lets say I've created a custom skill that runs a transcription on a dropped audio or video file, identifying speakers a prepending the speaker title to each segment of continuously spoken text. If I could nest skill cards, I could run sentiment analysis on each segment and generate a skill card representing that (as a timeline card) while also cross linking any other areas of the file which had the same sentiment rating (using the provided 'appears').

This is just one example, but the flexibility would would lend its self to many more complicated implementations, possibly negating the need for custom front end solutions in some cases.

Each json entry has a type. Along with text and image types we could have a 'card' type to indicate skill card nesting