fix(proxmox): embed SSH host key via NIXOS_HOST_KEYS_DIR so sops can decrypt on first boot

--pre-format-files placed the key on the QEMU builder VM's rootfs, not the
target disk. nixos-install chroots into the target and runs sshd-keygen, which
found no key in the chroot and generated a fresh (unregistered) one. sops then
could not decrypt on first boot because the key didn't match .sops.yaml, leaving
both root and nixos with '!' in /etc/shadow even after mutableUsers = false was
set (hashedPasswordFile pointed to paths sops never wrote).

Fix modules/platforms/proxmox.nix to embed the clan SSH host key in
environment.etc via NIXOS_HOST_KEYS_DIR at eval time -- the same pattern
lxc.nix uses. nixos-install's own activation places the key on the target disk,
sshd-keygen finds it already present and skips generation, and sops decrypts
correctly on first boot. Includes the same preserveSshHostKey/restoreSshHostKey
activation scripts as lxc.nix so subsequent nixos-rebuild switch calls (without
NIXOS_HOST_KEYS_DIR) don't remove the key as "obsolete" from environment.etc.

Update create-proxmox-resource.sh: switch VM builds from
  ./result-<target> --pre-format-files ... --build-memory 2048
to
  NIXOS_HOST_KEYS_DIR=$(pwd)/host-keys nix build --impure ... diskoImagesScript
  ./result-<target> --build-memory 2048
matching the LXC build path.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011uRcikkTp3D5VbXj2DwNpQ

CLAUDE.md

@@ -27,12 +27,13 @@ machines when deployed.
   template for a *real* host — every other host uses sops-nix
   (`hashedPasswordFile`, see "Security Notes" in `README.md`). Flag any *new*
   secret-like string you encounter instead of committing it.
-- `host-keys/` is gitignored — locally-generated *private* SSH host keys
-  for the auto-installer and non-LXC hosts (see `docs/auto-installer.md`).
-  Never commit its contents; if `git status` ever shows it as trackable,
-  something is wrong. Deployed LXC hosts use clan vars
-  (`vars/per-machine/<target>/openssh/`, committed and sops-encrypted)
-  instead — those ARE tracked by git and belong in the repo.
+- `host-keys/` is gitignored — used only by the auto-installer's own
+  environment for pre-seeding non-LXC host keys before first boot (see
+  `docs/auto-installer.md`). Never commit its contents; if `git status`
+  ever shows it as trackable, something is wrong. All deployed hosts use
+  clan vars (`vars/per-machine/<target>/openssh/`, committed and
+  sops-encrypted) for their SSH host keys — those ARE tracked by git and
+  belong in the repo.
 
 ### Two Proxmox nodes: `pve1.sweet.home` (production) and `pve-test.sweet.home` (sandbox)
 
@@ -206,11 +207,11 @@ instead of copying it.
 - `scripts/secrets/sync-host-keys.sh` — generates/registers SSH host keys
   and their `.sops.yaml`/`secrets/*.yaml` recipients for flake targets,
   idempotently (`--all`, `<target>`, `--remove`, `--regenerate-all-keys`,
-  all with `--dry-run`). For lxc-* targets it stores keys as clan vars
-  (`vars/per-machine/<target>/openssh/`, committed and sops-encrypted);
-  for other targets it uses the gitignored `host-keys/` directory. The
-  primary tool for provisioning a new host's secrets access — see
-  "Creating a new machine" in `docs/auto-installer.md`.
+  all with `--dry-run`). Stores keys as clan vars
+  (`vars/per-machine/<target>/openssh/`, committed and sops-encrypted) for
+  all flake targets. The primary tool for provisioning a new host's
+  secrets access — see "Creating a new machine" in
+  `docs/auto-installer.md`.
 - `scripts/secrets/prepare-host-key.sh` — narrower predecessor: generates a
   key by an arbitrary name without touching `.sops.yaml`. Still useful to
   pre-generate a key before its flake target exists yet, since

