vps-disaster-recovery
Fail
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill explicitly references sensitive private SSH key paths, which are used to authenticate with remote servers.
- Evidence:
~/.ssh/id_siliconvalleyand~/.ssh/id_vultrinSKILL.md. - [COMMAND_EXECUTION]: The skill provides scripts that execute administrative commands on remote servers via SSH, including package installation and system service management.
- Evidence: Use of
ssh,apt-get install, andsystemctlwithin the reconstruction scripts. - [PRIVILEGE_ESCALATION]: The instructions utilize
sudoto modify restricted system configuration files and manage global services. - Evidence: Modification of
/etc/ssh/sshd_configto disable password authentication and modification of/etc/tinyproxy/tinyproxy.conf. - [PERSISTENCE]: The skill implements a persistence mechanism by modifying the system's
crontabto ensure an SSH tunnel is automatically restarted every five minutes. - Evidence:
crontab -l 2>/dev/null | ... >> /tmp/cron_new; crontab /tmp/cron_newsequence inSKILL.md. - [INDIRECT_PROMPT_INJECTION]: The reconstruction script defines a variable
NEW_IPand interpolates it directly into a shell command, which represents a command injection surface if the agent accepts untrusted input for the IP address. - Evidence:
ssh -o StrictHostKeyChecking=no ... ubuntu@$NEW_IPwhere$NEW_IPis a placeholder for user-supplied data. - [DATA_EXPOSURE]: Hardcoded IP addresses (
43.159.133.35,106.54.241.187) and administrative usernames are exposed within the skill instructions.
Recommendations
- AI detected serious security threats
Audit Metadata