The Agent Skills Directory

[PERSISTENCE]: The skill is explicitly designed to maintain execution across sessions by registering background tasks with native operating system schedulers.
Evidence: The scripts/setup-crontab.sh script modifies the Linux user's crontab using the crontab - command.
Evidence: The scripts/setup-launchd.sh script creates and loads macOS LaunchAgents by writing .plist files to ~/Library/LaunchAgents/ and using launchctl load.
Evidence: The scripts/setup-schtasks.ps1 script creates scheduled tasks on Windows using the schtasks.exe command-line utility.
[DYNAMIC_EXECUTION]: The skill dynamically assembles and writes executable scripts or configuration files based on parameters passed to it at runtime.
Evidence: scripts/setup-launchd.sh generates a complete XML plist file with shell commands embedded in the ProgramArguments key.
Evidence: scripts/setup-crontab.sh constructs a command string that includes directory changes and output redirection before passing it to the cron daemon.
Evidence: scripts/setup-schtasks.ps1 builds a PowerShell command array to create a Task Scheduler entry with a custom shell command string.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it ingests untrusted commands for persistent background execution.
Ingestion points: User-provided commands and scheduling parameters are accepted via the /scheduler:schedule-add command defined in SKILL.md.
Boundary markers: The skill does not implement boundary markers or instructions to the agent to ignore potentially malicious content within the commands being scheduled.
Capability inventory: The skill has the capability to write to sensitive directory paths (like ~/Library/LaunchAgents) and execute arbitrary shell commands via system schedulers across three different operating systems.
Sanitization: While scripts/setup-launchd.sh performs basic XML entity escaping, there is no validation or sanitization of the actual command logic to prevent command injection or the scheduling of malicious tasks.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands through the use of system-level tools.
Evidence: SKILL.md explicitly allows the use of Bash(crontab:*), Bash(launchctl:*), and Bash(schtasks:*) to manage background processes.

scheduler