The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill makes extensive use of the subprocess.run function to interact with system utilities like du, df, find, docker, brew, npm, and pip. These calls are implemented using list-style arguments rather than shell strings, effectively preventing shell command injection vulnerabilities.
[DATA_EXPOSURE] (SAFE): The scripts access various system paths, including ~/Library/Caches and ~/Library/Application Support, to calculate disk usage and identify orphaned files. This access is strictly limited to metadata (filenames and sizes) necessary for the cleanup functionality, and no data is transmitted to external servers.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data from the local file system in the form of directory and file names. While this presents a theoretical surface for indirect prompt injection if an agent blindly trusts the output strings, the scripts do not evaluate these names as code or instructions, and they provide interactive confirmation steps for destructive actions like deletion.

macos-cleaner