The overview had its own collapsible panel with a handle button and swipe
handling. On phones it now renders inside ha-resizable-bottom-sheet in its
persistent mode, so it resizes like the automation editor's sheet and stops
at a strip with the tabs instead of closing. The sheet reports its height
for the map's insets and fit padding; wide layouts keep the floating panel.
A persistent sheet stops at a pixel minimum instead of closing when dragged
down, can open at its content height within a configurable share of the
viewport rather than at least 55%, reports its height after opening and
resizing, and lets a host shrink the grab area below the handle for content
that sits right under it. The map overview uses all of this on phones; the
automation editor keeps the previous behavior.
The overview rendered its own timeline items. It now uses ha-logbook-entry
in its narrow layout, so activity reads like the logbook everywhere: the
person, the zone they are in, and when. Arrivals and departures on the zone
tab are the same rows for the person who moved. The card hands hass to the
overview only for these rows.
The window seeding, unavailable skipping and zone transition rules moved
into pure functions with tests, so a regression in the People or Zones
timeline no longer needs manual checking.
The bottom inset that lifts the map controls above the drawer only applied
on phones, so on desktop a tall drawer covered the scale ruler. ha-map gains
left and right insets for its bottom controls, and the card reserves the
drawer's width on the side it occupies.
On phones the drawer body is capped at 60vh and clips its overflow for the
collapse animation, so activity entries beyond that were unreachable. The
activity box now scrolls within the drawer.
Only a selection fit knew about the drawer; the first fit and every
auto-fit used the whole viewport, so markers could land under the drawer.
ha-map takes a fit padding that automatic fits apply, and refits when it
changes, so the map settles once the drawer reports its size.
A failed history request, or an instance without the history integration,
read as no recent activity; the panel now says the activity could not be
loaded. The first history sample is the state at the window's start, not
a change inside it, so a person who has not moved for days no longer shows
that old state as recent activity: samples before the window only seed
the previous state, for zones and people alike.
Focusing a person or zone fitted the full viewport, so the marker could
land under the overview: the start side on desktop, the bottom sheet on
phones. Fits now take viewport padding, both engines apply it per fit,
and the card passes the overview's size as padding on the side it covers,
honoring right-to-left layouts.
An explicit fit that had to wait for the engine or for a size did not
pause auto-fit until it ran, so an auto-fit queued in between could take
its place. The pause comes first now.
The overview's selected tab could be an empty, disabled one, showing an
empty list while another tab had content and leaving aria-selected on a
disabled button; the selection falls back to the first tab with items,
which the tab stop already did.
The card kept its entity color subscription on whichever connection it
first saw. It now remembers that connection and subscribes anew when
hass brings a different one, as ha-map already does.
When the map card is the single card of a panel view it shows a
floating overview with People, Devices, and Zones tabs styled as an
animated segmented control; on mobile it becomes a bottom sheet drawer
that can be collapsed by tap or swipe. People list their zone or Away
state with the last update time, devices only appear when they have a
location, and zones show their occupant count while passive zones are
hidden in this view.
Selecting a row or map marker opens a detail view with a back button,
a name that opens the more info dialog, and a 24h activity timeline:
zone changes for a person or device, arrivals and departures per person
for a zone. Selecting focuses the map (street level for people, the
radius for zones, shown only while selected), marks the selected marker
with a thicker border, shows the accuracy circle when the position is
imprecise, and clicking the map deselects. In panel layout the overview
sits at the start side, so the zoom control and map buttons move to the
other side: ha-map gains a zoom-position property for that, and honours
--ha-map-bottom-inset so the attribution and scale stay clear of the
bottom sheet.
The zoom control sits at the physical top-left in either text direction,
but the alert used logical insets, so in RTL its clearance moved to the
right and it covered the control.
The card consumed the theme through hass, whose update gate only lets a
dark mode flip through, so switching between two light or two dark themes
kept the old palette. It now consumes the themes from the UI context, like
ha-map, and rebuilds its entity and trail colors whenever they change.
The map card rebuilds its entity colors and trail colors when the theme
changes, ha-map redraws editable circles on a theme change, and the
locations editor rebuilds its cached marker elements. A focused cluster
bubble keeps focus on the bubble that replaces it, and a caller naming a
marker by aria-labelledby keeps that name.
The trail colors are memoized with the registry entries as a key, like the
marker colors, so trails recolor with their markers when the registry
arrives or changes.
