You were right that HA uses the Shelly LOCAL API, not MQTT. Adds that path
(user chose "both") — control + status over http://<ip>/rpc, no MQTT setup on
the device:
- ShellyRpc (POST /rpc), ShellyHttpDriver (Switch/Light.Set, Reboot).
- ShellyStatusApplier: one shared apply path (normalize + input roles +
monotonic upsert + broadcast) reused by BOTH the MQTT ingest and the HTTP
poll, so transports can't drift. IngestShellyMessage refactored onto it.
- ShellyLocalOnboarder: probe an IP → GetDeviceInfo/GetStatus → create an
http-protocol device with its entities (reuses an MQTT-onboarded row by id,
no duplicate). AssignDevice uses it when a discovered Shelly is reachable;
falls back to MQTT-style if not.
- Manual "Gerät hinzufügen" modal (add by IP). shelly:poll scheduled every 10s
+ a re-poll after each command (PollShellyDevice) for near-live status.
driverFor picks http vs mqtt by protocol.
- Normalizer now drops housekeeping components (sys/wifi/cloud/mqtt/ws/…) so
GetStatus doesn't create junk entities.
6 ShellyHttpTest cases (Http::fake). Suite 68 green, 12/12 clean.
LIVE-VERIFIED against the real Shelly 1 Mini Gen3 at 10.10.30.78: onboarded
over local API (protocol http), entities switch:0 + input:0, kept online by the
10s poll — no MQTT configured on the device.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Codex R15 on the previous fix:
- [P2] reclassifyInput derived the raw input level from the already-displayed
(inverted) contact state, so toggling inversion twice didn't restore the
original. The ingest + device page now carry the raw `on` inside the contact
state, so flipping role/inversion is fully reversible. +1 round-trip test.
- Latent: Device::isOnline() could return null (null demo, no last_seen) and
break the type hint / views. Now returns a strict bool.
Suite 58 green, 12/12 tabs clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Multi-agent review (5 confirmed) + Codex found real issues in the rework:
- [P1 security] Shared `shelly` ACL had readwrite +/rpc, so a compromised
device could inject Switch.Set into ANY other device's <prefix>/rpc. Now
read-only on +/rpc (receive own commands) + write only homeos/rpc (reply).
Broker-verified: a shelly publish to victim/rpc is denied; laravel's is not.
- [P2 security] `+` wildcard reaches reserved homeos/ring namespaces → bogus
device. Ingest now rejects RESERVED_PREFIXES (homeos/ring/$SYS).
- [P2 security] Unbounded auto-onboarding = DB-exhaustion DoS. Added a device
cap (homeos.mqtt.max_devices, default 250).
- [P2 correctness] Every Shelly `input` became a phantom window contact (wall
switches shown as windows, possibly inverted). `input` is now a generic
sensor; the user PROMOTES specific inputs to window/door contacts on the
device page (invert-aware), stored in config->input_roles and applied in the
ingest — this is the "assign contacts" flow the user asked for.
- [P3 ux] Motion pill read `active`; producer writes `on`. Now reads both.
- [P1 migrations] Dedup computed survivor keys once; 3+ duplicates could
collide on unique(device_id,key). Re-query per duplicate (ring + mqtt_prefix).
+9 tests (reserved prefix, cap, input generic/promoted/inverted, demo echo).
Live-verified: input published → onboarded as input → assigned window contact
via UI → appears on Fenster page, persists across messages. Suite 57 green,
12/12 tabs clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Addresses "I have to configure MQTT per device; HA just works":
- Shared device account `shelly` (one credential every Shelly uses) with a
SCOPED ACL (status/events/online/rpc for any prefix; cannot touch homeos/,
ring/ or $SYS). gen-passwd seeds it from MQTT_SHELLY_PASSWORD.
- Auto-onboarding: IngestShellyMessage creates the device on the first
recognizable component (sys/wifi/cloud noise ignored), so pointing a Shelly
at the broker is all it takes — no manual "Zuweisen". Partial unique index on
config->>'mqtt_prefix' + race-safe create (merge-dedup migration).
- Settings → Geräte-MQTT card: server, username, reveal/copy password + steps
(config/homeos.php, MQTT_DEVICE_HOST). Discovery "Zuweisen" now upserts by
prefix (names/rooms an already-onboarded device, no duplicate) and no longer
forces per-device creds.
- Per-device credentials kept as opt-in hardening: a "generate" button on the
device page (pattern %u ACL retained).
Live-verified: publishing as `shelly` to a new prefix auto-creates the device
with switch+power state; a sys topic creates nothing. Suite 46 green, 12/12
tabs clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Codex R15 on the Ring bridge flagged two issues:
- [P1] Concurrent first-sight messages (retained info + motion on bridge
startup) could create duplicate devices — no uniqueness on ring_id. Added a
partial unique index on (config->>'ring_id'); resolveDevice now creates and,
on the unique violation, re-fetches the winner (race-safe).
- [P2] The install-gate cache was never invalidated, so ingest was dropped for
up to 15s after install and devices kept being created for up to 15s after
uninstall. AddonService now forgets the shared cache key on install/uninstall.
