Skill UI

This skill analyzes the predictable ways language models lose performance as context length increases. It details core patterns, including 'lost-in-middle,' context poisoning, and distraction. Use this knowledge to diagnose unpredictable agent failures, design highly reliable systems for long document processing, and implement advanced context engineering strategies.

Learn how to monitor, analyze, and optimize the expenditure of Large Language Models (LLMs) using Langfuse's advanced analytics suite. This guide covers techniques like automatic token tracking, programmatic cost reporting via the Metrics API, and implementing smart model routing to automatically select the most cost-effective model based on task complexity and input length. Essential for any developer managing AI budgets and ensuring operational efficiency.

Extends transformer context windows using RoPE, YaRN, ALiBi, and position interpolation so long documents (32k-128k+ tokens) can be processed, pre-trained models lengthened, and efficient positional encodings or extrapolation strategies deployed.

Use this tool to query, filter, and compare over 400 available AI models from OpenRouter. It allows users to check pricing, context lengths, model capabilities (e.g., tool calling), and select the best model based on specific needs, ensuring optimal performance and cost efficiency in AI development.

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.

RecallMax dramatically enhances AI agent memory capabilities by cleanly injecting massive external contexts (500K-1M tokens) without hallucination drift. It automatically summarizes lengthy conversations, preserving tone, intent, and key facts. Furthermore, it compresses multi-turn interaction histories into high-density token sequences, ensuring agents maintain perfect recall across complex and long-running sessions.

Provides an expert methodology to significantly reduce AI API spending (40-80%) without sacrificing user quality. The guide covers cost auditing, optimizing existing systems, and designing cost-efficient architectures using techniques like model routing, prompt caching, output length control, and prompt compression.

This guide details how to implement robust content safety guardrails for applications powered by Claude. It covers essential techniques such as system prompt hardening, input validation to detect injection patterns and length limits, and output validation to prevent system prompt leakage. It ensures compliance with Anthropic's policies, making the AI integration secure and reliable.

This comprehensive skill provides multi-layered guardrails for integrating with LLMs like Claude. It enforces security and compliance by implementing input validation (PII detection, length checks), system prompt restriction (scope control, safety rules), output filtering (preventing sensitive data leakage), cost governance, and model access policies. Ideal for building secure, robust, and governed AI applications.

The `hf-mem` skill estimates the necessary VRAM or system memory required to load Safetensors or GGUF model weights for inference directly from the Hugging Face Hub. It uses efficient HTTP Range requests, avoiding local downloads. Users can determine if a model fits on their GPU or instance, including optional estimations for the KV cache and setting limits like maximum sequence length.

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%).

A comprehensive scoring tool to validate localized or translated content before publishing. It calculates a composite score by analyzing technical accuracy (e.g., formatting preservation, placeholder integrity, length ratios), overall content quality, brand voice consistency, and adherence to target market compliance rules. Use this to replace subjective checks with structured, repeatable quality assurance.