.claude: Add shared permissions allow list, ignore local settings

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.claude/settings.json

@@ -0,0 +1,18 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(nix eval:*)",
+      "Bash(nix flake check:*)",
+      "Bash(nix build:*)",
+      "Bash(check-system:*)",
+      "Bash(build-system:*)",
+      "Bash(build-home:*)",
+      "Bash(git status:*)",
+      "Bash(git diff:*)",
+      "Bash(git log:*)",
+      "Bash(git show:*)",
+      "Bash(drill:*)",
+      "Bash(ping:*)"
+    ]
+  }
+}

.gitignore

@@ -4,3 +4,4 @@ result*
 !/.vms/.gitkeep
 /.keys/*.key
 *.swp
+/.claude/settings.local.json

AGENTS.md

@@ -9,6 +9,11 @@ tools refuse to write through a symlink and will error on `CLAUDE.md`).
 convention, a workflow gotcha, a design rationale — record it here (or in a repo doc this file points
 to, e.g. `home-switches.md`), not in agent memory. AGENTS.md is versioned and shared; memory is not.
 
+Claude Code permissions live in two files: `.claude/settings.json` (versioned, shared — the
+committed allow list of safe-to-auto-approve commands) and `.claude/settings.local.json` (personal,
+gitignored — where interactive "always allow" grants accumulate). Put durable, generally-safe
+commands in the shared file; leave one-off or machine-specific grants in the local one.
+
 ## Overview
 
 Personal Nix flake managing NixOS systems and home-manager configurations for a set of
@@ -53,6 +58,12 @@ Check everything (what CI runs): `nix flake check --no-build`.
 CI builds each attr of `.#ci.x86_64-linux` (systems, homes, packages, shell) and pushes to the
 Harmonia binary cache; see `.gitea/workflows/ci.yaml` and `ci/push-to-cache.sh`.
 
+For DNS lookups use **`drill`** (ldns) — `dig` isn't installed in this environment (it fails with
+exit 127, which is easy to miss if stderr is redirected). E.g. `drill -Q @<resolver> <name> A`.
+
+For privilege escalation use **`doas`**, not `sudo` — the boxes don't install `sudo` (it fails with
+`command not found`). E.g. `doas ip link set <if> up`.
+
 ## Architecture
 
 ### The custom module system
@@ -141,6 +152,14 @@ touching anything WAN/VLAN-related, and update it when the switch layout changes
   agree on VLAN numbering (e.g. `lib.my.c.home.vlans`), so a switch-side change usually pairs with a
   box change; `home-switches.md` documents the switch layout and per-switch config for the WAN design.
 
+### Home wireless APs (`vibe` / `wave`)
+The home Wi-Fi APs are also **not** managed by this flake: `vibe` (MikroTik cAP ax, RouterOS) and
+`wave` (Cudy AX3000 running OpenWrt/UCI). They are dumb APs — bridge clients onto the right VLAN,
+routers do DHCP/RA/firewall. The trunk/VLAN design, the OpenWrt flash + config for `wave`, and the
+per-AP management addressing live in **`home-aps.md`** at the repo root — read it before touching AP
+config, and update it when an AP changes. Only the DNS records live in the flake
+(`routing-common/dns.nix`).
+
 ## Secrets
 
 age-encrypted secrets in `secrets/`, managed with **ragenix**. Each module declares

