Skill UI

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 tool for searching the Model Context Protocol (MCP) across various GitHub locations, including PRs, issues, and discussions. It allows users to track the full lifecycle of protocol specifications, from initial proposals to final merged changes. Furthermore, it facilitates deep dives into pull request reviews and comments, enabling the extraction of key maintainer rationales and design intent for accurate technical understanding.

A comprehensive security skill designed to detect and mitigate indirect prompt injection attacks. It scans untrusted external content—including web pages, PDFs, and images—for hidden, obfuscated instructions before an LLM agent processes them. The pipeline combines content extraction (HTML, OCR), normalization, and dedicated ML models (e.g., LLM Guard, Prompt Guard 2) to ensure data safety.

This skill provides comprehensive methods and frameworks to detect sophisticated AI attacks, including model extraction, membership inference, and model inversion, which abuse inference APIs. It focuses on monitoring per-principal query volume, input distribution, and confidence exposure. Use it for securing public or partner inference APIs, or for conducting pre-deployment red-teaming to assess the model's extractability and overall security posture against MITRE ATLAS techniques.

Extract structured data from web pages using stored templates or one-shot DOM queries. Includes mandatory security gates for PII and prompt injection before data reaches the model. Supports reusable templates for recurring scraping tasks and browser automation workflows.

This skill deploys the OpenMed Model Context Protocol (MCP) server, exposing critical clinical NLP tools such as NER, PII extraction, and de-identification. It enables AI coding agents and chat clients to interact with OpenMed's capabilities via simple tool calls, running locally on-device to ensure user data privacy and security. Supports both stdio and Streamable HTTP deployment modes.

This skill performs the crucial ETL process of mapping OpenMed-extracted, terminology-coded clinical facts (conditions, drugs, measurements) into the OMOP Common Data Model (CDM) v5.4 tables. It is used downstream after entity extraction and source coding, transforming raw NLP findings into standardized, OMOP-conformant records for OHDSI analytics.

A comprehensive framework for conducting authorized security assessments on Large Language Models (LLMs) and AI agents. It covers critical vulnerabilities including advanced prompt injection (OWASP Top 10), tool abuse, RAG data poisoning, memory manipulation, and system prompt extraction. Ideal for red teaming, security auditing, and ensuring the robustness of AI systems in production environments.

Utilize zero-shot NER to extract custom entity types (e.g., Drug, Symptom, Device) from complex clinical or biomedical text without requiring any labeled training data or fine-tuning. OpenMed wraps GLiNER/GLiNER2, allowing users to define labels on the fly. The workflow involves building a model index and then running inference via CLI or Python API, making it ideal for rapid prototyping or unstable schemas.

Provides a hardened FastAPI endpoint for OpenMed's core functionalities: clinical Named Entity Recognition (NER), Personally Identifiable Information (PII) extraction, and de-identification. Designed for self-hosted microservices, it handles inference requests, model lifecycle management, and supports dynamic batching, ensuring secure, on-device processing of sensitive clinical data.

This skill provides one-shot chat completion using DeepSeek's deepseek-chat model. It is designed for non-reasoning tasks like summarization, extraction, or quick classification. The tool reads the DEEPSEEK_API_KEY environment variable and returns JSON output suitable for downstream automation. It supports various parameters such as temperature and max tokens, ensuring flexible interaction with the AI model for efficient text processing workflows.

A vendor-neutral, benchmark-grounded advisor that routes complex legal tasks to the optimal Large Language Model (LLM). It analyzes the specific legal vertical (e.g., contract drafting, legal research, info extraction) and user constraints (cost, speed, accuracy, jurisdiction) to recommend a primary model, a fallback, and crucial human verification steps. This tool provides decision support, not legal advice.