Skill UI

Performs a comprehensive external audit of a deployed repository's state, going beyond simple code review. It checks for critical production gaps, including RLS misconfigurations, webhook idempotency issues, secret exposure, deployment health, and prompt injection vulnerabilities. Ideal for pre-launch gates and ensuring code stability before going live.

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 security auditing for Model Context Protocol (MCP) servers and their associated tool metadata. It is designed to detect sophisticated supply chain attacks, including tool poisoning (indirect prompt injection), Server-Side Request Forgery (SSRF), tool shadowing, and unauthenticated exposure. It is essential for securing AI agent stacks before integrating new tools or during CI/CD processes.

Automate security regression testing for Large Language Model (LLM) applications by integrating Promptfoo and DeepTeam into CI/CD pipelines. This skill enforces a critical security gate, automatically testing endpoints against OWASP LLM Top 10, OWASP Agentic threats, and common jailbreaks/prompt injections. It ensures that security patches do not regress over time due to prompt or model changes, providing continuous risk monitoring.

This skill outlines the implementation of advanced runtime defenses for production Large Language Model (LLM) applications. It covers using multiple industry-leading guardrail systems—including Llama Guard, NeMo Guardrails, and LLM Guard—to detect and block adversarial attacks such as prompt injection, jailbreaks, and toxic content. It provides a comprehensive defense-in-depth strategy for ensuring model safety and output integrity in real-world deployments.

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.

PyRIT (Python Risk Identification Tool) is an open-source automation framework designed for advanced, multi-turn adversarial red-teaming of conversational LLMs. It simulates real-world attack scenarios by orchestrating an attacker model and a scorer model in a feedback loop. Features include Crescendo (gradual escalation) and Tree-of-Attacks with Pruning (TAP) techniques to detect complex vulnerabilities like prompt injection and jailbreaks in stateful dialogues.

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 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.

This guide provides a structured approach to red-teaming Large Language Models (LLMs) to detect sensitive data leakage from system prompts. It simulates advanced attacks—including prompt injection, instruction override, and encoding tricks—to identify embedded secrets, API keys, and proprietary business logic. Essential for validating adherence to OWASP LLM07 and securing AI applications against data exfiltration.

A comprehensive red-teaming skill for assessing the security posture of Retrieval-Augmented Generation (RAG) pipelines. It systematically probes two critical injection surfaces: poisoned retrieved context (indirect prompt injection) and embedding manipulation. Use this skill to validate retrieval guardrails, ensure data leak prevention, and demonstrate vulnerabilities in LLM-powered knowledge assistants.