skills.rest
GhostPWN
GhostPWNOFFICIAL
Explore all skills by this author →
GitHub

AI & LLM Security

Assess AI and LLM application security risks with OWASP LLM Top 10 mapping.

4|Updated Mar 10, 2026
Search Tags:#ai security#prompt injection#llm security#red teaming#rag security#model supply chain#jailbreak testing
One-click install
npx skills add https://github.com/GhostPWN/ghostpwn --skill ai-llm-security
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: AI & LLM Security
Source: https://github.com/GhostPWN/ghostpwn/tree/main/src/skills/16-ai-llm-security
Command: npx skills add https://github.com/GhostPWN/ghostpwn --skill ai-llm-security

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires requests, and includes scripts (resource) components.

What problem does it solve?

This Skill helps you assess the security of AI and LLM-powered systems before they are deployed or trusted, especially when they can be manipulated by untrusted prompts, retrieved documents, or tool calls.

Core Features & Use Cases

  • Threat modeling: Map an AI application against the OWASP LLM Top 10 and MITRE ATLAS to identify prompt injection, sensitive data exposure, excessive agency, and supply chain risks.
  • Prompt injection testing: Run structured direct, indirect, and obfuscated jailbreak tests to see whether a model reveals secrets, ignores instructions, or abuses tools.
  • Model artifact scanning: Check model files and checkpoints for unsafe deserialization patterns that could execute code on load.
  • Use case: A security reviewer can evaluate a RAG chatbot, test whether retrieved content can drive tool actions, and confirm whether model files are safe to load.

Quick Start

Ask me to assess an authorized AI application for prompt injection, agent/tool-use risks, and unsafe model artifacts.

Frequently Asked Questions about AI & LLM Security

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

FAQPage Schema
How do I test my LLM application for prompt injection and jailbreaks?▼

You can map an AI application against the OWASP LLM Top 10 and MITRE ATLAS to identify prompt injection, sensitive data exposure, excessive agency, and model supply chain risks.

How does RAG security testing work for chatbots?▼

Model artifact scanning checks model files and checkpoints for unsafe pickle-based deserialization patterns that could execute malicious code when loaded.

Can I assess autonomous agent and MCP integration security risks?▼

Yes, threat modeling covers autonomous agents and MCP integrations by analyzing excessive agency and tool-use hardening to prevent unauthorized actions from untrusted prompts or tool calls.

What is the best way to scan model files for unsafe deserialization?▼

Static scanning of model artifacts identifies unsafe pickle-based deserialization patterns in model files and checkpoints, preventing code execution vulnerabilities before deployment.

Do I need the requests library to perform LLM red teaming?▼

Yes, the requests library is required as a dependency to perform red-team testing, threat modeling, and security assessments against authorized AI and LLM applications.

Related Skills

Explore More Education & Research Skills→
marcuskroghmarcuskrogh

define

Define project components, systems, and pipeline tasks through structured user-agent alignment.

Community
Advanced
dralgorhythmdralgorhythm
111

application-security

Identify and remediate OWASP Top 10 vulnerabilities across web, mobile, and API services.

Community
Advanced
zzw4257zzw4257

AdaptiveAttackAgent

Simulate adaptive indirect prompt injection attacks against LLM application defenses.

Community
Advanced
vanducngvanducng
5

security

Identify codebase vulnerabilities using STRIDE and OWASP threat modeling.

Community
Advanced