The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently uses shell commands via subprocess.run and os.system for administrative tasks such as disk measurements, process monitoring, and executing cleanup operations. These are protected by a validation layer in apply-cleanup-selection.py and disk_safety.py that blacklists sensitive system roots (/System, /bin, etc.) and user credential directories.
[PERSISTENCE]: The skill installs a macOS LaunchAgent (com.local.mac-health-check.plist) to provide active background monitoring. This is a standard and transparent implementation for the skill's stated purpose of proactive health alerting.
[INDIRECT_PROMPT_INJECTION]: The skill ingests file metadata (filenames and paths) which could contain malicious instructions. It mitigates this risk by using a restricted 'metadata-only' runner for the AI agent, explicitly instructing it to ignore instructions embedded in filenames, and requiring a human reviewer to approve any cleanup plan through a separate interface.
[DYNAMIC_EXECUTION]: Runtime execution is used to apply user-selected cleanup commands and to open the local browser interface. The apply-cleanup-selection.py script restricts these commands to a specific whitelist (rm, brew, docker, etc.) and enforces strict path validation to prevent arbitrary execution.
[EXTERNAL_DOWNLOADS]: The skill uses ntfy.sh for push notifications and Homebrew to install required utilities like alerter and Mole. These are well-known services and are used exclusively for the skill's documented monitoring and maintenance functions.
[PRIVILEGE_ESCALATION]: While the skill uses sudo for cleaning certain system logs (Tier 6), the targets are restricted to a vetted allowlist of diagnostic and power logs, preventing unauthorized modification of critical system state.

maintaining-macos-health