The color module kept its own shared registry subscription with listener
counting and connection takeover. The lazy fullEntitiesContext already does
that for the whole app, so the module is now pure: it turns the registry
entries into a palette index, memoized per registry update. Maps ask for
the context only once they draw entities, so an editor's map, as in
onboarding, never requests the registry; the map card and the zone page
consume it directly.
The card kept its own color dictionary, so an entity could have a different
color on every map. It now uses the palette that ha-map and the zone editor
already share, ordered by registry creation; a configured color still wins.
Bubble avatars only carried the label text and picture, so an entity set to
the icon label mode showed initials once clustered and the attribute mode lost
its unit. The icon flag and unit now travel with the cluster data.
The doubled place labels are a tile issue, not part of the marker redesign,
and the population filter did nothing on the OSMF tiles, which fill in a
default population for every place. It moves to its own branch.
The zone page's updated() override never called its base, so once
willUpdate stopped misusing super.updated the subscribe mixin had no hook
left to run hassSubscribe when hass arrives after connection; the override
calls super.updated now.
Saves for one zone run in order, each looking the entry up when it starts,
so a second response can replace what the first produced, and a failure
only drops the pending values its own request carried. A passive zone that
is not drawn no longer anchors a cluster bubble. A map whose entity list
empties releases the color subscription, not only one whose list is unset.
A map whose entities were unset kept its registry subscription, holding
the shared stream open for nothing. It is released along with the
markers, and the three release sites share one helper.
The listener set deduplicated a reused callback while a separate counter
counted every call, so the two could disagree and the shared registry
stream could leak or be released early. Each subscription is now its own
listener, the count is the set's size, and unsubscribing twice is a no-op.
Marker and trail colors are resolved from theme variables when drawn, but
a theme change only swapped the base style, so the old palette stayed on
the markers until something else redrew them. Any change to the themes
now redraws entities and trails.
The color assignment gets unit tests: creation order per domain, ties
broken by registry id, the home zone kept out of the palette, muted
passive zones, a stable fallback for entities outside the registry, and
one shared registry stream released with the last subscriber.
An entity object without a color reached the accuracy circle's paint
with an undefined color; the map's own palette color fills in. The zone
page's markers are memoized on their data, while their colors come from
theme variables, so a theme change now bumps the color version and
recolors them at once.
Clusters are avatar bubbles now, so the styles for the old count badge
had nothing left to style. The shared entity color subscription is a
module-level singleton keyed to whichever connection subscribed first; a
map arriving with a different connection now takes the stream over
instead of reading colors from the old one.
A failed save dropped the pending edit and rethrew, which no caller
awaited, leaving an unhandled rejection and a marker that silently moved
back. The page now shows the error in a dialog instead.
The selected ring's color came from an inline outline-color that beat the
rule's outline shorthand by precedence, which reads as if the shorthand
won; the inline style now sets a variable the rule consumes, with the
primary color as fallback. A map whose connection context changes
resubscribes to the entity colors instead of staying bound to the old
connection. The zone page's willUpdate called the wrong lifecycle hook on
its base class.
Every map with a connection subscribed to the entity registry for the
creation-order colors, including editor maps that never draw an entity.
Onboarding's location picker is one, and its mocked server rejects the
registry request, failing the onboarding end-to-end tests. The
subscription now starts when a map first draws entities.
The engine used to guess at the host's save state: after a drag it
skipped one update that matched the pre-drag values, taking it for a
re-render with data the save had not yet replaced. That guess is wrong
when a save fails, since the old values are then the truth and the
marker stayed at an unsaved position. The engine now applies every host
update except while a drag is in progress.
The zone page, which knows when it is saving, keeps the dragged values
as pending edits and shows them until the saved data carries them; a
failed save drops them and the marker returns.
The engine sets the move cursor on the elements it actually drags, so the
zone marker and the named icon no longer carry a class for it, and the
editor's element cache no longer varies with editability.
Markers get a card-colored frame with a soft shadow instead of a thin
border in the entity color, and the selected marker gets a primary-color
ring outside that frame. With history trails shown the frame still takes
the entity color, so a marker can be matched to its trail, and the
markers redraw when trails appear or disappear.
