Zarr Python

Zarr-Python 3 gives you chunked, compressed N-dimensional arrays that live on disk, in memory, or in cloud object storage. If you're juggling multi-terabyte scientific datasets with NumPy, Dask, or Xarray, this is how you avoid loading everything into RAM at once. The chunking strategy matters more than you'd think: a (200, 200, 200) array can be 65× faster to read if you align chunks with your access pattern. Version 3.2+ defaults to the new Zarr format, drops Python 3.11 support, and plays nicely with S3 and GCS through fsspec. The skill covers compression codecs, sharding for when you've got millions of tiny chunks, and the difference between vindex and oindex that you'll inevitably need.

Install to Claude Code

npx -y skills add k-dense-ai/scientific-agent-skills --skill zarr-python --agent claude-code