home-aps.md

@@ -0,0 +1,145 @@
+# Home wireless APs
+
+Reference for the home Wi-Fi access points. Like the switches (`home-switches.md`), these are **not**
+managed by this flake — they are configured on-device (RouterOS on the MikroTik, OpenWrt/UCI on the
+Cudy). This file documents the shared VLAN/trunk design and each AP.
+
+Only the DNS records live in the flake (`nixos/boxes/home/routing-common/dns.nix`, `h.nul.ie` zone).
+Everything else here is applied by hand on the device.
+
+## The APs
+
+| | vibe | wave |
+|---|---|---|
+| Model | MikroTik **cAP ax** (`cAPGi-5HaxD2HaxD`) | Cudy **AX3000** (OpenWrt id `cudy,ap3000-v1`) |
+| OS | RouterOS 7.x | OpenWrt 25.12.x (MT7981B / Filogic 820) |
+| Radio | 2×2 both bands | 2×2 both bands (2 spatial streams) |
+| Uplink | trunk (multi-port; `ether2` is a wired LAN port) | single **2.5 GbE** trunk |
+| Management | core/hi/lo `.15` | core/lo `.14` (**not** `hi`; see below) |
+
+Both mirror the same two SSIDs. `wave` replaced an older AP of the same name; the new one is the
+Cudy running OpenWrt.
+
+> **Note on "AX3000":** MT7981B is 2×2 (2 spatial streams). The "3000" is *aggregate* Mbps —
+> 574 (2.4 GHz 2ss) + 2402 (5 GHz 2ss @ **160 MHz**) — not three streams. `iw` confirms 2×2
+> (`Available Antennas TX/RX 0x3`; "3 streams: not supported"). 160 MHz is what earns the "3000".
+
+## Shared design (dumb AP)
+
+Every AP is a **dumb AP**: it bridges wireless clients onto the right VLAN and does **no** routing,
+DHCP, RA or firewalling. The home routers (`river`/`stream`) own DHCP/RA/gateway (per-VLAN VRRP VIPs)
+and firewalling. The uplink is a **tagged trunk**:
+
+| VLAN | `lib.my.c.home.vlans` | Role | AP use |
+|---|---|---|---|
+| — (native) | `core` | switch/fabric management (1500) | backup management (untagged) |
+| 100 | `hi` | trusted LAN, **high-MTU** (jumbo 9000) | `vibe` management (it's jumbo-capable) |
+| 110 | `lo` | trusted LAN (1500) | main SSID `wlan0`; `wave` management |
+| 120 | `untrusted` | guest network | guest SSID `wlan1` |
+
+`hi` and `lo` are both **trusted** client VLANs — the only difference is MTU (`hi` carries jumbo
+9000, `lo` is standard 1500). An AP puts its own management on whichever it can do: `vibe` (jumbo)
+sits on `hi`, `wave` (eth0 capped at 2026) sits on `lo`. The main SSID lands on `lo` because Wi-Fi
+clients are 1500 regardless.
+
+### SSIDs
+
+| SSID | Bands | Security | VLAN |
+|---|---|---|---|
+| `wlan0` (main) | 5 GHz + 2.4 GHz | WPA2/WPA3-PSK (`sae-mixed`) | 110 (`lo`) |
+| `wlan1` (guest) | 2.4 GHz | WPA2-PSK (`psk2`) | 120 (`untrusted`) |
+
+**Passphrases are never stored in this repo.** `vibe` is the source of truth; read them out-of-band
+with `ssh admin@vibe '/interface wifi export show-sensitive'` (`.passphrase=` prints **unquoted**).
+
+## vibe (MikroTik cAP ax)
+
+RouterOS, one hardware-offloaded bridge `main` with `vlan-filtering=yes`. Access: `ssh admin@vibe`
+— key auth for `admin` is installed (`~/.ssh/id_rsa`), with `admin`/`admin` as a fallback.
+
+- **Uplink `ether1`** — trunk, tagged VLANs 100/110/120; native/untagged is the default VLAN 1
+  (PVID, no IP). `ether2` is a wired **access port** on VLAN 110 (`lo`). `l2mtu 9214`.
+- **Radios** — `wifi1` (5 GHz, 20/40/80) + `wifi2` (2.4 GHz, 20/40) both broadcast `wlan0`
+  (WPA2/WPA3-PSK), untagged onto VLAN 110. `wifi3` is a virtual AP on `wifi2` broadcasting `wlan1`
+  (WPA2-PSK), untagged onto VLAN 120. `country=Ireland`.
+- **Bridge VLANs** — 100 tagged `main,ether1`; 110 tagged `main,ether1` + untagged
+  `ether2,wifi1,wifi2`; 120 tagged `main,ether1` + untagged `wifi3`.
+- **Management** — `jim`/`dave`-style (core/hi/lo), on host `.15`: `192.168.64.15` on core
+  (native/untagged, backup), `192.168.68.15/22` + `2a0e:97c0:4d0:1::1:6` on the `hi` VLAN-100
+  interface (holds the default route, via the hi VIP `192.168.71.254`), and
+  `192.168.72.15/21` + `2a0e:97c0:4d0:2::1:6` on `lo` VLAN 110. No IP on `untrusted`.
+  `l2mtu 9214`, so `hi` carries jumbo (9000) here — `vibe` sits on `hi` because it *can* jumbo,
+  unlike `wave` (see its MTU note).
+- **Roaming** — 802.11k/v via a `/interface wifi steering` profile (`rrm=yes wnm=yes`,
+  `neighbor-group=home-aps`) assigned to `wifi1`/`wifi2`/`wifi3`.
+- **Resolver** — the hi VIP `192.168.71.254` / `2a0e:97c0:4d0:1::ffff`.
+
+## wave (Cudy AX3000, OpenWrt)
