Skill UI

An expert guide for deploying and optimizing Large Language Models (LLMs) entirely on local hardware. Mastering the entire local AI ecosystem, including inference engines like Ollama, vLLM, and llama.cpp. Provides deep knowledge on quantization formats (GGUF, EXL2, AWQ), VRAM calculation, and ensuring privacy-first, offline deployment for state-of-the-art models.

Fal.ai Media Studio via MCP unifies image, video, and audio generation using Nano Banana, Seedance, Kling, Veo 3, CSM-1B, and ThinkSound models, enabling prompt-driven drafts, high-fidelity renders, and multimodal content for creative workflows.

This skill automates the multi-round research review process. It iteratively reviews provided work using advanced AI models (Codex MCP), identifying critical weaknesses, implementing fixes, and re-reviewing until the content achieves a positive assessment or the maximum round limit is reached. Ideal for refining complex academic papers, technical designs, or large codebases that require rigorous, iterative quality assurance.

Deploys AI and NLP-powered detection to spot business email compromise by analyzing writing style, behavioral context, and intent, then integrates baselines, transformer models, and response workflows to surface impersonation attempts that rule-based filters miss.

PydanticAI is a robust Python framework that integrates Pydantic's type-checking and validation capabilities into Large Language Models (LLMs). It allows developers to build production-ready AI agents that guarantee structured, validated outputs, moving beyond simple raw strings. Key features include supported tool calling, dependency injection for testability, and multi-model compatibility across major providers like OpenAI, Anthropic, and Gemini. Use this when reliability and correct data structure are paramount.

This skill systematically probes an AI model's core behavioral patterns across six critical dimensions, including reasoning, refusal, formatting, and grounding. It generates a comprehensive, styled HTML report with radar charts and actionable insights, allowing users to understand a model's true capabilities, compare different models, or perform compliance and red-team assessments without needing an external API key.

AuraKit converts a single /aura command into a secure full-stack Claude pipeline with 34 modes, OWASP SEC-01~15 safeguards, runtime hooks, tiered models, token savings, and multilingual reviewers.

This skill is designed for performing local, on-premise evaluations and benchmarking of models hosted on the Hugging Face Hub. It facilitates running specialized evaluation frameworks like `inspect-ai` and `lighteval` against local hardware. Users can select optimal inference backends, including `vllm` for high throughput, or fall back to Hugging Face Transformers or `accelerate`, ensuring comprehensive smoke testing and task selection for LLMs. It focuses purely on the local execution process, separate from remote jobs or publishing workflows.

A comprehensive tool and API wrapper for accessing, parsing, and analyzing research papers published on Hugging Face and arXiv. Users can input various URLs or IDs (e.g., arXiv ID) to retrieve structured metadata, full markdown content, and automatically link associated resources such as models, datasets, and application spaces. Ideal for researchers and developers needing structured access to cutting-edge AI knowledge.

Fine-tune or train object detection, classification, and SAM/SAM2 segmentation models on Hugging Face Jobs, leveraging managed GPUs, automatic dataset checks, and Hub persistence without local GPU setup.

Run Hugging Face models directly in JavaScript/TypeScript for text, vision, and speech tasks without backend servers, using Transformers.js in Node.js or modern browsers with WebGPU, quantization, and pipeline abstractions for efficient inference.

Provides command-line recipes for hitting DeerFlow HTTP endpoints to start threads, stream runs, manage memory/models/skills/agents, and upload files for in-depth research tasks.