The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and immediately executes a shell script from a remote repository using a piped command. Evidence: Phase 5 of SKILL.md contains the command curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash -s -- --skip-browser.
[COMMAND_EXECUTION]: The skill makes extensive use of sudo to perform administrative tasks, including package installation, firewall configuration, and modifying system configuration files across Phases 2, 3, 4, 7, and 8.
[PRIVILEGE_ESCALATION]: The skill performs actions that significantly alter the system's security posture, such as modifying SSH configuration in /etc/ssh/sshd_config.d/99-agent-vps-hardening.conf and adding the agentuser to the docker group.
[PERSISTENCE]: The skill establishes multiple long-term persistence mechanisms on the VPS by creating and enabling systemd services (hermes-gateway.service, hermes-dashboard.service) and timers (hermes-git-backup.timer).
[DATA_EXFILTRATION]: The skill configures a synchronization script (sync-hermes-backup.sh) and systemd timer to automatically push data from the VPS to an external GitHub repository.
[CREDENTIALS_UNSAFE]: The skill generates sensitive credentials, including a random password for the dashboard stored in /root/hermes-dashboard/credentials.txt and an unencrypted SSH deploy key in ~/.ssh/hermes_backup_deploy_key.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by connecting the agent to messaging platforms (Telegram/Discord) while maintaining high-privilege capabilities.
Ingestion points: Telegram/Discord messaging gateway (Phases 6, 7)
Boundary markers: Absent
Capability inventory: Full shell access (sudo), file system access, and network operations
Sanitization: Absent

hermes-vps-setup