Skill UI

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 skill structures a highly controlled, bounded development life cycle for AI agents. It mandates a structured three-phase workflow: Planning (Spec), Implementation (Build), and Verification (Review). It is essential for building complex features from scratch, especially when dealing with ambiguous requirements, high risk, or needing explicit human approval gates to ensure the resulting code meets all defined specifications.

This comprehensive skill provides an index of official blueprints and implementation guides for building full-stack, AI-powered applications using Weaviate. It covers advanced patterns such as Retrieval-Augmented Generation (RAG), agentic workflows, data exploration, and multimodal document search, guiding users through best practices for setup, clients, and frontends.

Tamarind Bio is a cloud platform providing managed GPU access to hundreds of open-source computational biology tools. It facilitates complex tasks such as protein structure prediction (AlphaFold), de novo protein design, antibody engineering, protein-ligand docking, and molecular dynamics simulations. Users submit jobs via a unified API and receive predicted structures and biophysical scores without needing local high-performance computing resources.

This skill provides utilities to manage and drain the deliberation queue (brain_frames) of the ACC system. It allows users to list waiting frames, retrieve necessary context, perform deliberation on open tasks, and submit continuations using `acc_act(runtime="continue")`. It is essential for checkpointed work saved by headless runs, ensuring continuous task resolution and calibration against predicted outcomes in complex AI workflows.

This skill provides a hybrid, persistent, and searchable knowledge memory system specifically designed for enhancing AI agents' capabilities. It allows developers to record, retrieve, and search key knowledge, architectural decisions, and operational memories automatically. Use it to build agents with long-term context and reliable knowledge management, ensuring findings are synced back into project documentation.

The Anti-Deception Harness detects manipulative user prompts, such as manufactured urgency, pressure to validate, or undue reliance on authority. It calls an external tool to provide a structured scaffold detailing integrity procedures and counter-evidence, ensuring responses are honest and robust, preventing compromised or sycophantic answers. Ideal for high-stakes decision support.

This skill guides developers on implementing Conversational Language Understanding (CLU) using the official azure-ai-language-conversations Python SDK. It demonstrates how to analyze conversation intent and entities, build advanced NLP features, and integrate language understanding into applications using ConversationAnalysisClient. Best practices emphasize using DefaultAzureCredential and context managers for robust, production-ready code.

The `hf-mem` skill estimates the necessary VRAM or system memory required to load Safetensors or GGUF model weights for inference directly from the Hugging Face Hub. It uses efficient HTTP Range requests, avoiding local downloads. Users can determine if a model fits on their GPU or instance, including optional estimations for the KV cache and setting limits like maximum sequence length.

This professional tool helps developers and researchers identify the optimal AI model for a specific task, such as coding, reasoning, or image recognition. It queries official HuggingFace leaderboards, filters models based on hardware constraints (e.g., available VRAM or RAM), and provides a detailed comparison table including benchmark scores, parameter size, and licensing information.