README.md

@@ -72,8 +72,8 @@ nix eval --json .#nixosConfigurations --apply builtins.attrNames | jq -r '.[]'
 | `modules/common/` | Shared NixOS config, Home Manager, aliases imported by every host |
 | `modules/nix-cache/` | Binary cache and remote builder client/server modules |
 | `modules/installer/` | Auto-installer environment (ISO, also served as PXE netboot) — see `docs/auto-installer.md` |
-| `host-keys/` | Gitignored, locally-generated SSH host keys for the auto-installer (non-LXC hosts) — see `docs/auto-installer.md`. LXC hosts use clan vars (`vars/per-machine/<target>/openssh/`) instead |
-| `vars/per-machine/` | Clan vars: committed, sops-encrypted SSH host keys for deployed LXC hosts; read by `create-proxmox-resource.sh` at deploy time |
+| `host-keys/` | Gitignored; only used by the auto-installer environment for pre-seeding SSH host keys before first boot — see `docs/auto-installer.md`. All deployed hosts use clan vars (`vars/per-machine/<target>/openssh/`) instead |
+| `vars/per-machine/` | Clan vars: committed, sops-encrypted SSH host keys for all deployed hosts; read by `create-proxmox-resource.sh` at deploy time |
 | `docs/` | Operational notes for cache, builders, lock updates, boot services, the auto-installer, and Proxmox image builds |
 | `scripts/` | Codex setup, validation, host-key, release-bump, and Proxmox resource helpers |
 
@@ -163,8 +163,9 @@ sops-nix-everywhere: it has a hardcoded login password instead (no stable
 per-boot host key for sops-nix to derive from on ephemeral media) — see
 "Host keys" in `docs/auto-installer.md` for why, and how the private keys it
 *does* pre-seed for target hosts stay out of git via the gitignored
-`host-keys/` directory. Deployed LXC hosts instead use clan vars
-(`vars/per-machine/<target>/openssh/`, committed and sops-encrypted).
+`host-keys/` directory. All deployed hosts use clan vars
+(`vars/per-machine/<target>/openssh/`, committed and sops-encrypted) for
+their SSH host keys.
 
 This repository's git *history* still contains secrets committed before this
 migration (see `remove-sensetive-info-refactor.md`) — those are being

modules/common/configuration.nix

@@ -61,13 +61,20 @@
     !include ${config.sops.templates."nix-github-token.conf".path}
   '';
 
-  #Set root password
-  users.users.root = {
+  users = {
+    # With mutableUsers = false, update-users-groups.pl enforces hashedPasswordFile
+    # on every activation regardless of whether the account already exists in
+    # /etc/shadow. The default (true) only applies hashedPasswordFile to newly-
+    # created accounts — which means a freshly-built proxmox disk image (where
+    # activation runs without a usable sops key, so both accounts land in shadow
+    # with ‘!’) will never have its passwords fixed by subsequent boots.
+    mutableUsers = false;
+
+    users.root = {
       hashedPasswordFile = config.sops.secrets."root-hashedPassword".path;
     };
 
-  # Define a user account. Don't forget to set a password with ‘passwd’.
-  users.users.${vars.primaryUser} = {
+    users.${vars.primaryUser} = {
       isNormalUser = true;
       extraGroups = [ "wheel" ]; # Enable ‘sudo’ for the user.
       packages = with pkgs; [
@@ -80,6 +87,7 @@
         "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIGygkCljN6uKpdJbHTOQtn8ZnH+wKXDLAwrDFbLrE/65 nixos@nixos"
       ];
     };
+  };
 
 
   # Enable the OpenSSH daemon.

modules/platforms/proxmox.nix

@@ -1,9 +1,81 @@
-{ ... }:
+{ lib, flakeTarget, ... }:
 
