pufferlib

SUMMARY

What it does

This skill provides version-aware guidance for using PufferLib, a reinforcement learning library. It covers environment adaptation, vectorization, policy construction, training, evaluation, and safe checkpoint review. The skill distinguishes between two incompatible versions: the published 3.0.0 (Python/Gymnasium/PettingZoo emulation) and the current 4.0 source (native C environments). It includes safe defaults, installation and provenance checks, environment and vectorization workflows, policy and training guidance, logging considerations, and checkpoint inspection procedures. The skill is part of the Scientific Agent Skills collection by K-Dense and is licensed under MIT.