The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts/new-launchagent.sh, which utilizes launchctl (the native macOS service manager) to load and unload background tasks. The script performs validation on inputs such as labels and paths to ensure the generated plists are valid and safe to execute.
[PRIVILEGE_ESCALATION]: The new-launchagent.sh script uses sudo for operations that require root privileges, specifically when installing agents to the system-wide /Library/LaunchDaemons directory. This is standard macOS administrative behavior and is explicitly gated by a --system flag.
[EXTERNAL_DOWNLOADS]: An example auditing fixture script (broken-heal.sh) contains a health check using curl to googleapis.com, which is a well-known and trusted domain for connectivity testing. No unauthorized or suspicious external downloads were detected.
[PERSISTENCE]: While the skill is dedicated to managing launchd jobs (a form of persistence), its purpose is to provide a structured and auditable way for users to manage their own background processes. It includes best practices for 'quiet watchdogs' that stand down when their remediation goals are unachievable, reducing the risk of unintended persistence loops.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risks of automated remediation and provides remediation guidance (e.g., Pattern 1: Premise-state self-check) to ensure that watchdogs do not act on stale or untrusted signals, serving as a defensive measure against unintended behavior.

macos-watchdog