Skill UI

A comprehensive tool for detecting and flagging Personally Identifiable Information (PII) within text, code, and configuration files. It is essential to use this before committing code, logging data, or generating model responses to prevent the accidental leakage of sensitive data such as emails, passwords, SSNs, or API keys.

A comprehensive template providing full-stack development guidelines and best practices. This guide details the architecture (Next.js, FastAPI), file structure, common coding patterns (API response handling, state management hooks), and structured methods for integrating AI models (e.g., Claude). Use this template when starting a complex, multi-service application project.

This tool is designed to help optimize and refine prompts used with Large Language Models (LLMs). By structuring inputs, it ensures that the instructions given to AI systems are clear, comprehensive, and contextually rich, resulting in significantly higher quality and relevance of the generated output. Useful for developers, writers, and researchers.

This skill provides guidance on implementing common architectural patterns and best practices within the PyTorch framework. It is designed for developers building sophisticated deep learning models, helping to improve structure, efficiency, and code maintainability when working with neural networks.

This skill provides a structured framework for designing and implementing composable recommendation, ranking, and feed pipelines. It models the industry-standard six-stage pattern: Source, Hydrator, Filter, Scorer, Selector, and SideEffect. Use it for any system that selects "the top K items for a (user, context)", such as social feeds, ad ranking, RAG rerankers, or content recommendation engines, ensuring a robust and scalable architecture.

This skill is designed for complex, iterative decision-making processes requiring deep computation, such as stochastic optimization, high-dimensional search, or repeated model rollouts. It mandates the creation of a structured 'Ledger' to rigorously record every trial, prior winner, fresh information, and decision mark. It ensures that conclusions are based on durable evidence, managing risk through explicit promotion gates and preventing premature certainty.

A practical decision framework for determining whether to use regular expressions or Large Language Models (LLMs) when parsing structured text (e.g., forms, quizzes, invoices). It recommends starting with deterministic, low-cost regex for 95-98% of structured data, and only invoking expensive LLM calls for identifying and validating ambiguous edge cases, optimizing both cost and accuracy.

RouterBase acts as a unified, OpenAI-compatible API gateway, allowing applications to route requests across diverse large language models (LLMs) like GPT, Claude, and Gemini. It helps developers migrate existing OpenAI SDK calls, implement sophisticated primary/fallback model selection based on cost, latency, and quality, and manage multimodal payloads (image, audio, video) from a single, standardized endpoint.

This tool performs comprehensive safety scanning on untrusted inputs, such as user submissions, API payloads, or webhook data. It detects sophisticated threats like prompt injection, jailbreak attempts, and adversarial attacks using advanced AI defense mechanisms, ensuring data integrity and security before model processing or execution.

This pattern demonstrates how to build a robust, thread-safe data persistence layer in Swift using the 'actor' model. It combines fast in-memory caching with reliable file-backed storage, automatically eliminating data races and the need for manual locks. Ideal for offline-first applications and managing shared mutable state in modern Swift concurrency.

A guide to the Swift 6.2 concurrency model, which mandates default single-threaded execution to eliminate accidental data races. Learn how to use MainActor for isolated consistency and utilize the @concurrent attribute for explicitly offloading CPU-intensive tasks to background threads, ensuring safe and predictable application architecture.

This skill intercepts the AI response flow to allow users to explicitly control the desired depth, length, and token usage of the generated answer before the model begins writing. By analyzing the prompt's complexity and the user's intent (e.g., 'short version', 'detailed answer'), it estimates the appropriate token range and prompts the user to select a granularity level (25%, 50%, 75%, or 100%).