+
+Single-port AP, so the port is a VLAN **trunk** carrying management + both SSIDs.
+
+### Management addressing
+
+`wave` takes host `.14`, on **`lo` (primary) and `core` (backup)** — deliberately **not** `hi`,
+unlike the switches and `vibe`. `hi` is the jumbo (9000) VLAN, but `wave`'s eth0 caps at 2026 (see
+MTU note), so there's no reason to put it there; `lo` is 1500 with a proper VRRP VIP for the default
+route + resolver, and `core` has no VIP/v6 so it can only be a backup. No IP on `untrusted`. Records
+in `dns.nix`:
+
+| Name | VLAN | Address |
+|---|---|---|
+| `wave-core` | core (native/untagged) | `192.168.64.14/24` — backup, like the switches (no VIP → backup only) |
+| `wave` | lo 110 | `192.168.72.14/21`, `2a0e:97c0:4d0:2::1:5` — primary; holds the default route + resolver (lo VIP `192.168.79.254` / `2a0e:97c0:4d0:2::ffff`) |
+
+**Firewall:** management (SSH/LuCI) reachable from `core`/`lo` only; `untrusted` is a separate
+zone with `input REJECT` (and `wave` has no IP there) — **no management via the guest VLAN**.
+
+### brian switch port
+
+`wave` hangs off **brian** (UniFi). Its port is a **trunk**: tagged VLAN **110/120** (`lo` + guest),
+and **native/untagged = core** (the fabric's management VLAN, carrying `wave-core`). VLAN 100 (`hi`)
+is **not** needed here — `wave` isn't on `hi` (see Management addressing). Configure via the UniFi
+controller (brian has no CLI); see `home-switches.md`.
+
+### Flashing OpenWrt (Cudy AX3000 / `cudy_ap3000-v1`)
+
+Hardware: MT7981B, 512 MB RAM, 256 MB SPI-NAND, 1× 2.5 GbE (RTL8221B), 2×2 WiFi 6.
+
+> ⚠️ **Serial caveat:** units with a serial starting `2543…` (post ~Nov 2025) use a different flash
+> chip and can brick with older firmware. Match firmware to the unit.
+
+OpenWrt can't be flashed directly over stock. Two-stage, via a Cudy **transition** firmware (Cudy
+OpenWrt download page / `support@cudy.com`; `warnning.txt` in that bundle has the steps):
+1. Stock Cudy UI: update to **≥ 2.4.7** (adds TFTP `recovery.bin` recovery), then flash the Cudy
+   **intermediate** firmware (`cudy_ap3000-v1-sysupgrade_*.bin`), "keep settings" **unchecked**.
+   It reboots into an OpenWrt-based build at `192.168.1.1` (SSH `root`, empty password).
+2. From there, `sysupgrade -n` to vanilla OpenWrt (`…-cudy_ap3000-v1-squashfs-sysupgrade.bin` from
+   `downloads.openwrt.org`; this release ships **no** factory image — sysupgrade only).
+
+Stock default (out of box) is a DHCP client falling back to **`192.168.10.254`**; the stock UI is a
+customised LuCI (only 80/443, no SSH) with a first-boot "create admin password" wizard — so the
+stock-side flashing is done from a browser, not headless.
+
+### On-device config notes
+
+- Package manager is **`apk`** (not `opkg`). WiFi runs **`wpad-mbedtls`** (full — swapped from the
+  default `wpad-basic-mbedtls`, which lacks 802.11v). **802.11k + 802.11v** (`ieee80211k` +
+  `bss_transition`) are enabled on all SSIDs. ⚠️ Swapping wpad **live** leaves the mac80211 vifs
+  stuck in a start→teardown loop (`nl80211 ... No such device`); a `wifi reload`/`network restart`
+  won't recover it — **reboot** after `apk add wpad-mbedtls`.
+- Radios: `radio0` = 2.4 GHz, `radio1` = 5 GHz (keyed by `band`, don't assume). 5 GHz is pinned to
+  **channel 36 / HE160** (any 160 MHz block in IE is DFS; ch36 has the shortest ~60 s CAC).
+- Bridge: `br-lan` with `vlan_filtering`, single port `eth0` — tagged `110/120`, untagged/PVID
+  VLAN 1 (= native/core). SSIDs attach via `network` = `lo`/`untrusted` (= `br-lan.110`/`.120`).
+- Dumb-AP: no DHCP pools, `odhcpd.maindhcp=0`, `delegate=0` on the L3 interfaces.
+- **MTU:** all interfaces are **1500**. The `mtk_eth_soc` 2.5 GbE (`eth0`) caps at **2026 bytes**
+  (`ip link set eth0 mtu 9000` → `SIOCSIFMTU: Invalid argument`), so `wave` can't join `hi`'s jumbo
+  (9000) fabric like `vibe` does — which is precisely **why `wave` is managed on `lo`, not `hi`**
+  (see Management addressing). Nothing on `wave` needs > 1500.
+- **LuCI:** enabled, login `root` / `admin`. **SSH:** key-only (`PasswordAuth`/`RootPasswordAuth off`).
+- `iperf3` installed for throughput testing.
+
+### Access
+
+- SSH: `ssh root@wave` (key-only; `wave`/`wave-core` resolve once `dns.nix` is deployed).
+- LuCI: `http://192.168.72.14/` (or `http://wave/`), `root` / `admin`.

