vps-server-management – Agents & Automation skill for AI agents

VPS Server Management

When to Use

• Use when the user asks to operate an authorized VPS or agent running on a remote host.
• Use when SSH, deployment, restart, status, or log inspection is needed with explicit permission.

Source of truth: library/infrastructure.md (read it for the latest — IPs/expirations change).

Servers (Hostinger VPS) — 3 total

| Hostname | IP | OS | Purpose | Expires |

|---|---|---|---|---|

| openclaw-server | <IP> | Ubuntu 24.04 (Dokploy) | OpenClaw — personal instance | <expiry> |

| n8n-server | <IP> | Ubuntu 24.04 (n8n) | All n8n workflow automations (primary) | <expiry> |

| hermes-server | <IP> | Ubuntu 24.04 | Hermes Agent — Discord gateway (Vilnius, LT) | <expiry> |

The server table is an upstream example, not the current user’s inventory; library/infrastructure.md and library/hermes/ are not bundled here. Resolve the authorized host from the actual workspace inventory, verify its SSH host key, and use a named least-privileged account; elevate only for the operation that requires it.

Access levels (never share higher than needed)

1. App login — e.g. app.example.hstgr.cloud. Build/edit workflows, no server access. Safest to share.

2. VPS SSH — root@<IP>. Docker, files, system config. Trusted technical people only.

3. Hostinger hPanel — hpanel.hostinger.com. Billing, reboot, OS reinstall. Exposes SSH creds + browser terminal, so it grants server access too. The user only.

Managing a VPS via an agent

Start with read-only service status and scoped redacted logs. For multi-step work, use an existing authorized remote session or a scoped on-host agent with normal approval controls. Before an update or restart, identify the affected service, backup/rollback path and expected interruption; verify service health afterward. Do not disable agent safeguards merely to simplify SSH operations.

When checking on a remote/on-box agent, send the user one concise status line each time: what it is doing and whether it is on track.

Claude Code cmux note: after Claude finishes, it may prefill a predicted next user message; that draft is Claude, not the user speaking.

Agents on servers

• OpenClaw → openclaw-server (managed via Dokploy).
• Hermes → hermes-server (Discord gateway). Setup/config docs in library/hermes/.
• n8n → n8n-server.

Hermes ops (on hermes-server)

hermes --version            # shows version + commits behind
hermes update               # auto-snapshots, updates deps, rebuilds web UI, restarts gateway itself
hermes gateway status|restart
journalctl --user -u hermes-gateway --since '5 min ago' --no-pager   # gateway logs (systemd USER service)

• Default model lives in ~/.hermes/config.yaml under model.provider + model.default — NOT in .env. Change via hermes model (interactive) or edit the yaml directly, then hermes gateway restart to propagate to gateways.
• Treat EBADENGINE as evidence to check supported runtime ranges. Do not assume an unsupported runtime is safe because installation completed; plan a compatible dependency/runtime change and verify the gateway before declaring the update healthy.
• Deeper docs (Discord/Slack/WhatsApp setup, file structure, vision config): library/hermes/.

Limitations

• Adapted from davidondrej/skills; verify local paths, tools, credentials, and agent features before acting.
• For commands, remote access, scheduling, browser automation, or file-changing workflows, get explicit user approval and confirm the target environment first.