+let
+  # Bakes this exact flake target's pre-generated SSH host key straight
+  # into /etc/ssh/ -- mirrors lxc.nix's builtins.getEnv pattern (impure
+  # and empty under normal `nix build`/`nix eval`, so this is a no-op
+  # unless explicitly opted into with NIXOS_HOST_KEYS_DIR=... --impure).
+  #
+  # Unlike --pre-format-files (which places files on the QEMU builder VM's
+  # rootfs, not the target disk), embedding via environment.etc here means
+  # nixos-install's own activation step installs the key onto the target
+  # disk. sshd-keygen then finds it already present and skips generation,
+  # so the disk image boots with the clan-registered key and sops can
+  # decrypt on first boot.
+  #
+  # Without this, nixos-install's sshd-keygen activation generates a fresh
+  # key (unregistered in .sops.yaml), sops decryption fails permanently,
+  # and password hashes are never applied -- confirmed live: passwords
+  # stayed '!' even with mutableUsers = false because hashedPasswordFile
+  # pointed to a path that sops never wrote.
+  hostKeysDirStr = builtins.getEnv "NIXOS_HOST_KEYS_DIR";
+  hasHostKeysDir = hostKeysDirStr != "" && builtins.pathExists hostKeysDirStr;
+  hostKeysDir = /. + hostKeysDirStr;
+
+  privKeyFile = hostKeysDir + "/${flakeTarget}_ssh_host_ed25519_key";
+  pubKeyFile = hostKeysDir + "/${flakeTarget}_ssh_host_ed25519_key.pub";
+  hasKeyForThisTarget =
+    hasHostKeysDir
+    && builtins.pathExists privKeyFile
+    && builtins.pathExists pubKeyFile;
+in
 {
   imports = [
     ../hardware-configuration/vm/proxmox.nix
     ../boot/efi.nix
     ../disko/proxmox.nix
   ];
+
+  environment.etc = lib.mkIf hasKeyForThisTarget {
+    "ssh/ssh_host_ed25519_key" = {
+      source = privKeyFile;
+      mode = "0600";
+    };
+    "ssh/ssh_host_ed25519_key.pub" = {
+      source = pubKeyFile;
+      mode = "0644";
+    };
+  };
+
+  # NixOS's etc activation removes any /etc file that was in the previous
+  # generation's environment.etc but is absent from the current one. Since
+  # the SSH key is only in environment.etc during the --impure build (when
+  # NIXOS_HOST_KEYS_DIR is set), normal rebuilds would remove it as
+  # "obsolete". These scripts mirror lxc.nix's approach: save the live key
+  # before etc runs, restore it after. Without the explicit deps, the
+  # topological sort places preserveSshHostKey after etc (confirmed live on
+  # lxc-tor-relay: position 7 vs etc's position 5), so the key is gone
+  # before it can be saved.
+  system.activationScripts = {
+    preserveSshHostKey = ''
+      if [ -f /etc/ssh/ssh_host_ed25519_key ]; then
+        cp /etc/ssh/ssh_host_ed25519_key /run/sshd-host-key-preserve.tmp
+        cp /etc/ssh/ssh_host_ed25519_key.pub /run/sshd-host-key-preserve.pub.tmp
+      fi
+    '';
+
+    restoreSshHostKey = {
+      deps = [ "etc" ];
+      text = ''
+        if [ ! -f /etc/ssh/ssh_host_ed25519_key ] && [ -f /run/sshd-host-key-preserve.tmp ]; then
+          install -m 0600 /run/sshd-host-key-preserve.tmp /etc/ssh/ssh_host_ed25519_key
+          install -m 0644 /run/sshd-host-key-preserve.pub.tmp /etc/ssh/ssh_host_ed25519_key.pub
+        fi
+        rm -f /run/sshd-host-key-preserve.tmp /run/sshd-host-key-preserve.pub.tmp
+      '';
+    };
+
+    etc = { deps = [ "preserveSshHostKey" ]; };
+    setupSecrets = { deps = [ "restoreSshHostKey" ]; };
+  };
 }

scripts/proxmox/create-proxmox-resource.sh

@@ -302,6 +302,12 @@ platform_prefix="lxc"
 [[ -z "$cores" ]] && cores="$PROXMOX_DEFAULT_CORES"
 [[ -z "$memory" ]] && memory="$PROXMOX_DEFAULT_MEMORY_MB"
 
