OK so I'm not OP on that post but I can

[fluid rivet]

systemd --user without linger
Originally, ran nix‑openclaw as a user service with no linger.

What happened:
When SSH session ended, user systemd instance stopped.
Gateway silently went offline.

What we changed:
sudo loginctl enable-linger openclaw

Why: long‑running agents must survive user logouts on headless VMs.

Auto‑generated TOOLS.md in workspace (the original comment here)

What I tried: Let nix‑openclaw generate workspace/TOOLS.md.

What happened:
It wrote into workspace every time, conflicting with our policy/tooling repo.

This adds friction to managing tools/policy centrally across agents.

What I changed:
ignore/remove it and keep tool policy in a curated repo/dir.

Why: tools/policy should be opt‑in and configurable, not auto‑written for use cases like this

Nix‑shipped config keys vs gateway schema

What I tried: Used the Nix‑generated config as‑is.

What happened:
Some keys shipped in /nix/store were rejected by the gateway (e.g. messages.queue.byProvider).
Agent failed to start until we removed unsupported keys.

What I changed:
Copied config to writable location and sanitized unsupported fields.
Why: configuration must align with the gateway version, or the service fails.

This is probably unique to people building mutliple agents across earlier and later version etc as well but worth noting as well I think

I realize that was a lot, let me know if any/all of that doesn't make sense or even if it would be helpful to review my setup, I'm just one guy, but I've been using this stuff in a specific way including one openclaw who's job is mostly to manage proxmox, VM templates, and spin up / down other openclaws as needed so like I said maybe mine is just a unique edgecase, but I purposely wanted to use nix so it would be a reproducible as possible which is how i found the repo in the first place