skills.rest
brucesongs
brucesongsCOMMUNITY
Explore all skills by this author →
GitHub

ai-agent-security

Run six-phase red-team engagements against stateful, tool-using AI agents.

60|14|Updated Apr 27, 2026
Search Tags:#mcp#sandbox#red-team#security-testing#mult-agent#ai-agent-security#memory-poisoning
One-click install
npx skills add https://github.com/brucesongs/kali-claw --skill ai-agent-security-brucesongs
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ai-agent-security
Source: https://github.com/brucesongs/kali-claw/tree/main/skills/ai-agent-security
Command: npx skills add https://github.com/brucesongs/kali-claw --skill ai-agent-security-brucesongs

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Identify and exploit vulnerabilities in stateful, tool-using AI agents to reveal attack surfaces and real-world risks.

Core Features & Use Cases

  • Six-phase agent red-teaming workflow covering perimeter reconnaissance, MCP enumeration, tool poisoning, indirect injection, memory manipulation, multi-agent compromise, autonomous hijack, and comprehensive reporting.
  • Payloads, playbooks, and practical guidance for testing agent sandboxes, memory stores, and inter-agent trust models, mapped to OWASP LLM Top 10 and emerging Agent ATT&CK.
  • Real-world use cases include pre-deployment security reviews of deployed agents, penetration testing of agent tool graphs, and development of effective guardrails and detections.

Quick Start

Follow the six-phase AI agent security playbook to scope, execute, and document an end-to-end red-team engagement against an AI agent.

Frequently Asked Questions about ai-agent-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I perform a red-team security test on an autonomous AI agent?▼

To red-team an autonomous AI agent, execute a six-phase engagement workflow covering perimeter reconnaissance, MCP enumeration, tool poisoning, memory manipulation, and comprehensive reporting to reveal real-world attack surfaces.

What is MCP protocol enumeration in AI agent security testing?▼

MCP protocol enumeration is a red-teaming phase that identifies exposed tool interfaces and external call surfaces in tool-using AI agents to map potential vulnerabilities and security risks.

How do I test for memory poisoning vulnerabilities in stateful AI agents?▼

Test for memory poisoning by injecting RAG and memory manipulation payloads into the agent's persistent context across turns to evaluate how stateful memory stores can be compromised.

Does this red-teaming workflow map to the OWASP LLM Top 10?▼

Yes, the red-teaming payloads and playbooks for testing agent sandboxes, memory stores, and inter-agent trust models are explicitly mapped to the OWASP LLM Top 10 and emerging Agent ATT&CK frameworks.

What are the limitations of penetration testing AI agent tool graphs?▼

Penetration testing AI agent tool graphs is limited by sandbox boundaries and requires an engagement contract to scope testing, as unauthorized exploitation of inter-agent trust models can cause autonomous hijack risks.

Related Skills

Explore More Software Engineering Skills→
legioncodeinclegioncodeinc
72

mcp-protocol-stinger

Audit MCP servers and tool contracts for protocol compatibility.

Official
Advanced
Kaikei-eKaikei-e
9

security-auditor

Audit codebases for security vulnerabilities and produce structured OWASP-aligned findings reports.

Community
Advanced
xjtu-wangxjtu-wang
4

competition-prompt-injection

Analyze prompt-injection and related attacks within a sandboxed agent chain.

Community
Advanced
MatrixFounderMatrixFounder
4

skill-adversarial-security

Audit codebases and LLM integrations for security vulnerabilities using OWASP standards.

Community
Advanced