The Agent Skills Directory

[COMMAND_EXECUTION]: All provided scripts (modify-mail.sh, search-maillogs.sh, set-spam.sh, test-imap.sh, test-pop3.sh, test-smtp.sh) execute system commands using sudo. This grants the AI agent root-level administrative access to the underlying operating system and the Virtualmin hosting environment.
[CREDENTIALS_UNSAFE]: The scripts test-imap.sh and test-pop3.sh accept a --pass parameter to test mailbox connectivity. Passing passwords as command-line arguments is insecure because they are visible in the system's process list (e.g., via the ps command) and are typically recorded in the shell's command history file (.bash_history).
[DATA_EXFILTRATION]: The search-maillogs.sh script provides a mechanism to query Postfix mail logs. These logs contain sensitive metadata, including sender and recipient email addresses, timestamps, and connecting IP addresses, which could be exploited to harvest communication patterns or user information.
[PROMPT_INJECTION]: The search-maillogs.sh script acts as an ingestion point for untrusted data from system logs. If an attacker sends an email with malicious instructions in the headers (like the Subject or From fields), those instructions will appear in the log output.
Ingestion points: System mail logs via virtualmin search-maillogs in search-maillogs.sh.
Boundary markers: None. The output is processed directly by the agent.
Capability inventory: The skill has broad sudo access to modify system settings and manage mail services across all files in the scripts/ directory.
Sanitization: None. The script passes raw log data back to the agent without filtering or escaping.

virtualmin-mail