The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system-level commands to manage task schedules, including launchctl on macOS, systemctl on Linux, and schtasks.exe on Windows. It also provides the capability to execute arbitrary shell scripts when the task type is set to 'script'.
[DYNAMIC_EXECUTION]: The skill dynamically generates wrapper scripts (.sh and .ps1) by populating templates with user-supplied data. These generated scripts are then registered with the operating system's scheduler and executed at the specified intervals.
[PERSISTENCE]: The core functionality of the skill is to establish persistence by creating scheduled jobs that run independently of the agent session. This involves creating and loading LaunchAgents on macOS, systemd user timers on Linux, and scheduled tasks on Windows.
[INDIRECT_PROMPT_INJECTION]: There is a significant command injection risk due to insufficient sanitization of user-provided data. While the task 'target' is escaped for single-quoted strings, other parameters such as 'working_directory' and 'output_directory' are directly interpolated into double-quoted variable assignments in the bash and PowerShell templates. A user or an indirect injection source could provide a path containing subshells or command delimiters (e.g., $(command) or " ; command ; ") to execute unauthorized code with the privileges of the scheduled task.
Ingestion points: User-provided task configuration via the 'add' command, specifically the working directory and output directory paths.
Boundary markers: The templates lack robust delimiters or escaping for several interpolated variables.
Capability inventory: The skill possesses the ability to write to the file system, modify OS-level persistence mechanisms, and execute shell commands.
Sanitization: The skill implementes escaping for the task target using platform-specific rules, but fails to apply similar protections to directory paths used in the generated scripts.

manage