Skill UI

Provides expert guidance for GRPO training with TRL, covering dataset prep, reward engineering, structured outputs, and multi-objective alignment when preference pairs are scarce.

A comprehensive guide to building production-grade evaluation systems for Large Language Models (LLMs). This skill covers advanced methodologies, including Direct Scoring and Pairwise Comparison, alongside critical techniques for mitigating systematic biases (e.g., Position Bias, Length Bias). Learn how to select appropriate metrics and structure prompts to ensure reliable, consistent, and objective quality assessment across various AI applications.

This skill integrates LangChain 1.0 and LangGraph 1.0 into standard OpenTelemetry (OTEL) backends (e.g., Jaeger, Honeycomb, Tempo). It solves critical observability gaps by ensuring correct prompt content capture, propagating spans across complex subgraphs, and providing five LLM-specific Service Level Objectives (SLOs) like latency and cost. Ideal for multi-tenant, compliance-sensitive, or existing OTEL stack environments where LangSmith is unsuitable.

This skill provides a structured framework for capturing, organizing, and maintaining persistent facts about the user. By segmenting facts into topic-specific files (e.g., coding preferences, tech stack, goals), it creates a reliable, objective knowledge base. This structure is crucial for AI agents and applications to achieve highly personalized interactions and provide accurate context.

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 framework for creating and executing intelligent, goal-oriented action plans (GOAP). This system analyzes preconditions, optimizes costs, and provides adaptive replanning capabilities. It is designed for complex objectives that require multiple, dependent steps, ensuring robust task completion even when unexpected state changes occur.

A comprehensive guide for researchers submitting to major speech conferences (like INTERSPEECH). It details how to harden the reproducibility of speech results, covering crucial 'decay channels' such as corpus versions, text normalization rules (WER), objective metrics (EER), subjective protocols (MOS), and model stochasticity. The guide emphasizes pinning specific toolkits, commit hashes, configurations, and hardware details to ensure scientific rigor and verifiability.

This skill processes verbose, stream-of-consciousness voice dictations (from tools like macOS Dictation or Wispr Flow) and transforms them into highly structured, token-efficient prompts. It removes filler words, repetitions, and tangents, organizing the core intent into standard sections (Context, Objective, Constraints) suitable for advanced LLMs like Claude Code.

Multi-agent decision swarm that activates 20 legendary Tianya forum experts in parallel for a user objective. A coordinator assigns the task, each expert provides specialized insights, and a synthesis agent merges consensus and disagreements into a structured report with risks, recommendations and action plans.

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.