderboard
AI Security Verification Standards
Testable standards and checklists that define verifiable AI security.
- https://github.com/OWASP/AISVS - OWASP Artificial Intelligence Security Verification Standard — 213 testable requirements (“Verify that…”) across 12 control families and 44 sections, graded L1/L2/L3, covering training data, model supply chain, RAG memory, agent and MCP security; designed to be verified in parallel with ASVS at the matching level
Scanners for Smart Contracts
Red Team vs Blue Team
Supply Chain Analysis(SCA)
Container and Cluster
Services fingerprint detection
Man-In-The-Middle
The framework
Wireless Pentest
Mobile Apps Packages Analysis
Binary Executables Analysis
Privacy Compliance
Subdomain Enumeration or Takeover
评论