+if [[ "$type" == "vm" && -n "$disk_size" ]]; then
+  echo "WARNING: --disk-size is LXC-only for create mode and is ignored for VMs." >&2
+  echo "  VM disk size comes from proxmoxImageSize in variables.nix (currently ${disk_size}G was requested)." >&2
+  echo "  To expand after creation, use: --modify --vmid <n> --grow-disk <GB>" >&2
+fi
+
 # --- discover / resolve the flake target from --host --------------------
 # Emits "<target>\t<hostName>" pairs for every ${platform_prefix}-* flake
 # target -- the one source both --list and the --host lookup below read
@@ -354,6 +360,14 @@ fi
 # feeds straight into the guest's real hostname) disagree with host.nix.
 [[ -z "$name" ]] && name="$host"
 
+# For VM builds: the diskoImagesScript (run via QEMU on the node) writes the
+# raw disk image as <hostname>.raw into the CWD it was called from (the remote
+# repo dir), not to /var/lib/vz/import/ or anywhere else.  Import directly from
+# there -- no intermediate mv that can fail crossing filesystem boundaries or
+# leave a stale file on error.
+vm_built_raw=""
+[[ "$type" == "vm" ]] && vm_built_raw="${remote_repo_dir}/${host}.raw"
+
 # --- refuse to duplicate a host that's already live on the node ---------
 # Queries the node itself (qm/pct's own name/hostname config), not any
 # static list in this repo -- a file can't track whether a resource still
@@ -688,10 +702,14 @@ if [[ -n "$image" ]]; then
 elif [[ "$force_rebuild" -eq 1 ]]; then
   echo "--force-rebuild: skipping the existing-image check on ${node}."
 else
-  echo "==> Checking whether ${node} already has ${remote_path}..."
+  # VMs: check for the raw image in the remote repo dir (where disko writes it).
+  # LXC: check for the tarball in iso_storage (where the LXC build stages it).
+  _check_path="$remote_path"
+  [[ "$type" == "vm" ]] && _check_path="$vm_built_raw"
+  echo "==> Checking whether ${node} already has ${_check_path}..."
   if [[ "$dry_run" -eq 1 ]]; then
-    echo "[dry-run] would check: ssh ${ssh_target} -- test -f ${remote_path}"
-  elif ssh "$ssh_target" "test -f '${remote_path}'" 2>/dev/null; then
+    echo "[dry-run] would check: ssh ${ssh_target} -- test -f ${_check_path}"
+  elif ssh "$ssh_target" "test -f '${_check_path}'" 2>/dev/null; then
     echo "Found it -- reusing, skipping build (use --force-rebuild to override)."
     image_already_remote=1
   else
@@ -766,23 +784,29 @@ REMOTE_SCRIPT
     fi
   else
     if [[ "$dry_run" -eq 1 ]]; then
-      echo "[dry-run] would build on ${node}: nix build --no-use-registries --no-accept-flake-config${nix_opts_display} \\"
+      echo "[dry-run] would build on ${node}: NIXOS_HOST_KEYS_DIR=\$(pwd)/host-keys nix build --impure --no-use-registries --no-accept-flake-config${nix_opts_display} \\"
       echo "[dry-run]   .#nixosConfigurations.${flake_target}.config.system.build.diskoImagesScript"
-      echo "[dry-run] would run:   ${sudo_display}./result-${flake_target} \\"
-      echo "[dry-run]                --pre-format-files host-keys/${flake_target}_ssh_host_ed25519_key /etc/ssh/ssh_host_ed25519_key \\"
-      echo "[dry-run]                --pre-format-files host-keys/${flake_target}_ssh_host_ed25519_key.pub /etc/ssh/ssh_host_ed25519_key.pub \\"
-      echo "[dry-run]                --build-memory 2048"
-      echo "[dry-run] would stage the result at ${remote_path}"
+      echo "[dry-run] would run:   ${sudo_display}./result-${flake_target} --build-memory 2048"
+      echo "[dry-run] image will be at ${vm_built_raw} (imported from there; no mv to /var/lib/vz/import/)"
       local_image="<built-image>.raw"
     else
       echo "==> Building Disko image for ${flake_target} on ${node}..."
       # See the LXC branch above for why this is one %q-quoted command