Zones are drawn as colored circles containing the zone icon, using the
entity's color with the zone radius and GPS accuracy circles matching
it. Marker clusters render as a bubble of up to three rounded square
avatars with a +N chip; when all members are in the same zone and the
bubble would overlap that zone's circle, it floats above the circle,
pinned to the zone's coordinate, with a tail pointing at it. Individual
markers are rounded squares, and while history trails are shown bubble
avatars get borders in their entity color so trails can be matched.
Avatars in bubbles are clickable and open the entity, and the whole
bubble zooms in on its members.
ha-map gains per-entity hide_radius, hide_accuracy, and selected
options for hosts that manage a selection; ha-entity-marker exposes its
background and border width via CSS variables and a selected state with
a thicker border. The OSMF vector tiles carry some places twice (node
and boundary centroid); towns and larger now require a population,
keeping the node, and smaller places get a collision padding.
setMarkerAccessibility only filled in what was missing and nothing removed
it, so a cached element rebuilt with a new title or a different role kept
the old attributes. The helper now records what it set, replaces exactly
that on a rebuild, and clears it when the element leaves the map, while a
caller's own attributes stay untouched.
The flat meters-to-degrees conversion broke down for large circles near
the poles, wrapping longitude and producing latitudes beyond 90. Handle
placement and the polygon now use the great-circle destination, and the
fitted bounds a spherical box that spans all longitudes once the circle
reaches a pole, so all three agree with distanceMeters everywhere.
Every editable marker was a focusable button whose activation fired an
event, whether or not the host did anything with it: the zone page and
the location selector do not. Activation is now opt-in per location; the
onboarding search results opt in, and a marker that does not act on it is
draggable without button semantics. The engine contract separates taking
pointer input from being a keyboard-focusable button for that.
Marker titles and handle labels are localized when drawn, so a language
change, or localization arriving after the first draw, redraws them.
It only re-selected the row the user had just clicked, after the shared
request-selected helper had reset it, which fired the selection event a
second time and zoomed to the zone twice. The row highlight comes from
the list itself and nothing else read the value, so the list no longer
mirrors a selection at all.
Drawing, fitting, and handle placement each converted meters to degrees
on their own, with different pole guards, so above 89.4 degrees the
resize handle left the drawn circle. They now share one conversion that
also agrees with distanceMeters; only the pole itself is guarded.
Arrow keys on the resize handle also reached MapLibre's keyboard handler
and panned the map while resizing; the handle keeps them.
The notice followed a map that fills the editor's fixed height, so it
landed outside the box: clipped in onboarding, spilling into the form in
the location selector. It now sits over the top of the map, clear of the
zoom control.
The resize handle was the one marker whose click reached MapLibre's map
click, which a pin-on-click host takes as a request to move the location
there; the handle stops it like the other markers do.
The notice appears after the engine has been chosen, so helper text
would not reach a screen reader; it is an alert now. The engine test's
frame stub returned an id that landed in the engine's pending frame after
the callback had cleared it, so every redraw after the first was skipped.
The Leaflet fallback draws named location icons statically, but the
stylesheet still gave the ones marked editable a move cursor. The engine
sets that cursor on the elements it actually drags, so the rule goes.
The fallback notice blamed the browser, while MapLibre can also be
unavailable after a failed start or a lost WebGL context; it now
describes what the map offers. The resize handle's east point only guards
the pole itself, so its distance agrees with distanceMeters wherever a
zone can be.
The move cursor came from stylesheets keyed on whether a location may be
edited, so it showed on circles that cannot move and on the static
Leaflet fallback. The MapLibre engine now sets it on the elements it
actually made draggable and clears it when handing them back.
The fallback notice said the browser cannot edit locations, while
clicking the map to place one still works there; it now names what is
missing, dragging markers and resizing zones. The editor's
internationalization context is reactive, like its neighbours.
Arrow-key resizing changed the radius without arming the pre-edit guard
that pointer drags use, so a host re-render with the old radius while a
key was held, or its echo right after the commit, snapped the circle
back. A key resize now counts as a drag from the first key press until
the commit.
MapLibre leaves its positioning class and transform on a marker element
after removal. Hosts like ha-locations-editor cache their elements and
hand them to whatever engine draws next: after a fatal WebGL failure that
is the Leaflet fallback, which shows the locations statically. A removed
element, and every element still on the map when the engine is destroyed,
is reset so it arrives as a plain element. The editable circle's center
gets the same treatment.
