Personal systems
Ten years of iCloud, sorted by an agent that asks before it deletes
Duplicates, screenshots, downloads from 2016, three copies of every tax return. An agent walks the whole library, proposes what to merge, move or bin, and waits for a yes.
Cloud storage grows until it costs money, and then you pay rather than sort it. I paid for years. Then I built the thing that sorts it.
How it works
The agent walks the library and builds an index: what every file is, where it lives, how big it is, and whether there is another copy of it somewhere else. Duplicates are matched on content, not name, which is how it found the same PDF saved four times under four names.
From the index it produces a review queue, not a deletion list. Every proposed action comes with a reason. “These 340 screenshots are older than two years and none has been opened since. Archive?” “These two folders contain the same photos. Keep the one with the original dates?” I approve in batches from Telegram. Nothing is deleted without a yes, and everything deleted goes to a holding folder for thirty days first.
The taxonomy
The second job was a folder structure that a system can file into: Finance, Home, Health, Work, Projects, Archive, with a rule for each. New files that land in Downloads get proposed a home. The taxonomy is boring on purpose. Clever structures are the reason nobody can find anything.
The monthly report
On the first of the month I get a short report: storage used, what was archived, what is waiting for review, and the five largest new files. It takes a minute to read and it means the mess never comes back.
What broke
The first duplicate pass matched on file size and name and produced a list that would have deleted originals and kept edited copies. It now hashes content and keeps the earliest file by creation date unless I say otherwise. The lesson generalises: a deletion system has to be pessimistic by design.