The agent spawns an isolated Pi subprocess that investigates independently and returns a summary. The parent session's context grows by the summary size only, not by all the files the subagent read.

Abstract: Jupyter notebooks have become central in data science, integrating code, text and output in a flexible environment. With the rise of machine learning (ML), notebooks are increasingly used ...

The CLAUDE.md file at the root of this repo is the core -- it's the configuration that makes everything work.