Skill UI

This skill uses Python and boto3 to analyze AWS CloudTrail logs. It establishes statistical baselines of normal API activity and detects various anomalies, such as unusual event sources, first-time API usage, geographic IP deviations, and high-frequency calls. This is critical for identifying credential compromise, privilege escalation, and unauthorized resource access in a security incident.

A comprehensive security tool designed to detect and defend against subtle prompt injection attacks hidden within untrusted multi-modal content, such as web pages, PDFs, and images. It scans for obfuscated payloads (e.g., hidden CSS, metadata, zero-width characters) by combining content normalization, heuristic analysis, and dedicated ML detectors, ensuring that LLM agents only consume sanitized, trustworthy input.

This skill utilizes NVIDIA's open-source garak framework to conduct comprehensive red-teaming assessments on Large Language Models (LLMs). It tests for critical vulnerabilities such as prompt injection, jailbreaks, data leakage, and toxic content generation by sending thousands of adversarial probes. Ideal for pre-deployment security validation, API guardrail testing, and generating defensible evidence for AI risk assessments.

This tool executes the advanced `metaharness/darwin` security benchmark (ADR-155). It rigorously evaluates the security detection capabilities of AI systems against a controlled ground-truth corpus, providing critical metrics like TPR, FPR, and patch-pass rates. It compares the champion model against multiple baselines, offering empirical data crucial for iterative security improvement and drift detection.