Ai Security

This goes after the LLM attack surface with practical payloads and detection patterns. You get direct injection techniques, jailbreak methods, RAG pipeline poisoning, and model extraction testing vectors. It covers the full AI supply chain from malicious pickle files in HuggingFace models to LangChain template injection. The tool calling exploitation section is especially useful if you're assessing agentic systems where one compromised agent can poison others downstream. Includes YARA rules for prompt injection detection and concrete examples of adversarial input crafting. This is offensive work, so you're actively trying to break guardrails and extract training data. Good for red team engagements against AI products or validating your own LLM application defenses before someone else finds the holes.

GitHub stars: 280
last updated: May 28, 2026

View on GitHub

Install

$ npx -y skills add hypnguyen1209/offensive-claude --skill ai-security --agent claude-code

Installs into .claude/skills of the current project.

Files

AI/ML Security

When to Activate

Red-teaming an LLM/chatbot/copilot for direct & indirect prompt injection and multi-turn jailbreaks.
Testing a RAG pipeline for document/embedding poisoning, embedding inversion, and cross-tenant retrieval leakage.
Auditing an AI agent / MCP server for tool poisoning, excessive agency, and command injection (RCE).
Scanning a model artifact (HuggingFace, .pt/.pkl/.bin/.gguf) for deserialization payloads before loading it.
Assessing a model API for extraction/distillation, membership inference, and adversarial-suffix robustness.
Mapping findings to OWASP LLM Top-10 (2025) + MITRE ATLAS for a report.