The Agent Skills Directory

[SAFE]: The skill implements a read-only audit mechanism using standard Python libraries to measure directory sizes and report free disk space.
[SAFE]: The instructions include robust 'Hard safety rules' that explicitly prohibit destructive actions (deletions), privilege escalation (sudo), and the inspection of sensitive data such as browser profiles, credentials, or shell history.
[SAFE]: The scripts/scan.py script uses lstat and scandir to safely collect filesystem metadata while respecting filesystem boundaries (st_dev) and user ownership (st_uid), preventing unauthorized data exposure or performance issues.
[SAFE]: The script follows the 'least privilege' principle by identifying cleanup candidates but leaving the actual execution to the user, enforced by a separate confirmation gate.

janitor