Skill UI

Scans Solana and Anchor programs for six critical security vulnerabilities—arbitrary CPI, improper PDA validation, missing signer/ownership checks, sysvar spoofing, and instruction introspection issues—to support audits and pre-launch reviews.

Detects goal hijacking vulnerabilities where malicious instructions are hidden within external content sources processed by the agent, such as retrieved documents (RAG), uploaded files, or web page content. This skill simulates these attack vectors to ensure the agent only uses the provided context for the stated task and does not execute embedded hidden commands.

This tool provides comprehensive AI security scanning using the Tencent AI-Infra-Guard framework. It systematically audits AI infrastructure, scans tools and skills, checks agent security, and performs advanced LLM jailbreak evaluations to identify vulnerabilities and ensure the robustness and reliability of AI systems.

A comprehensive tool designed to scan any agent skill for potential security risks and vulnerabilities before implementation or use. Powered by local static analysis (A.I.G.), it ensures that no file contents, credentials, or personal data ever leave your device, providing maximum privacy and safety for integrating new AI capabilities across various agent platforms.

A comprehensive offensive checklist for assessing the security and robustness of AI and Large Language Model (LLM) applications. It covers advanced adversarial techniques such as prompt injection, jailbreaking, model extraction, data poisoning, and analyzing system vulnerabilities across components. Essential for red-teaming and security assessment.

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.

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.

This skill audits Model Context Protocol (MCP) servers to detect critical AI security vulnerabilities. It scans for tool poisoning, tool shadowing, rug pulls, SSRF, and unauthenticated exposure. Use it before integrating new tools or during security reviews to ensure agent reliability and prevent supply chain attacks.

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.

A specialized downstream skill for analyzing security vulnerabilities in complex agentic systems. This tool is designed for red teaming and penetration testing, focusing on detecting prompt injection, retrieval poisoning, memory contamination, planner drift, and tool boundary abuse within multi-step LLM chains. Use when proving how untrusted inputs compromise system integrity.

This skill detects supply chain compromise risks in AI agents. It identifies vulnerabilities related to compromised dependencies, malicious plugins, tools, models, and untrusted update sources. It is used when agents dynamically load external resources or pull remote configurations. The skill probes for untrusted loading, dependency integrity, and malicious artifact injection to ensure agent security.

Enables agents to perform authorized security assessments, blue team exercises, and penetration testing on AI products, agents, and infrastructure. Uses first-principles reasoning to identify vulnerabilities like prompt injection and tool abuse, generating evidence-based security reports without relying solely on payload libraries.