Features arise out of the composition of fundamental units of the system, they're not normally first class units themselves. Can you give an example?

For example using nested .gitignore files vs using one root .gitignore file. I guess this principle is related to this one:

Imagine the code as a graph with nodes and edges. The nodes should be grouped in a way that when you display the graph with grouped nodes, you see few edges between groups. Removing a group means that you need to cut maybe 3 edges, not 30. I.e. you don't want something where every component has a line to every other component.

Also when working on a feature - modifying / adding / removing, ideally you want to only look at an isolated group, with minimal links to the rest of the code.