A blazingly fast DataFrame library built on Apache Arrow that slots in when pandas becomes a bottleneck but your data still fits in RAM. The expression-based API is cleaner than pandas chaining, lazy evaluation gives you automatic query optimization, and everything parallelizes by default. Think 1-100GB datasets, ETL pipelines, or anywhere you're tired of waiting for pandas to finish. The learning curve is real if you're coming from pandas since there's no index and the syntax is different, but the speed gains and cleaner code usually justify it. Use lazy mode with scan_csv instead of read_csv for anything sizeable, and stick to native expressions instead of Python functions to keep things fast.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill polars