Skill UI

An advanced AI tool utilizing Meta's Segment Anything 2 (SAM2) model for interactive, pixel-perfect object segmentation. Users can click on any object within a frame or video to automatically generate precise masks. Key features include point/box prompting, video tracking across multiple frames, and seamless integration into Annotation Studio workflows, making it ideal for large-scale data labeling and video analysis pipelines.

This debug bundle provides comprehensive guidance for working with Together AI's OpenAI-compatible API. It supports critical LLM lifecycle tasks, including inference, fine-tuning, and model deployment for over 100 open-source models (e.g., Llama, Mixtral). It also includes detailed troubleshooting steps and solutions for common API errors (like 401 Unauthorized or 429 Rate limit), ensuring smooth development.

This guide provides a comprehensive tutorial on running various AI inference tasks using the Together AI API, which is compatible with the OpenAI standard. It demonstrates key functionalities including chat completions, streaming responses, image generation, and creating embeddings. This is ideal for developers who need to test open-source models, compare performance across different LLMs (like Llama or Mixtral), or integrate powerful generative AI features into their applications using Python or Node.js.

This guide provides comprehensive instructions for setting up and authenticating the Together AI SDK. It covers installation using both Python and Node.js, configuring the necessary API key, and verifying connectivity to utilize the OpenAI-compatible API for powerful open-source model inference and fine-tuning tasks. Ideal for developers starting with Together AI.

Provides comprehensive guidance for performance tuning, inference, and fine-tuning using the Together AI platform. This skill supports working with Together AI's OpenAI-compatible API, covering model deployment, utilizing various open-source models (like Llama, Mixtral), and implementing best practices for efficient and cost-effective batch inference.

A comprehensive guide to implementing production-ready patterns for interacting with the Together AI API. This includes a singleton client for seamless OpenAI-compatible integration, robust error wrappers with retry logic, and a fluent request builder. It covers best practices for LLM inference, model setup, and handling diverse response types (chat, embedding, fine-tuning).

This skill provides high-performance, real-time object detection using the Intel OpenVINO runtime. It operates via a Docker container, ensuring cross-platform compatibility. Supports detection across various Intel accelerators (NCS2, iGPU, Arc GPU) and CPUs. Users can customize parameters like confidence threshold, target classes, processing FPS, and model precision to optimize performance for specific hardware and accuracy needs.

This tool functions as a Command Line Interface (CLI) alias for implementing the specific 'pua:mama' protocol within advanced coding models like Codex or Claude Code. It loads and enforces a structured 'mama skill/protocol,' maintaining rigorous verification standards while adjusting the visible narrative style to a specific 'Chinese-mom nagging mode.' It manages configuration by interacting with `~/.pua/config.json`, ensuring existing user fields are preserved.

This skill automatically enriches research wiki pages that are left as bare scaffolds after initial ingestion. It reads multiple external sources (like arXiv/alphaxiv) and fills in 10 structured `_TODO._` sections (e.g., Problem/Gap, Key Results, Open Questions) with detailed prose summaries. Use this skill to transform incomplete research scaffolds into a comprehensive, structured knowledge base.

This skill is designed to prevent AI coding assistants from exhibiting sycophantic agreement bias. When active, it forces a structured, critical assessment of user claims. The process involves extracting the core claim, assessing it independently with evidence (for/against), and concluding solely based on that assessment, ensuring objective and evidence-based responses.

A comprehensive tool for generating high-quality audio content, including instrumental background music (BGM), jingles, ambient beds, and full songs with sung vocals. It supports complex creative parameters such as mood, tempo, vocal style, and target language, ensuring original and copyright-safe output through specialized generation models.

This skill provides a comprehensive guide for building scalable Retrieval-Augmented Generation (RAG) pipelines and persistent agent memory using Pinecone as the vector database. It covers index setup, embedding large corpuses, implementing dense and hybrid semantic search, and advanced metadata filtering for real-world production systems. Use this for building searchable knowledge bases and context-aware AI agents.