home-manager/modules/common.nix

@@ -216,6 +216,7 @@ in
           jq
           yq-go
           nix-tree
+          treemd
         ];
 
         sessionVariables = {

home-switches.md

@@ -0,0 +1,208 @@
+# Home switches
+
+Reference for the two MikroTik switches on the home network — **jim** and **dave** — plus the
+Ubiquiti switch **brian**, and how the home boxes and the Digiweb WAN hang off them. These switches
+are **not** managed by this flake; they are configured by hand (RouterOS on jim/dave, UniFi on
+brian). It covers the physical topology, the VLAN map, and how the Digiweb WAN reaches river.
+
+In short: the Digiweb ISP VLAN (10) is trunked straight through to river (which runs PPPoE on it),
+and the ONT's untagged management is PVID'd onto VLAN 140 at brian, its edge switch. VLAN 10 is
+carried untranslated because a single ONT makes it unique on the fabric — see
+[the WAN path](#the-digiweb-wan-path-trunked-vlan-10--pvid-140) and
+[why not translation](#why-not-translation-for-one-ont).
+
+## The switches
+
+| | jim | dave | brian |
+|---|---|---|---|
+| Identity | `jim-sw` | `dave-sw` | (UniFi) |
+| Model | CRS326-24G-2S+ | CRS504-4XQ | Ubiquiti Switch Pro XG 8 PoE |
+| Switch chip | Marvell 98DX3236 | Marvell 98DX4310 (+ Atheros 8227 for the 1G mgmt port) | — |
+| OS | RouterOS 7.18 | RouterOS 7.18 | UniFi |
+| Ports | 24×1G + 2×SFP+ | 4×QSFP28 (100G, breakout-capable) + 1G mgmt | 8×10GBASE-T PoE + 2×25G SFP28 |
+| Bridge | `main`, `vlan-filtering=yes` | `main`, `vlan-filtering=yes` | UniFi VLAN profiles |
+
+jim and dave run a single hardware-offloaded bridge (`main`) with VLAN filtering. Access to the
+MikroTiks is SSH as `admin` / `admin` by short hostname (see [Accessing the switches](#accessing-the-switches)).
+Only jim and dave can do hardware VLAN translation (`/interface ethernet switch rule` on the Marvell
+chips); brian cannot rewrite tags, only trunk/PVID them.
+
+## Physical topology
+
+The ONT terminates on brian; jim's `wan-pon-in` (`sfp-sfpplus2`) is a spare SFP+ port.
+
+```
+            Virgin Media cable modem
+                        │  VLAN 130 (wan1 / wan2 / wan-in)
+        ┌───────────────┴──────────────────────────────────────┐
+        │ jim    CRS326-24G-2S+    (Marvell 98DX3236)          │
+        │ 1G edge: fort, pronter, laptop-dock, palace-kvm,     │
+        │          ups, ether15-20, wan1/wan2/wan-in           │
+        │ wan-pon-in = sfp-sfpplus2  (spare SFP+)              │
+        └───────┬──────────────────────────────────────────────┘
+                │ dave-uplink = sfp-sfpplus1  (10G trunk)
+                │ also: palace, stream (1G secondaries),
+                │       castle (2.5G, normally down)
+        ┌───────┴──────────────────────────────────────────────┐
+        │ dave    CRS504-4XQ    (Marvell 98DX4310)             │
+        │ jim-downlink = qsfp28-3-1                            │
+        └──┬──────────────────┬───────────────────┬────────────┘
+           │ palace           │ castle            │ brian-downlink
+           │ = qsfp28-1-1     │ (100G)            │ 802.3ad LAG
+           │ (100G)           │                   │ (brian1 + brian2)
+           │                  …                   │
+   ┌───────┴────────┐                     ┌───────┴─────────────┐
+   │ palace host    │                     │ brian               │
+   │   └ river (VM) │                     │ Switch Pro XG 8 PoE │
+   └────────────────┘                     └───────┬─────────────┘
+   river WAN + LAN ride the 100G link             │    hosts the ONT
+                                                  │
+                                          ┌───────┴─────────────┐
+                                          │ ONT   (Digiweb)     │  untagged mgmt
+                                          │ PPPoE via ONT       │  192.168.100.1
+                                          └─────────────────────┘  + VLAN 10; PVID 140
+```
+
+Notes:
+- **river** runs as a VM on the **palace** host; its uplink is dave's 100G `palace` port. jim also
