Skill UI

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.

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.

This skill provides comprehensive tools to detect data poisoning and model backdooring attacks throughout the machine learning lifecycle. It checks data provenance, detects label-flip anomalies (using Cleanlab), and analyzes model activations and spectra (using ART) to identify malicious triggers or compromised weights, ensuring ML supply chain integrity.

This skill provides a comprehensive defense-in-depth framework for securing autonomous AI agents. It implements least-privilege tool allowlisting, identity binding, and human-in-the-loop controls to mitigate risks like prompt injection and tool poisoning. Use this when building agents that interact with sensitive systems (e.g., payments, file writes) to ensure governance, auditability, and compliance.