Custom sources and layers are now recorded when added and only inserted
into the map once the style has loaded. A circle moved while a swapped
style was still loading therefore reaches the record, and a swap in that
window carries it over; the deferred insert skips what the swap already
placed.
Clicking or dragging a zone on the map, and saving one in the dialog,
selected its row in code. That selection fires the same request-selected
event a click ends with, so the page had to tell the two apart, and the
row could not be shown selected and stay clickable at the same time: the
highlight never actually appeared. The list now only follows the user's
own clicks, which removes that bookkeeping along with the zoom that a
radius drag used to trigger.
The MapLibre engine's editable circle against the fake map: dragging the
center moves the circle and the handle and reports the drop, dragging the
handle resizes and snaps it back, arrow keys resize in steps committed on
release, the advertised maximum follows a larger radius, one update
echoing pre-drag values is ignored, the center activates by click, Enter
and Space but not after a drag, and removal clears markers, layers and
listeners.
ha-map's reconciliation of editable locations through a fake engine with
editing support: circles and markers are drawn, moved in place, rebuilt
when their appearance changes and removed when dropped, move, resize and
activation events carry the location id, and an engine without editing
support shows the locations statically.
Editable markers, the circle center on both engines, and the static
fallback center activate on Enter and Space, as their button role
promises. Locations without a name get a localized one, since onboarding
and the location selector pass none. The resize handle's advertised
maximum follows the radius instead of being exceeded by it, and the
handle stays near the circle at high latitudes.
Clicking the active zone row zoomed to it again upstream but was
swallowed by the check that told a map-driven selection from a click; a
one-shot flag on selections made in code replaces that inference. The
engine rebuilds its sources from its own record on a style swap, so a
radius changed while the style was unloaded is not lost.
ha-locations-editor (zone config, the location selector, onboarding)
used raw Leaflet layers and leaflet-draw, which pinned it to the Leaflet
engine. Its interaction surface is three gestures - drag a marker, drag
a zone center, drag a zone radius handle - so no editing library is
needed.
MapEngine gains an optional editing capability (draggable markers and an
addEditableCircle primitive) plus panTo and containsLocation. Only the
MapLibre engine implements editing: the circle geometry plus a draggable
center marker and a draggable radius handle on the east edge, the radius
derived from the great-circle distance between them, redrawn per frame
while dragging. The handle has a touch-sized hit target and is a
keyboard-operable slider with an accessible name. The Leaflet engine
stays a viewing fallback: without WebGL2 an editor shows its locations
as static markers and a notice that this browser cannot edit on the map.
ha-map exposes editableLocations and fires
editable-location-moved/resized/clicked and editing-available-changed.
Locations are reconciled by id: position and radius changes move the
existing handles, while a changed icon, name, color, or editability
rebuilds the marker, so edits made in the form fields show up on the
map on either engine. ha-locations-editor becomes a thin layer over
that with its public contract unchanged, so its consumers need no
changes; it reads localize from the internationalization context instead
of taking hass.
The zone panel no longer zooms the map when a selection comes from the
map itself. Clicking or dragging a marker selects its list item, and
mwc-list reports that programmatic selection with the same "property"
request-selected event a click ends with. The two are told apart by
timing: a selection from the map has already set the active entry when
its event arrives, while a click's event arrives before the handler
sets it. Skipping the programmatic event also keeps the item selected,
which the shared request-selected helper would otherwise reset.
The layers property and the engine selector are removed along with the
leaflet-draw dependency and its patch: Leaflet is now purely the viewing
fallback, selected by WebGL2 support alone. The Leaflet engine keeps a
leafletMap field for the ha-map fit tests, which run on it in jsdom.
Prevent scrim close on backup restore dialog
Clicking outside the restore dialog dismissed it unless the encryption
key field was dirty. Restoring a backup is destructive, so guard the
dialog at every step.