+  has 1G `palace`/`stream` ports, but those are secondary links and do **not** carry the WAN.
+- **stream** (the second router box) is dual-homed to both jim and dave (STP picks the active path).
+- **castle** is dual-homed but **not** via STP: its primary uplink is dave's **100G** `castle` port
+  (`et100g`, active), and it has a secondary **2.5G** link to jim's `castle` edge port (`et2.5g`,
+  **normally down** — no live failover). ⚠️ **castle's root disk is NVMe-oF over the fabric** (via
+  `et100g`→dave), so rebooting **dave** — or downing castle's `et100g` — freezes castle mid-I/O.
+  Do dave maintenance (upgrades/reboots) from a host that doesn't depend on dave for storage or
+  network, or with castle cleanly powered off; don't drive it from castle.
+- **brian** is a Ubiquiti **Switch Pro XG 8 PoE** (8×10GBASE-T), downlinked from dave over an
+  **802.3ad LAG** (`brian-downlink` = `brian1` + `brian2`, layer-2 hash). It hosts the ONT.
+
+## VLANs
+
+| VLAN | Name | Purpose |
+|---|---|---|
+| — (native) | core | Switch management, `192.168.64.0/24` (jim `.10`, dave `.11`, brian `.13`) |
+| 100 | hi | High-performance / jumbo network (MTU 9000) |
+| 110 | lo | Standard LAN |
+| 120 | untrusted | Guest / untrusted network |
+| 130 | wan | **stream's WAN** — Virgin Media cable modem (untagged on jim's `wan1`/`wan2`/`wan-in`) |
+| 140 | wan-pon-ont | ONT management, `192.168.100.0/24` (PVID'd at the ONT edge) |
+| 10 | pon-isp | Digiweb ISP transport — **trunked straight through** to river, PPPoE runs on it |
+| 141 | wan-pon-isp | **Reserved** — the translated ISP VLAN for the future multi-ONT design |
+
+Switch L3 presence (`/interface vlan` on `main`) exists **only** for VLANs the switch is managed
+from — `hi` (100) and `lo` (110), plus native core. WAN and guest VLANs deliberately have no switch
+L3 interface.
+
+## The Digiweb WAN path (trunked VLAN 10 + PVID 140)
+
+The ONT presents two things on one wire:
+- **untagged** management traffic (`192.168.100.x`), and
+- **tagged VLAN 10** carrying the Digiweb ISP session (the BRAS requires VLAN 10).
+
+With a **single ONT** there's no reason to translate anything — VLAN 10 is unique on the fabric, so
+we just carry it end to end and let river run PPPoE directly on it:
+
+1. **Untagged mgmt → VLAN 140, at the ONT's edge switch (brian).** brian sets the ONT port's PVID to
+   140 so the untagged management traffic becomes VLAN 140, and allows tagged VLAN 10 through the
+   same port. river takes `192.168.100.100/24` on VLAN 140 (matching stream's modem-mgmt `.100`) to
+   reach the ONT web UI at `192.168.100.1`. Doing the PVID at the ONT-facing edge keeps it clean —
+   the untagged frames never share a domain with anything else.
+
+2. **VLAN 10 (ISP) trunked straight through, untranslated.** brian → dave → palace carry tagged
+   VLAN 10 by ordinary bridge-VLAN membership. No `/interface ethernet switch rule`, no pinning, no
+   asymmetric-learning issues — it's just a normal tagged VLAN. river attaches PPPoE to VLAN 10
+   directly (`wan-pon-isp` netdev = VLAN `pon-isp` = 10; baby-jumbo MTU 1508 so PPP nets a clean
+   1500).
+
+Net result: **river runs PPPoE single-tagged on VLAN 10 and holds a VLAN 140 address to reach the
+ONT.** See `nixos/boxes/home/palace/vms/river.nix` for the river side.
+
+```
+ONT ──(untagged + VLAN10)── brian ──(VLAN140 + VLAN10)── dave ──(VLAN140 + VLAN10)── river
+        ONT port             │ PVID140 + tagged 10        │ plain bridging
+                             └─ brian-downlink LAG ── dave ┘
+```
+
+### Why not translation (for one ONT)?
+
+Translation would swap VLAN 10 → 141 with two pinned hardware ACL rules to keep VLAN 10 off the rest
+of the fabric. That buys nothing with a single ONT — VLAN 10 is already unique, so trunking it is
+simpler and rule-free. Translation only earns its keep when **two** ONTs both deliver VLAN 10 and
+would collide (below).
+
+## Switch configuration
+
