Skill UI

This skill provides expert capabilities for training and fine-tuning transformer language models using the TRL (Transformers Reinforcement Learning) library. It supports state-of-the-art post-training methods, including Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), Group Relative Policy Optimization (GRPO), and Reward Model training. Use this skill to align and customize foundation models for advanced tasks via CLI commands.

Specialized skill to rewrite and humanize AI-generated natural language files (like notes or documentation). It removes common 'AI-isms' such as sycophantic openings, stock vocabulary, over-explaining, and excessive signposting. Users can choose between a fast, deterministic regex cleanup or a deep LLM rewrite (default mode) using Claude to enhance burstiness and match a natural human voice, all while strictly preserving code blocks, URLs, and headings.

This guide provides a comprehensive reference on Agent Skills, a standardized and powerful method for providing structured context, workflows, and tools to Large Language Models (LLMs). It details the anatomy of the SKILL.md file, the three-stage progressive disclosure loading process, and best practices for creating both capability (tool-based) and process (workflow-based) primitives. Use this resource when designing, reviewing, or debugging agent-driven applications to ensure reliable and context-efficient performance.

This skill executes the DeepSWE benchmark, a comprehensive coding-agent evaluation suite. It uses OpenRouter and mini-swe-agent to score various Large Language Models (LLMs) on 113 tasks, providing reproducible results for model comparison, development assessment, and research benchmarking. Use this when a standardized, deep evaluation of LLM coding capabilities is required.

Diagnoses full-stack failure modes in agent and LLM applications, including wrapper regressions, memory contamination, tool discipline failures, hidden repair loops, and rendering corruption. Produces severity-ranked findings and code-first fixes for production agent systems.

A specialized tool for generating long-form content, including guides, blogs, and tutorials, that maintains a unique, human-like brand voice. It ensures structural integrity and factual accuracy, preventing the generic, boilerplate feel often associated with raw LLM output. Ideal for content teams needing sophisticated, reliable narrative development.

A comprehensive guide covering patterns and architectures for automating code execution using LLMs. It spans simple sequential pipelines (e.g., daily dev scripts) to advanced, RFC-driven multi-agent Directed Acyclic Graph (DAG) orchestration. Ideal for building CI/CD style continuous development pipelines, managing complex state, and automating quality gates without human intervention.

This tool enables complex web interactions using plain natural language commands. Instead of relying on unreliable selectors, it utilizes a powerful page-agent and LLM loop to interpret user intent, performing actions like filling forms, clicking buttons, and extracting data from dynamic websites. Ideal for testing and automation in pages with ambiguous or changing markup.

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.

This tool audits the token consumption across all components (agents, skills, rules, MCP servers) within a Claude Code session. It analyzes token overhead, identifies bloat (e.g., overly long descriptions, redundant components), and generates a detailed context budget report. Use it to maintain optimal context window usage, improve performance, and prevent context saturation.

A comprehensive pattern for building robust and cost-efficient LLM application pipelines. This guide details combining model routing based on task complexity, immutable cost tracking, limited retry logic, and prompt caching to manage API expenditures while maintaining high quality. Essential for production systems handling variable-complexity, high-volume tasks.

This skill runs comprehensive benchmarks against structural and adversarial corpora. It measures the performance and cost efficiency of a booster model compared to leading LLM baselines (Gemini, Anthropic). Results are persisted in verifiable JSON reports, crucial for auditing claims and verifying booster win rates before publishing.