       # string rather than separate ssh argv elements.
-      printf -v remote_cmd 'bash -s -- %q %q %q %q %q %q' \
-        "$remote_repo_dir" "$flake_target" "$remote_dir" "$remote_filename" "$NIX_EXTRA_OPTS" "$sudo_prefix"
+      # $7 = image_name (hostname, the diskoImagesScript's own output filename).
+      #
+      # NIXOS_HOST_KEYS_DIR + --impure: modules/platforms/proxmox.nix reads
+      # this env var at eval time (like lxc.nix) to embed the clan SSH host
+      # key in environment.etc. nixos-install's own activation then places the
+      # key on the target disk, so sshd-keygen finds it already present and
+      # skips generation. --pre-format-files put the key on the QEMU builder
+      # VM's rootfs (not the target disk), so sshd-keygen regenerated a fresh
+      # key -- one not registered in .sops.yaml -- and sops could never decrypt.
+      printf -v remote_cmd 'bash -s -- %q %q %q %q %q %q %q' \
+        "$remote_repo_dir" "$flake_target" "$remote_dir" "$remote_filename" "$NIX_EXTRA_OPTS" "$sudo_prefix" "$host"
       ssh "$ssh_target" "$remote_cmd" <<'REMOTE_SCRIPT'
 set -euo pipefail
-repo_dir="$1"; target="$2"; dest_dir="$3"; dest_name="$4"; nix_extra_opts_str="$5"; sudo_pfx="$6"
+repo_dir="$1"; target="$2"; dest_dir="$3"; dest_name="$4"; nix_extra_opts_str="$5"; sudo_pfx="$6"; image_name="$7"
 declare -a NIX_OPTS=()
 [[ -n "$nix_extra_opts_str" ]] && eval "NIX_OPTS=(${nix_extra_opts_str})"
 cd "$repo_dir"
@@ -794,24 +818,26 @@ if [[ ! -f "host-keys/${target}_ssh_host_ed25519_key" ]]; then
   echo "and ensure it was synced here before starting the build." >&2
   exit 1
 fi
-nix build --no-use-registries --no-accept-flake-config "${NIX_OPTS[@]}" \
+# Build diskoImagesScript with NIXOS_HOST_KEYS_DIR so proxmox.nix embeds the
+# clan SSH key in environment.etc (same as lxc.nix). This causes nixos-install
+# to place the key on the target disk, so sshd-keygen finds it and skips
+# generation -- the disk image boots with the registered key, sops decrypts.
+NIXOS_HOST_KEYS_DIR="$(pwd)/host-keys" nix build --impure \
+  --no-use-registries --no-accept-flake-config "${NIX_OPTS[@]}" \
   ".#nixosConfigurations.${target}.config.system.build.diskoImagesScript" \
   --out-link "result-${target}"
-$sudo_pfx "./result-${target}" \
-  --pre-format-files "$(pwd)/host-keys/${target}_ssh_host_ed25519_key" /etc/ssh/ssh_host_ed25519_key \
-  --pre-format-files "$(pwd)/host-keys/${target}_ssh_host_ed25519_key.pub" /etc/ssh/ssh_host_ed25519_key.pub \
-  --build-memory 2048
-built="$(find . -maxdepth 1 -name '*.raw' -newer "result-${target}" | head -1)"
-if [[ -z "$built" ]]; then
-  echo "ERROR: no .raw image found in ${repo_dir} after build." >&2
+# Remove any stale .raw from a previous failed build so the post-build check
+# below is unambiguous (diskoImagesScript writes to CWD as ${image_name}.raw).
+$sudo_pfx rm -f "${image_name}.raw" 2>/dev/null || true
+$sudo_pfx "./result-${target}" --build-memory 2048
+if [[ ! -f "${image_name}.raw" ]]; then
+  echo "ERROR: ${image_name}.raw not found in ${repo_dir} after build -- disko/QEMU may have failed." >&2
   exit 1
 fi
-$sudo_pfx mkdir -p "$dest_dir"
-$sudo_pfx mv "$built" "${dest_dir}/${dest_name}"
-echo "Built and staged: ${dest_dir}/${dest_name}"
+echo "Built image: ${repo_dir}/${image_name}.raw"
 REMOTE_SCRIPT
-      local_image="$remote_path"
-      echo "Built on ${node}: ${remote_path}"
+      local_image="$vm_built_raw"
+      echo "Built on ${node}: ${vm_built_raw}"
     fi
   fi
 fi
@@ -889,14 +915,35 @@ else
 --net0 virtio,bridge=${bridge} --bios ovmf --machine q35 --scsihw virtio-scsi-pci \
 --efidisk0 ${storage}:1,efitype=4m,pre-enrolled-keys=0 --agent enabled=1"
 
