Skill UI

A multi-layered defense system designed to detect sophisticated prompt injection attacks targeting LLM-powered applications. It combines regex pattern matching for known signatures, heuristic scoring for structural anomalies, and advanced DeBERTa-based classification. Ideal for building robust input validation layers, securing RAG pipelines, and performing proactive AI security scanning.

This skill detects AI-generated deepfake audio used in voice phishing (vishing) attacks. It functions by extracting critical spectral features (including MFCCs, spectral centroid, and zero-crossing rate) and passing them through advanced machine learning models (like Random Forest and Gradient Boosting). The system supports batch forensic analysis, provides confidence scores, and generates detailed reports to verify the authenticity of recorded voice calls, making it essential for cybersecurity incident response and fraud investigation.

This framework detects NTLM relay attacks across Active Directory environments by correlating Windows Security Event 4624 (LogonType 3). It identifies key indicators such as IP-to-hostname mismatches, Responder/LLMNR poisoning artifacts, and non-enforced SMB/LDAP signing, providing comprehensive threat detection for T1557.001.

This skill is designed for running comprehensive evaluations of Hugging Face Hub models on local hardware. It supports multiple evaluation frameworks, including `inspect-ai` and `lighteval`, and provides flexibility in backend selection (e.g., `vllm`, Transformers, or `accelerate`) to optimize for performance and compatibility. Use this for local GPU benchmarking, smoke testing, and backend selection, but note that it does not handle remote job orchestration or result publishing.

This skill provides comprehensive analysis and structured data extraction for academic research papers hosted on Hugging Face and arXiv. It can read paper content in markdown format, retrieve key metadata (authors, abstracts, GitHub repos), and identify linked assets such as associated models, datasets, and Spaces. Use this when the user provides a Hugging Face or arXiv URL, or explicitly requests a summary or explanation of an AI research topic.

This skill provides a managed cloud environment for training and fine-tuning advanced vision models. Users can build and customize models for object detection (e.g., DETR, YOLOS), image classification, and segmentation (SAM/SAM2). By leveraging Hugging Face Jobs, training is executed on powerful cloud GPUs without requiring local GPU setup, and all resulting models are automatically saved to the Hugging Face Hub.

This framework implements robust input and output validation guardrails for LLM-powered applications. It prevents prompt injection, detects and redacts PII, filters toxic content, and validates model outputs against defined schemas. Use it when deploying AI agents that handle sensitive data or require strict adherence to corporate policies, ensuring a secure and trustworthy AI interaction layer.

This sophisticated tool generates publication-quality AI illustrations and diagrams for academic papers. It implements a multi-stage, iterative workflow, using advanced LLMs (like Claude and Gemini) to parse user requests, optimize complex layouts, verify academic styling (e.g., CVPR/NeurIPS standards), and render high-fidelity figures. Ideal for researchers needing professional, complex visuals for their manuscripts.

This tool assesses whether coding agents adhere to defined skills, rules, or agent definitions. It automatically generates scenarios with varying prompt strictness, executes the agent, and generates a comprehensive report detailing the compliance rate of tool calls against predefined workflow specifications using advanced LLM classification.

This skill provides continuous, structured monitoring of deep learning model training metrics. It analyzes data from platforms like WandB and local logs to detect critical anomalies such as NaN values, divergence, plateaus, and sudden spikes. By providing a clear decision (CONTINUE, WAIT, or STOP) and detailed evidence, it ensures compute resources are only spent on viable training runs.

Transformers.js allows developers to run state-of-the-art machine learning models directly within JavaScript/TypeScript environments (browser, Node.js). It supports a comprehensive range of tasks including NLP (text generation, summarization, translation), Computer Vision (image classification, object detection), and Audio processing. This capability enables building fully client-side, multimodal AI applications without requiring a dedicated Python backend or server infrastructure.

Assesses AI/ML systems for prompt injection, jailbreak, model inversion, data poisoning, and agent tool abuse using MITRE ATLAS mapping, signature detection, and scoring metrics before deployment.