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.

Checks AI agent implementations against the OWASP Agentic Security Initiative Top 10 risks, producing findings on controls, vulnerabilities, and recommendations for prompt injection, tool governance, escalation, logging, identity, policy, supply chain, and behavioral monitoring before production or audit.

Conducts comprehensive AI prompt safety reviews that analyze bias, misinformation, injection, and privacy risks while delivering effectiveness evaluations, security guidance, and best-practice recommendations for prompt engineers.

Provides comprehensive security guidelines for integrating Anthropic Claude API. Covers essential best practices including secure API key management (rotation, isolation), defending against prompt injection attacks by separating system instructions from user input, robust input sanitization, and output validation to prevent accidental data leakage. Essential for building production-grade, secure AI applications.

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 guide details how to implement robust content safety guardrails for applications powered by Claude. It covers essential techniques such as system prompt hardening, input validation to detect injection patterns and length limits, and output validation to prevent system prompt leakage. It ensures compliance with Anthropic's policies, making the AI integration secure and reliable.

This guide details essential best practices for securing applications integrated with Anthropic's Claude API. It covers critical security measures including secure API key management (server-side only), robust input validation to prevent cost attacks, defense mechanisms against prompt injection, and implementing per-user rate limiting. Essential reading for maintaining data privacy and system integrity in production environments.

A comprehensive evaluation suite designed to benchmark the performance of Large Language Models (LLM) and Vision-Language Models (VLM) specifically for home security applications. It covers 143 diverse tests across 16 critical domains, including tool use, prompt injection resistance, contextual reasoning, alert triage, and detailed scene analysis, ensuring robustness for smart home AI systems.

This skill defines canonical, ordered middleware patterns for LangChain and LangGraph agents. It addresses critical production-level LLM deployment issues, including PII leakage across tenants, token budget overconsumption, prompt injection defense (Guardrails), and cache key corruption. Use this to build secure, scalable, and cost-aware AI applications.

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.

This comprehensive, zero-trust security auditor scans third-party AI agent skills before installation. It performs a rigorous 6-phase review process, analyzing surface patterns, script integrity, permissions, and supply chain risks. It is designed to detect malicious payloads, credential theft, instruction overrides, and prompt injection attacks, ensuring secure AI development and deployment.

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.