+  # VMs built on the node: import from the repo dir (where disko/QEMU wrote it).
+  # VMs from --image: import from remote_path (where scp uploaded it).
+  _import_path="${remote_path}"
+  [[ -z "$image" ]] && _import_path="${vm_built_raw}"
   if [[ "$dry_run" -eq 1 ]]; then
-    echo "[dry-run] ssh ${ssh_target} -- ${sudo_display}qm importdisk ${vmid} ${remote_path} ${storage}"
+    echo "[dry-run] ssh ${ssh_target} -- ${sudo_display}qm importdisk ${vmid} ${_import_path} ${storage}"
     echo "[dry-run] (would parse the resulting disk identifier from that output)"
     echo "[dry-run] ssh ${ssh_target} -- ${sudo_display}qm set ${vmid} --scsi0 ${storage}:<parsed-disk-id>"
   else
-    importdisk_output="$(ssh "$ssh_target" "${sudo_prefix} qm importdisk ${vmid} ${remote_path} ${storage}")"
+    if ! importdisk_output="$(ssh "$ssh_target" "${sudo_prefix} qm importdisk ${vmid} ${_import_path} ${storage}" 2>&1)"; then
+      echo "ERROR: qm importdisk failed:" >&2
+      echo "${importdisk_output}" >&2
+      exit 1
+    fi
     echo "$importdisk_output"
-    disk_id="$(echo "$importdisk_output" | grep -oP "(?<=Successfully imported disk as ')[^']+" | sed 's/^unused[0-9]*://')"
+    # PVE output format: "unusedN: successfully imported disk '<storage>:<vol>'"
+    # (lowercase "successfully", no "as"; the primary regex targets this form; the
+    # || true inside the substitution prevents set -e from aborting when grep finds
+    # no match -- without it the script would silently exit before reaching the
+    # fallback whenever the PVE format doesn't match).
+    disk_id="$(echo "$importdisk_output" | grep -oP "successfully imported disk '\\K[^']+" || true)"
+    if [[ -z "$disk_id" ]]; then
+      # Fallback for other PVE output variants: read qm config directly.
+      unused_line="$(ssh "$ssh_target" "${sudo_prefix} qm config ${vmid}" | grep '^unused[0-9]*:' | head -1 || true)"
+      if [[ -n "$unused_line" ]]; then
+        disk_id="${unused_line#*: }"
+        echo "Note: disk ID resolved from qm config: ${disk_id}"
+      fi
+    fi
     if [[ -z "$disk_id" ]]; then
       echo "ERROR: couldn't parse the imported disk identifier from qm importdisk's output above." >&2
       echo "The VM shell (${vmid}) and imported disk both exist -- finish attaching it by hand:" >&2
@@ -905,6 +952,12 @@ else
       exit 1
     fi
     remote "${sudo_prefix} qm set ${vmid} --scsi0 ${disk_id}"
+    # The disk data is now in ZFS; remove the source raw file (only for images
+    # we built on the node -- --image uploads are the operator's to manage).
+    if [[ -z "$image" ]]; then
+      ssh "$ssh_target" "${sudo_prefix} rm -f '${_import_path}'" 2>/dev/null || \
+        echo "Warning: couldn't remove ${_import_path} from ${node} -- you can delete it manually" >&2
+    fi
   fi
   remote "${sudo_prefix} qm set ${vmid} --boot order=scsi0"
   remote "${sudo_prefix} qm start ${vmid}"

