The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/scan.py script executes system commands such as du, find, sw_vers, and uname to calculate directory sizes and gather system architecture information.
[COMMAND_EXECUTION]: The scripts/server.py script uses osascript on macOS to move files to the trash via the Finder and utilizes ctypes to call SHFileOperationW on Windows for similar operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes directory paths and file names collected during the scanning phase to provide analysis and recommendations.
Ingestion points: Untrusted data (file names and paths) enters the agent context through the reading of /tmp/storage_scan.json.
Boundary markers: None explicitly present in the instructions.
Capability inventory: The skill can delete files, move them to the trash, and open directories in the file manager using server.py.
Sanitization: A comprehensive path validation system in server.py checks all targets against a list of over 20 critical system paths and prefixes, rejects relative paths or traversal attempts, and resolves symlinks to ensure they do not point to protected areas.
[SAFE]: The local server used for managing deletions implements multiple security layers: it binds only to 127.0.0.1, requires a 24-character random session token for all requests, verifies the Host header to block DNS-rebinding attacks, and maintains an operations log at ~/Library/Logs/disk-cleaner/operations.log.

disk-cleaner