Skill UI

This comprehensive benchmark suite evaluates the performance of Vision-Language Models (VLM) on complex video anomaly detection tasks using the SmartHome-Bench dataset. It covers a wide range of real-world smart home scenarios, including security breaches, elderly fall detection, and unusual pet behavior. Unlike single-frame analysis, this benchmark requires deep multi-frame video understanding, making it a robust tool for testing the advanced AI capabilities of video monitoring systems.

Offensive cloud security methodology for AWS, Azure, and GCP. Covers identity discovery, credential harvesting, enumeration, privilege escalation, persistence, data exfiltration, lateral movement, serverless abuse, Kubernetes paths, and detection evasion during authorized assessments.

A comprehensive guide detailing advanced security threats in Machine Learning and AI systems. Covers model supply chain attacks (RCE via pickle, poisoning), adversarial examples (FGSM, PGD, physical-world), data privacy breaches, and LLM-specific vulnerabilities. Essential for building robust, secure AI pipelines.

A comprehensive expert playbook detailing advanced Large Language Model (LLM) prompt injection techniques. Covers direct methods (instruction override, role play), indirect attacks (RAG poisoning, web data manipulation), and sophisticated tool/function calling abuse for data exfiltration and security bypass. Essential for rigorous AI application security testing.

Automated system designed to monitor paste sites (e.g., Pastebin, GitHub Gists) for leaked credentials, API keys, and sensitive data dumps. It uses advanced scraping, keyword matching, and regex patterns to provide early detection of data breaches and security incidents.

This skill provides a repeatable assessment for RAG systems, testing for critical vulnerabilities in vector stores and embedding models. It specifically addresses embedding inversion, cross-tenant data leakage, and knowledge-base poisoning, aligning with OWASP LLM08:2025. Use when performing authorized security audits of retrieval layers and multi-tenant isolation.

A comprehensive security skill designed to detect and mitigate indirect prompt injection attacks. It scans untrusted external content—including web pages, PDFs, and images—for hidden, obfuscated instructions before an LLM agent processes them. The pipeline combines content extraction (HTML, OCR), normalization, and dedicated ML models (e.g., LLM Guard, Prompt Guard 2) to ensure data safety.

NVIDIA garak is an open-source LLM vulnerability scanner designed for comprehensive red-teaming and security assessment. It systematically probes large language models (via OpenAI-compatible APIs, Hugging Face, etc.) using thousands of adversarial prompts. It covers critical vulnerabilities such as prompt injection, jailbreaking, data leakage, and toxicity. Use this tool to establish a repeatable security baseline before or after deployment, validate the effectiveness of safety guardrails, or generate quantitative evidence for AI risk assessments.

This skill provides comprehensive techniques for security-testing Retrieval-Augmented Generation (RAG) pipelines. It probes for indirect prompt injection, which occurs when malicious instructions are embedded in retrieved documents, and tests for embedding-space manipulation. Using advanced tools like NVIDIA garak, Promptfoo, and PyRIT against vector stores (FAISS, Chroma, etc.), users can validate retrieval guardrails and secure CI/CD gates against adversarial inputs and data leakage.

Provides security guidelines and best practices for LLM-based trading agents, covering risk management, data protection, and safe deployment.

This guide provides rigorous methodology for authors submitting security and machine learning research to top conferences. It emphasizes the need to 'freeze the world' of measurement by pinning target versions, microarchitectural states, and network conditions. It details how to report probabilistic evidence (like attack success rates) and manage ethical data release, ensuring findings remain valid and verifiable over time.

This skill deploys the OpenMed Model Context Protocol (MCP) server, exposing critical clinical NLP tools such as NER, PII extraction, and de-identification. It enables AI coding agents and chat clients to interact with OpenMed's capabilities via simple tool calls, running locally on-device to ensure user data privacy and security. Supports both stdio and Streamable HTTP deployment modes.