scripts/secrets/sync-host-keys.sh

@@ -11,17 +11,16 @@
 #   sync-host-keys.sh --regenerate-all-keys Remove and freshly regenerate
 #                                           every locally-managed key.
 #
-# "Generate/register" is idempotent and additive only: an existing
-# host-keys/ file is never touched, and .sops.yaml only ever gains an
-# anchor/alias it doesn't already have -- safe to re-run any time, e.g.
-# right after adding a new host to flake.nix.
+# "Generate/register" is idempotent and additive only: an existing clan
+# var is never overwritten, and .sops.yaml only ever gains an anchor/alias
+# it doesn't already have -- safe to re-run any time, e.g. right after
+# adding a new host to flake.nix.
 #
-# --remove and --regenerate-all-keys only ever operate on anchors that have
-# a corresponding host-keys/<name>_ssh_host_ed25519_key file. Anchors
-# without one (&admin, and any anchor for an already-deployed host whose
-# real /etc/ssh key was registered by hand, e.g. &docker/&server/&nix-cache
-# today) are never listed, removed, or regenerated -- this tooling only
-# ever touches keys it itself manages.
+# --remove and --regenerate-all-keys only ever operate on anchors that
+# have a corresponding clan var (vars/per-machine/<name>/openssh/) or
+# host-keys/ file. Anchors without either (&admin) are never listed,
+# removed, or regenerated -- this tooling only ever touches keys it itself
+# manages.
 set -euo pipefail
 
 repo_root="$(cd "$(dirname "$0")/../.." && pwd)"
@@ -56,13 +55,14 @@ Usage: $0 --all [--dry-run]
   <flake-target>            Same, for just one target (e.g. lxc-server).
                             Reports if it already has one.
   --remove                  Interactively pick one locally-managed key to
-                            remove from .sops.yaml and host-keys/.
+                            remove from .sops.yaml and vars/per-machine/
+                            (or host-keys/ for legacy keys).
   --regenerate-all-keys     Remove every locally-managed key and generate
-                            fresh replacements for every current flake
-                            target. Destructive -- requires typed
+                            fresh clan-var replacements for every current
+                            flake target. Destructive -- requires typed
                             confirmation.
   --dry-run                 Combine with any of the above: print what would
-                            change (host-keys/ files, .sops.yaml anchors and
+                            change (clan vars, .sops.yaml anchors and
                             key_groups, which secrets/*.yaml would be
                             re-encrypted) without touching anything. No keys
                             generated, no files written, no sops calls,
@@ -135,10 +135,17 @@ discover_targets() {
 }
 
 locally_managed_hosts() {
+  {
     for f in "$keydir"/*_ssh_host_ed25519_key.pub; do
       [[ -e "$f" ]] || continue
       basename "$f" _ssh_host_ed25519_key.pub
     done
+    local d
+    for d in "${repo_root}/vars/per-machine"/*/openssh/ssh_host_ed25519_key/secret; do
+      [[ -f "$d" ]] || continue
+      basename "$(dirname "$(dirname "$(dirname "$d")")")"
+    done
+  } | sort -u
 }
 
 add_keys_json="[]"
@@ -308,7 +315,7 @@ cmd_remove() {
   local hosts
   mapfile -t hosts < <(locally_managed_hosts)
   if [[ "${#hosts[@]}" -eq 0 ]]; then
-    echo "No locally-managed keys in host-keys/ -- nothing to remove."
+    echo "No locally-managed keys found (checked host-keys/ and vars/per-machine/) -- nothing to remove."
     return
   fi
 
@@ -317,7 +324,9 @@ cmd_remove() {
   for host in "${hosts[@]}"; do
     local registered="not registered in .sops.yaml"
     grep -qE "^  - &${host} age1" "$sops_yaml" && registered="registered in .sops.yaml"
-    printf '  %d) %s (%s)\n' "$i" "$host" "$registered"
+    local where="host-keys/"
+    clan_ssh_key_exists "$host" "$repo_root" && where="clan-vars"
+    printf '  %d) %s [%s, %s]\n' "$i" "$host" "$where" "$registered"
     i=$((i + 1))
   done
 
