The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides functionality for monitoring and managing system processes and services. It includes critical safety protocols, such as a list of protected processes (init, systemd, sshd) that must never be terminated and a requirement to attempt graceful termination (SIGTERM) before force killing (SIGKILL).
[PROMPT_INJECTION]: An indirect prompt injection surface is present where the skill reads system process metadata. Ingestion points: Process data retrieved via list_processes and get_process_info in SKILL.md. Boundary markers: No explicit delimiters are defined for process metadata. Capability inventory: Command execution via os.kill and service management via systemctl in SKILL.md. Sanitization: The provided implementation snippets use Process IDs (PIDs) for targeting processes, which avoids shell-based command injection vulnerabilities during termination.

process-management