llm-security-tester

What problem does it solve?

LLM-integrated applications face critical security risks including prompt injection, sensitive data leakage, supply chain attacks, and unbounded consumption that can lead to data breaches, financial loss, and compliance violations. This Skill eliminates the gap between generic security scanning and LLM-specific threat coverage by providing targeted analysis for the unique attack surface of AI-powered systems.

Core Features & Use Cases

• Full OWASP LLM Top 10 v2 (2025) Coverage: Detects all 10 LLM-specific vulnerability categories, including new 2025 additions like LLM07 System Prompt Leakage and LLM08 Vector and Embedding Weaknesses.
• MITRE ATLAS v5.4.0 Mapping: Aligns every finding to recognized adversary tactics for consistent threat prioritization and compliance reporting.
• Language-Specific Remediation Patterns: Provides safe, production-ready code examples for Python, C#, Java, and TypeScript to fix each identified vulnerability.
• Use Case: For a SaaS product with a RAG knowledge base and agentic customer support tools, use this Skill to scan for indirect prompt injection in ingested documents, MCP tool poisoning, and cross-tenant vector store leaks, then implement concrete fixes to meet 2026 security best practices.

Quick Start

Use the llm-security-tester skill to scan my LLM-powered customer support agent for OWASP LLM Top 10 and MITRE ATLAS vulnerabilities, including prompt injection and MCP tool poisoning risks.