Skill UI

This skill integrates with the Runway ML platform, enabling developers to programmatically generate high-quality video and edit images using the official Python SDK. Use this when your workflow requires advanced creative AI features, such as converting text prompts into videos, or managing complex creative tasks within a CI/CD pipeline. The skill provides necessary patterns and error handling for successful API calls.

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.

A comprehensive reference architecture for building scalable, cost-effective, and robust AI services. This system manages the full AI lifecycle, including intelligent model routing (for quality/cost trade-offs), response caching (using Redis), asynchronous batch processing, and fine-tuning pipeline management via Together AI's OpenAI-compatible API. Ideal for enterprise applications requiring high availability across multiple open-source models.

This utility handles asynchronous webhook events from Together AI, enabling developers to build robust, event-driven Machine Learning pipelines. Instead of constantly polling the API for status updates (e.g., fine-tuning completion, batch job status, model deprecation), you can subscribe to these webhooks. It includes logic for signature verification, event routing, and idempotency, making it ideal for mission-critical AI workflows.

A comprehensive guide detailing advanced security threats in Machine Learning and AI systems. Covers model supply chain attacks (RCE via pickle, poisoning), adversarial examples (FGSM, PGD, physical-world), data privacy breaches, and LLM-specific vulnerabilities. Essential for building robust, secure AI pipelines.

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.

This skill autonomously optimizes artifacts like code or models using Hypothesis Tree Refinement. It runs iterative experiments with dev/test evaluators to prevent overfitting, coordinating subagents in isolated worktrees. Ideal for improving training recipes, agents, or pipelines without manual supervision.

A comprehensive tool for identifying poisoned training data, backdoored ML models, and supply-chain compromises. It leverages IBM's Adversarial Robustness Toolbox (ART) for activation clustering and spectral signature analysis, uses Cleanlab for label quality checks, and enforces integrity via weight-hash verification and safetensors. Essential for securing ML pipelines before training, deployment, or during supply chain audits.

This skill provides comprehensive techniques for security-testing Retrieval-Augmented Generation (RAG) pipelines. It probes for indirect prompt injection, which occurs when malicious instructions are embedded in retrieved documents, and tests for embedding-space manipulation. Using advanced tools like NVIDIA garak, Promptfoo, and PyRIT against vector stores (FAISS, Chroma, etc.), users can validate retrieval guardrails and secure CI/CD gates against adversarial inputs and data leakage.

Automates the creation and publication of polished, functional Gradio demos on Hugging Face Spaces for any user-provided LoRA model. This skill handles the complex workflow of designing the optimal UI, selecting the correct inference pipeline (e.g., text-to-image, video-to-video), and generating all necessary files (app.py, requirements.txt, README.md) to deploy a complete, ready-to-use AI playground.

This tool formats prompts for various Large Language Model (LLM) providers, including Anthropic, OpenAI, Gemini, and Ollama. It is essential for building robust Retrieval-Augmented Generation (RAG) pipelines. Users can format messages, create HNSW indexes from documents, route relevant context, and manage provider status, ensuring compatibility across different AI ecosystems.

Pattern for controlling LLM API costs while preserving quality. Combines complexity-based model routing, immutable budget tracking, retry logic limited to transient failures, and prompt caching. Use when building LLM API apps, processing batches with varying difficulty, or enforcing budget guardrails in production.