Hi #C0AGE0SJPU1 👋

I'm Makhmud, a 2nd year CS student at UAE University with a background in Computer Vision and ML (ResNet, YOLO, OpenCV).

I've been exploring the Jupyter-Native Metaflow project. I installed Metaflow, ran flows locally, and tried NBRunner in a notebook. I intentionally split a FlowSpec across multiple cells to understand the failure — got

ModuleNotFoundError: The cell doesn't contain any class that inherits from 'FlowSpec'

. I traced it back to

utils.py

where

format_flowfile()

uses

ast.parse()

on the cell source looking for a complete class definition which obviously fails when the class is split.

My understanding of the solution direction: a

%%step

cell magic that intercepts cell content, collects individual step method definitions, and assembles them into a complete FlowSpec before passing to NBRunner — similar to how Kale handles cell-tagging for Kubeflow pipelines.

Before writing my proposal I had a few questions for @hundreds-rainbow-67050:

Is the

%%step

magic approach the preferred direction, or has the team considered alternatives?

Should the extension live as a standalone package or inside the main Metaflow repo?

Is there a good first issue or area of the codebase you'd recommend I explore before submitting?

GitHub: https://github.com/makhmudlp