The Agent Skills Directory

[COMMAND_EXECUTION]: The script devops.py utilizes subprocess.run(shell=True) to execute system utilities including docker, ps, ss, and netstat. Executing commands through a shell increases the attack surface for command injection if input sanitization is bypassed.
[DATA_EXFILTRATION]: The skill is designed to read sensitive system log files such as /var/log/auth.log, /var/log/syslog, /var/log/messages, and /var/log/kern.log. These files frequently contain sensitive information regarding user logins, authentication attempts, and internal system events which could be exposed to the agent or exfiltrated via the built-in network capabilities.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface. It ingests untrusted data from external sources, specifically system logs and HTTP response bodies via urllib.request.urlopen. Content from these attacker-controllable sources is displayed to the agent without boundary markers or sanitization, potentially allowing malicious instructions embedded in logs or web responses to influence agent behavior.

agentic-devops