+How each switch is set up for the Digiweb WAN path. **Confirm any change on the box before applying**
+(see [Accessing the switches](#accessing-the-switches)).
+
+**brian (UniFi)** — hosts the ONT:
+- The ONT port has **native/untagged network = VLAN 140** (PVID) and is a **tagged member of VLAN 10**,
+  so the ONT's untagged management lands on 140 and its tagged ISP frames pass through.
+- The `brian-downlink` LAG up to dave trunks **tagged 140 + tagged 10** (alongside the LAN VLANs).
+
+**dave (RouterOS)** — trunks both WAN-pon VLANs to `brian-downlink` and `palace`. The ISP VLAN 10 row:
+```
+/interface bridge vlan add bridge=main vlan-ids=10 tagged=brian-downlink,palace
+```
+VLAN 140 also spans `brian-downlink,palace` (it carries a few other members too). No switch rules —
+this is plain tagged bridging.
+
+**jim (RouterOS)** — carries **none** of the Digiweb WAN path: no translation rules, and no VLAN
+10/140/141 rows. `wan-pon-in` (`sfp-sfpplus2`) sits at `pvid=1` as a spare port. jim only handles
+stream's VLAN-130 WAN and the LAN VLANs.
+
+## Future: multiple ONTs (per-port VLAN translation)
+
+If a second ONT arrives (e.g. a Digiweb line for stream, or a second river), trunking breaks: both
+ONTs deliver **tagged VLAN 10**, and plain bridge-VLAN filtering can't tell them apart. That's when
+translation earns its place — a switch rule matches on the **ingress port**, so each ONT's VLAN 10
+becomes a *distinct* fabric VLAN:
+
+- ONT-A port: VLAN 10 → **141** (→ river)
+- ONT-B port: VLAN 10 → **142** (→ stream / second river)
+- mgmt: PVID each ONT port onto its own VLAN (140, 143, …) so both ONTs' `192.168.100.1` stay in
+  separate L2/L3 domains.
+
+The forward direction isolates naturally (each ONT maps to a different fabric VLAN). The **return**
+direction is where port targeting is mandatory: both translate *back* to VLAN 10, so bridge VLAN 10
+now has two members and a plain FDB-miss flood would leak one ONT's upstream to the other. Each
+return must be pinned to its port with `new-dst-ports`:
+```
+# ONT-A: 141 in on palace → 10, forced out ONT-A's port
+# ONT-B: 142 in on stream → 10, forced out ONT-B's port
+```
+Each ONT port must also be a tagged member of bridge VLAN 10 for correct egress tagging (the missing
+piece that otherwise shows up as pppd "Timeout waiting for PADO"). The pins bypass the FDB, so the
+two ISP sessions never mix.
+
+**Why a new switch:** jim (the only box with spare SFP+ *and* the translation feature) has just
+**one** free SFP+ port, so it can't host two ONTs. The plan is a dedicated
+**CRS305-1G-4S+** (4×SFP+, same Marvell rule support) to land multiple ONTs and do the per-port
+translation there, feeding distinct fabric VLANs up to dave.
+
+## Accessing the switches
+
+The switches resolve by **short hostname** on the home network — the home routers serve their
+records in the home zone (`nixos/boxes/home/routing-common/dns.nix`: `jim` → hi `.10`, `dave` → hi
+`.11`, `brian` → core `.13`). From a box on the home network just `ssh admin@jim` / `admin@dave`.
+
+**Key auth** for `admin` is installed on jim/dave (and the `vibe` AP) — `ssh -i ~/.ssh/id_rsa
+admin@jim` works keyless (imported via `/user ssh-keys import`). Password `admin`/`admin` remains as
+a fallback. Non-interactive password pattern (avoids the ssh-agent hang) if the key isn't available:
+
+```
+sshpass -p admin ssh -o IdentityAgent=none -o PubkeyAuthentication=no \
+  -o PreferredAuthentications=password -o StrictHostKeyChecking=accept-new \
+  -o UserKnownHostsFile=/tmp/sw_known_hosts admin@jim
+```
+
+**Always confirm config changes on the switch** (print the affected menu, apply, re-verify). brian
+is UniFi — configured through its controller, not RouterOS CLI.
+
+## Management IPs
+
+| | core (`192.168.64.0/24`) | hi (`192.168.68.0/22`) | lo (`192.168.72.0/21`) |
+|---|---|---|---|
+| jim | `.10` (on `main`) | `.10` | `.10` |
+| dave | `.11` (on `management`, the 1G Atheros port) | `.11` | `.11` |
+| brian | `.13` (core) | — | — |

