Skill UI

PufferLib is a high-performance library designed for advanced reinforcement learning simulations and training. It excels in parallel environment simulation, offering optimized vectorization and native support for multi-agent systems. Utilizing algorithms like PuffeRL (optimized PPO+LSTM), it achieves millions of steps per second, significantly accelerating research and deployment across single-agent and multi-agent environments.

A detailed guide for authors submitting theoretical machine learning papers (like COLT) regarding the appropriate inclusion and design of numerical simulations. It emphasizes that numerics should only visualize proved bounds, phase transitions, or separations, maintaining the focus on theoretical proofs rather than empirical data. Provides rules for plotting, reporting variability, and maintaining academic honesty.

This skill details the PluginEval quality methodology, covering scoring dimensions, rubrics, and formulas. It explains evaluation layers including static linting, LLM judging, and Monte Carlo simulations. Use it to understand quality measurement, interpret scores, improve triggering accuracy, set thresholds, or explain badges to partners.