Sharing something that’s saved me a few times. I run a small Proxmox cluster + Docker workloads + Unraid for media, and the thing that kept biting me was services that thought they were being backed up but weren’t — usually because I’d swapped a volume mount or moved a container and forgotten to update the backup target. The disk fails, you go to restore, and the restore is from six weeks ago. Or it’s a stateful service that was never in the backup config at all.

I now run this prompt weekly against an inventory file I maintain by hand. It’s a single LLM prompt — works with Claude, GPT-4, or Gemini. Paste in the inventory, get back a risk-flagged report with the things you’d want to know about before you need them.

The inventory format (one service per line, comma-separated):

name, type, host, last_backup_iso

Example:

Postgres-01, database, vm-101, 2026-05-17
Plex, media, lxc-201, 2026-05-09
Vaultwarden, password-mgr, vm-103,
Caddy, reverse-proxy, vm-101, 2026-05-16
Gitea, code, vm-102, 2026-05-22
Home Assistant, smarthome, lxc-204,

Leave last_backup_iso blank if the service has no known backup — that’s exactly the case you want flagged. I generate this list by running pvesh ls / + docker ps -a --format '{{.Names}}' and stitching the output, but a hand-maintained file works fine.

The prompt:

You are a senior systems operator auditing my homelab. Here is my service inventory:

[paste the comma-separated list above]

For each service:
- Flag RED if last_backup is older than 7 days
- Flag YELLOW if last_backup is older than 24 hours
- Flag NEVER if the service has no last_backup at all AND its type indicates state
  (databases, password managers, smart-home configs, code repos, document stores,
  media metadata stores — but not reverse proxies, stateless web frontends, etc.)
- Suggest a target backup frequency by service type:
  databases → daily; password managers → daily;
  document/code stores → daily; smart-home configs → weekly;
  media libraries → weekly metadata + occasional full;
  reverse proxies → on-config-change only

Output format:
1. Risk-sorted markdown table (RED/NEVER/YELLOW/OK, in that order)
2. Top 3 specific actions ordered by impact, written as a one-line task each
3. Any pattern you notice that suggests a configuration gap (not just per-service)

Example output running this on the sample inventory above (real run, not staged):

Service	Type	Host	Last backup	Flag
Vaultwarden	password-mgr	vm-103	—	NEVER
Home Assistant	smarthome	lxc-204	—	NEVER
Plex	media	lxc-201	2026-05-09	RED (14d)
Caddy	reverse-proxy	vm-101	2026-05-16	OK (stateless)
Postgres-01	database	vm-101	2026-05-17	YELLOW (6d)
Gitea	code	vm-102	2026-05-22	OK

Top 3 actions:

1. Add Vaultwarden to backup target NOW — losing the password vault is a credential disaster, not a recovery inconvenience.
2. Add Home Assistant config + database to backup — losing automations + entity history takes weeks to reconstruct from memory.
3. Investigate why Plex metadata stopped being captured 14 days ago — likely a volume-mount change after the last container update.

Pattern noticed: All “never backed up” services share the host vm-103 / lxc-204. Check whether this host class was excluded from your backup job’s target list — common after Proxmox node renames or LXC migrations.

---

What this won’t do: it’s not a monitor — it doesn’t query Proxmox or Docker for you. You have to keep the inventory file current. Worst case it tells you what you already knew. Best case (and the case that’s actually saved me) it spots the silently-stopped one.

Caveats:

• Output quality scales with model. Claude 3.5+ / GPT-4 / Gemini 2.5+ give the structured table reliably. Smaller models get sloppy.
• The “type” field matters more than you’d think — vague types (“server”, “thing”) produce worse risk classification.
• Re-run the inventory + prompt monthly; the “NEVER” column is where the real risk lives and it tends to grow as you add services.

Happy to answer questions about the inventory format, prompt tweaks, or how I generate the list. Also have a longer set of these for migration risk, security audits, and capacity planning — but the one above is the highest-leverage single prompt I run.