+3 tests (unique-index guard, repeated-messages-reuse-device, cache
invalidation). Suite 43 green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Second half of the Ring integration (handoff §12):
- RingTopics + RingNormalizer (H3): parse ring/<loc>/<cat>/<id>/<entity>/state,
map ding/motion/contact/lock/battery; bridge status topic drives the addon
connection state. Defensive — unknown topics are ignored, never junk devices.
- IngestRingMessage (H4, single ingest path): auto-creates cloud-flagged Ring
devices on first sight, monotonic race-safe state upsert, broadcasts live.
Gated on the addon being installed (cached) so a still-running bridge can't
recreate devices after uninstall.
- Shared AppliesDeviceState trait: the monotonic upsert now lives once, used by
both Shelly and Ring ingest (MqttTest guards the Shelly path).
- Listener subscribes ring/#. ring-mqtt container (opt-in `addons` compose
profile) as least-privileged `ring` MQTT user (ACL: ring/# only); gen-passwd
seeds the account. "Cloud" badge on Ring devices in list + detail.
11 Ring tests (topic parse, normalizer, bridge status, auto-create, install
gate, unknown-topic guard, out-of-order). Suite 40 green; 12/12 tabs clean.
Live-verified: real MQTT ring/.../ding + info publish → cloud device created
with ding + battery state. (Real Ring OAuth login runs in the ring-mqtt sidecar.)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- metrics table + Metric model; metrics:sample command (scheduled every minute,
pruned to 24h) records MQTT throughput (a Redis counter the ingest jobs bump)
and host CPU load + memory% from /proc.
- Chart.js line-chart Alpine island (themed from CSS tokens, dual axis) on the
Host page shows MQTT/min, CPU load and memory %. wire:ignore so the 10s health
poll doesn't re-init it. Nav check 10/10 clean (0 console errors).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- AssignDevice derives the Shelly MQTT prefix from the mDNS service instance name
(the device id) instead of the slugified topic identifier (which carries the
_shelly._tcp suffix); non-Shelly devices get no prefix.
- IngestDiscoveryMessage broadcasts DeviceDiscovered only on first creation, not on
every retained/periodic re-publish.
- The generic Confirm modal can target its event at the opening component (->to),
so a shared event name can't be caught by an unrelated component.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Python discovery sidecar (zeroconf mDNS over host network) publishes findings
to homeos/discovery/<source>/<id> as the sidecar user. Compose service with
network_mode: host + NET_RAW. Verified live: it found real devices on the LAN
(a printer and Shellys) via mDNS.
- Listener also subscribes homeos/discovery/#; IngestDiscoveryMessage upserts
discovery_findings (preserving assigned/ignored) and broadcasts DeviceDiscovered
on the private discovery channel.
- "Neue Geräte" page lists findings live with Assign (modal → creates a Device
and links the finding) and Ignore/Restore. Per-device broker credentials are
provisioned at onboarding (noted in the assign hint).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two concurrent workers could both read the old last_seen_at, pass the in-memory
comparison and save in reverse order, rewinding presence. The guard now lives in
the WHERE clause of a single conditional UPDATE, so last_seen_at only ever
advances even under concurrent ingestion.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- ACL: drop the shared `shelly` account + `+` wildcard (allowed cross-device
spoofing/control). Devices now authenticate with per-device credentials
(username = topic prefix, provisioned at onboarding) bound to their own prefix
via `pattern %u`. gen-passwd.sh creates only laravel + sidecar.
- last_seen_at is set from the message receive time (observed_at), monotonically,
so a delayed/retried/stale ingest job can't mark a device online incorrectly.
- Device::isOnline: a real device with no last_seen is OFFLINE (never connected);
only demo devices are assumed reachable. Added `demo` flag + presence tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
With multiple Horizon workers, ingest jobs for one entity can finish out of
order and overwrite newer state with older. The listener now stamps each message
with a µs receive time (observed_at); IngestShellyMessage applies state only when
the incoming message is newer (race-safe via a conditional update + unique guard),
and broadcasts only when applied. Added a feature test for the ordering guard.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Real bidirectional MQTT so devices are live, not mock (handoff §13.3):
- Mosquitto 2 broker (auth + per-client ACLs for laravel/shelly/sidecar from day
one); passwd generated by docker/mosquitto/gen-passwd.sh (gitignored).
- mqtt-listener daemon: subscribes `+/status/#`, parse + dispatch only (H2),
exponential reconnect backoff, graceful SIGTERM. php-mqtt/laravel-client.
- Ingest path (H4): IngestShellyMessage resolves device by mqtt_prefix, upserts
device_states, refreshes last_seen, broadcasts DeviceStateChanged
(ShouldBroadcastNow) on the private `home` channel.
- Control path (H1): DeviceDriver contract + ShellyMqttDriver (command topic +
Shelly.Reboot RPC) behind DeviceCommandService, which audits every command to
the new `commands` table. Device detail toggles + restart route through it;
flash reflects the real result.
- Live UI: dashboard + device pages listen via Echo (#[On('echo-private:home,
.DeviceStateChanged')]) and re-render instantly.
- Vendor specifics isolated in Support/Mqtt + Support/Drivers (H3).
Verified end-to-end in a real browser: publishing an MQTT status turned a light
"An" on the dashboard in 3.0s with no reload, 0 console errors. R12 30/30;
15 feature tests green (incl. ingest + command audit). README/bootstrap document
the broker passwd step.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
HomeOS Bootstrap