@@ -334,7 +343,7 @@ cmd_remove() {
   local target="${hosts[$((choice - 1))]}"
 
   if [[ "$dry_run" -ne 1 ]]; then
-    read -rp "Really remove '${target}'? Its host-keys/ files will be deleted and it will lose access to every secrets file it can currently decrypt. (y/N): " confirm
+    read -rp "Really remove '${target}'? Its key files will be deleted and it will lose access to every secrets file it can currently decrypt. (y/N): " confirm
     if [[ ! "$confirm" =~ ^[Yy]$ ]]; then
       echo "Cancelled."
       return
@@ -347,11 +356,13 @@ cmd_remove() {
   apply_edit_plan "$plan"
 
   if [[ "$dry_run" -eq 1 ]]; then
-    echo "[dry-run] would delete host-keys/${target}_ssh_host_ed25519_key(.pub)."
+    echo "[dry-run] would delete host-keys/${target}_ssh_host_ed25519_key(.pub) if present."
+    echo "[dry-run] would delete vars/per-machine/${target}/openssh/ if present."
     echo "[dry-run] Nothing was changed. Re-run without --dry-run to apply this."
   else
     rm -f "${keydir}/${target}_ssh_host_ed25519_key" "${keydir}/${target}_ssh_host_ed25519_key.pub"
-    echo "Removed host-keys/${target}_ssh_host_ed25519_key(.pub)."
+    rm -rf "${repo_root}/vars/per-machine/${target}/openssh"
+    echo "Removed key for ${target} (host-keys/ and/or vars/per-machine/ as applicable)."
     echo
     echo "Review the diff, then commit and push."
   fi
@@ -361,7 +372,7 @@ cmd_regenerate_all() {
   local hosts
   mapfile -t hosts < <(locally_managed_hosts)
   if [[ "${#hosts[@]}" -eq 0 ]]; then
-    echo "No locally-managed keys in host-keys/ -- nothing to regenerate."
+    echo "No locally-managed keys found (checked host-keys/ and vars/per-machine/) -- nothing to regenerate."
     return
   fi
 
@@ -390,8 +401,8 @@ cmd_regenerate_all() {
   apply_edit_plan "$plan"
 
   if [[ "$dry_run" -eq 1 ]]; then
-    echo "[dry-run] would delete ${#hosts[@]} host-keys/ file pair(s)."
-    echo "[dry-run] would then generate fresh replacements for the same hosts"
+    echo "[dry-run] would delete ${#hosts[@]} key pair(s) from host-keys/ and/or vars/per-machine/."
+    echo "[dry-run] would then generate fresh clan vars replacements for the same hosts"
     echo "[dry-run] (not simulated further here -- run without --dry-run, or"
     echo "[dry-run] preview a specific target with: $0 <target> --dry-run)."
     echo
@@ -403,8 +414,9 @@ cmd_regenerate_all() {
   local host
   for host in "${hosts[@]}"; do
     rm -f "${keydir}/${host}_ssh_host_ed25519_key" "${keydir}/${host}_ssh_host_ed25519_key.pub"
+    rm -rf "${repo_root}/vars/per-machine/${host}/openssh"
   done
-  echo "Removed ${#hosts[@]} host-keys/ file pair(s)."
+  echo "Removed ${#hosts[@]} key pair(s)."
 
   echo
   echo "Regenerating fresh keys for every current flake target..."

variables.nix

@@ -166,7 +166,7 @@
   # build (modules/disko/proxmox.nix, config.system.build.diskoImagesScript
   # — see docs/proxmox-images.md). Root fills whatever's left after the ESP
   # and swap partitions within this total.
-  proxmoxImageSize = "20G";
+  proxmoxImageSize = "50G";
 
   # nix-cache's Nix store garbage collection retention
   # (modules/nix-cache/server.nix).