diff --git a/SKILL.md b/SKILL.md
index c8d6a80..5edf8c4 100644
--- a/SKILL.md
+++ b/SKILL.md
@@ -37,6 +37,13 @@ user-invocable: true
     keywords=["boat", "dock, "launch"] />
 </references>
 
+<references>
+  <reference file="reference/home-assistant.md"
+    summary="Home Assistant: access/WS API, dual Nest integration (cloud + Matter), 2026-06-15 Matter fabric loss + recommission steps, main Lovelace dashboard"
+    categories=["home-automation"]
+    keywords=["home assistant", "HA", "home.thelonglakelodge.com", "matter", "nest", "thermostat", "climate", "dashboard", "lovelace", "HVAC", "in floor heating", "environmentals", "core_matter_server", "172.24.24.50"] />
+</references>
+
 <references>
   <reference file="reference/website.md"
     summary="Lodge website: site-publish flow, /welcome PDF redirect, out-of-band PDF update via aws-cli"
diff --git a/reference/home-assistant.md b/reference/home-assistant.md
new file mode 100644
index 0000000..48be8b3
--- /dev/null
+++ b/reference/home-assistant.md
@@ -0,0 +1,87 @@
+# Home Assistant
+
+<overview>
+HA OS at `home.thelonglakelodge.com`, proxied via fritzlab sjc001 remote-proxy
+to `172.24.24.50:8123` (also directly reachable on the lab network). Supervisor
+Observer on `:4357` (read-only health, no auth). API token in `~/.env` as
+`HA_TOKEN` (long-lived user token — has core API + service calls, but NOT the
+`/api/hassio/*` Supervisor proxy → 401). No SSH (`:22` closed, no terminal addon).
+No automatic backups configured (`backup/info` → 0 backups).
+
+Local control via WebSocket: `wss://home.thelonglakelodge.com/api/websocket`,
+auth with `{"type":"auth","access_token":HA_TOKEN}`. Useful WS cmds:
+`config_entries/get`, `config/device_registry/list`, `config/entity_registry/list`,
+`matter/node_diagnostics{device_id}`, `lovelace/config{url_path:null}` /
+`lovelace/config/save`. Addon control via REST service:
+`POST /api/services/hassio/addon_restart {"addon":"core_matter_server"}`.
+</overview>
+
+<thermostats>
+## Nest thermostats are integrated TWICE
+
+The 8 Nest thermostats appear via two independent paths:
+- **Cloud (Google Nest)** — `nest` integration, entities `climate.*_google_nest`
+  (+ `climate.lodge_basement` = Basement Living Room) and per-zone
+  `sensor.*_temperature` / `sensor.*_humidity`. Reliable; cloud-dependent.
+- **Matter (local)** — `matter` integration via the `core_matter_server` addon,
+  entities `climate.*_nest`. Local control, but see incident below.
+
+8 zones: Office, Kitchen, Bunkhouse Kitchen (forced-air "HVAC"); Eat-in Kitchen,
+Primary Bathroom, Basement Living Room, Basement Bedroom, Tub Room (in-floor heat).
+Matter fabric id `1E5211CA4C7B5275`, node ids 1–5,7,8,9 (temp sensors are bridged
+endpoint-101 sub-devices of their thermostat node).
+
+`script.set_away_temperatures` (HVAC 55–78 / in-floor 55) and
+`script.set_occupied_temperatures` (HVAC 68–76 / in-floor 67) set all zones,
+targeting by **label**: `hvac` (3 forced-air) and `in_floor_heat` (5 in-floor),
+both applied to the cloud `*_google_nest` devices (resolved in-script via
+`label_devices(..) | map('device_entities') | ... | select('match','climate')`
+since the labels are on devices, not entities). Commands are **serialized one
+zone at a time with an 8 s delay** — firing all 8 at once trips the Google Nest
+SDM per-user `ExecuteDeviceCommand` rate limit (429 RESOURCE_EXHAUSTED). (Matter copies carry parallel
+`matter_hvac` / `matter_in_floor_heat` labels — used by the scripts before the
+2026-06-15 repoint to cloud.) `google_nest` labels all 8 cloud devices.
+</thermostats>
+
+<matter-outage>
+## Matter fabric loss (2026-06-15)
+
+At 2026-06-15 13:53 UTC an HA Core/Supervisor restart (Supervisor → 2026.06.1)
+left the `core_matter_server` addon (v8.5.0) with an **empty fabric** — all 11
+Matter devices report `node_not_found`, all 59 `climate.*_nest` etc. entities
+`unavailable`. Integration reconnects fine (config entry `loaded`); the server
+simply has zero nodes. Addon restart + config-entry reload do NOT restore them
+(storage lost, not transient), and there is no backup to restore from.
+
+Cloud Nest is unaffected → thermostat control intact via `*_google_nest`.
+
+Recovery = re-commission each thermostat on-site (non-destructive, no factory
+reset needed — Nest supports Matter multi-admin):
+1. On lodge WiFi, Google Home app (lodge account) → tap thermostat → Settings →
+   "Linked Matter apps & services" → "Link apps & services" → generate a Matter
+   setup code (11-digit / QR). Code is single-use, ~5 min TTL.
+2. HA → Settings → Devices & Services → Matter → Add device → enter the code.
+   Wait ~1–2 min; node appears (bridged temp sensor comes with it).
+3. Repeat per thermostat (one code at a time). Rename to match.
+4. Delete the stale unavailable Matter devices afterward.
+Then re-point dashboard climate tiles back to `climate.*_nest` if local control
+is wanted (see dashboard note).
+</matter-outage>
+
+<dashboard>
+## Main Lovelace (`lovelace` / "Overview", storage mode)
+
+Views: `main` (single grid card, columns:1, headed sections: locks/lights,
+Helpful Scripts, Water Sensors, then climate as three 2-col thermostat-card
+grids — **Lodge HVAC** (office, kitchen), **In-Floor Heat** (5 in-floor zones),
+**Bunkhouse** (bunkhouse kitchen)),
+`Environmentals` (history page), and `Lock Management`. Edit via
+`lovelace/config` / `lovelace/config/save` (WS).
+
+HVAC climate tiles point at the **cloud** `*_google_nest` entities (repointed
+off the dead Matter `*_nest` during the 2026-06-15 outage). The `Environmentals`
+page is three per-category vertical-stacks (Lodge HVAC / In-Floor Heat /
+Bunkhouse), each with Temperature, Humidity, and Heating/Cooling history-graphs
+(72h) scoped to that category's entities.
+If Matter is re-commissioned and local control is preferred, repoint these back.
+</dashboard>