nixos/boxes/home/routing-common/dns.nix

@@ -242,10 +242,14 @@ in
             shytzel IN A ${net.cidr.host 12 prefixes.core.v4}
             brian IN A ${net.cidr.host 13 prefixes.core.v4}
 
-            wave IN A ${net.cidr.host 12 prefixes.hi.v4}
-            ; wave IN AAAA ${net.cidr.host (65536+3) prefixes.hi.v6}
-            vibe IN A ${net.cidr.host 13 prefixes.hi.v4}
-            vibe IN AAAA ${net.cidr.host (65536+4) prefixes.hi.v6}
+            vibe-core IN A ${net.cidr.host 15 prefixes.core.v4}
+            vibe IN A ${net.cidr.host 15 prefixes.hi.v4}
+            vibe IN AAAA ${net.cidr.host (65536+6) prefixes.hi.v6}
+            vibe-lo IN A ${net.cidr.host 15 prefixes.lo.v4}
+            vibe-lo IN AAAA ${net.cidr.host (65536+6) prefixes.lo.v6}
+            wave-core IN A ${net.cidr.host 14 prefixes.core.v4}
+            wave IN A ${net.cidr.host 14 prefixes.lo.v4}
+            wave IN AAAA ${net.cidr.host (65536+5) prefixes.lo.v6}
 
             ups IN A ${net.cidr.host 20 prefixes.lo.v4}
             palace-kvm IN A ${net.cidr.host 21 prefixes.lo.v4}