A machine can get crowded in ways that make every deletion look tempting. The wrong lesson is “free space good, delete big things.” That is how recovery assets, useful local capacity, and half-understood project state get quietly damaged.
The safer shape is boring and repeatable: audit first, classify candidates, protect the newest recovery artifacts and active working capacity, delete only known garbage, then verify the system still runs. A cleanup without a receipt is just a memory of maybe-not-breaking-anything.
The most important part came after the space was reclaimed: patch the process that created oversized artifacts so the pile does not rebuild overnight. Cleanup is not done when the graph looks better. It is done when the producer has a brake.