Claude-Session: https://claude.ai/code/session_017eZQ6BTz6MervuG3Zb8RLp
Co-authored-by: Claude <[email protected]>
* Add row targets UI to entity state and numeric state in card condition editor
* Use automation descriptions instead of target ui
* Type check
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Format condition editor
* Fix contextual condition descriptions
* Resolve state references in condition descriptions
* Share automation condition headers with visibility conditions
---------
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Run ha-map on a runtime-selected engine: MapLibre GL native or Leaflet
ha-map no longer drives Leaflet directly. All primitive operations -
camera, HTML element markers, meter-radius circles, history paths,
clustering with a pluggable icon builder, scale ruler, dark mode - go
through a MapEngine interface, and the engine is selected at runtime:
- MapLibreMapEngine renders the vector base map natively where WebGL2 is
available, without Leaflet in the loop: continuous fractional zoom,
DOM markers synced to the basemap every frame, GeoJSON zone and
accuracy circles below the labels, GeoJSON history trails with hover
popups, and a pixel-distance cluster grid recomputed when the camera
settles. Dark mode swaps the style while carrying the custom layers
over, and rotation and pitch stay disabled for north-up dashboards.
- LeafletMapEngine wraps the existing behavior unchanged (vector tiles
through the maplibre adapter with the raster fallback, markercluster)
for browsers without WebGL2 - the legacy floor includes iOS 12-14 and
kiosk browsers without hardware acceleration - and for
ha-locations-editor, which manages raw Leaflet layers with
leaflet-draw and declares engine=leaflet.
A permanently lost WebGL context rebuilds the map on the Leaflet engine
instead of leaving a dead canvas. Zoom levels keep Leaflet semantics
across engines. MapLibre's stylesheet is shipped to /static/map for the
native engine's controls and popups.
* Enhance keyboard accessibility for icon click
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Fix dark mode style application logic
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Address code review on the map engine
A fatal engine event during setup was dropped because the in-flight
setup owned the loading flag; the fallback request is now recorded and
setup switches to Leaflet once init settles. Entity markers accept Enter
and Space, as both engines make them focusable buttons. History points
show their popup on tap as well as hover, since touch has no hover.
Tests cover the engine choice and fallback paths, including a fatal
event mid-setup, and the MapLibre engine itself against a fake map:
style swaps, queued layer work, custom layer carry-over, failed style
requests, token refusal recovery, context-loss grace, and clustering.
* Address second code review round on the map engine
A cluster whose members share a spot, or that sits at maximum zoom,
could never be opened: fitting its bounds changed nothing and the next
regroup recreated it. Activating such a cluster now opens it: all its
members are shown in a bubble with a tail pointing at their spot, each
reachable on its own, until the map moves again.
An engine whose init failed, or was abandoned by a disconnect or a fatal
event during setup, was never destroyed and could keep a WebGL context;
setup now destroys any engine that did not become the active one.
The Leaflet engine sizes the cluster element like MapLibre does, so
cluster bubbles no longer shrink to their text.
* Keep the engine's own record of its map sources and layers
Carrying zone circles and history paths over a style swap relied on
MapLibre serializing the outgoing style into transformStyle's previous
argument. The engine now keeps the source and layer specifications it
added and rebuilds them into the new style itself, so the carry-over no
longer depends on what MapLibre hands over, and a missing previous style
cannot drop them.
* Address review: setup teardown, marker input, cluster focus
An engine still setting up could not be torn down: ha-map only held it
once init resolved, so removing the element mid-load kept its WebGL
context alive and left the loading guard set, ignoring a reconnect. The
engine being set up is now tracked and destroyed on disconnect, a stale
setup notices it was superseded, and the MapLibre engine settles a
pending init when destroyed.
Non-interactive markers let pointer input through, as they do on
Leaflet. Opening a cluster from the keyboard moves focus to its first
member, and a member that has focus when the bubble closes hands it to
the icon that replaces it.
* Tear down an engine that fails while still setting up
A fatal event during setup only flagged the fallback and waited for
init to settle, which relied on MapLibre still firing style.load. The
engine being set up is destroyed instead, which settles its init, so the
setup hands over to Leaflet without depending on the failed engine.
* Make history point popups readable on MapLibre
MapLibre's stylesheet is linked into the map's shadow root and wins over
the component styles on equal specificity, property by property: its
white popup background applied while our white text color did too,
leaving the timestamp invisible. The popup rules now carry !important,
as the Leaflet tooltip rules already do for the same reason.
---------
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Add show_values toggle to energy and power Sankey cards
Adds a boolean config option that renders each node's value directly
on its diagram label (e.g. "Solar\n3.4 kWh"), stripping HTML from the
valueFormatter output for ECharts plain-text labels.
Fix show_values label alignment on Sankey nodes
Collapse internal whitespace in valueFormatter output after stripping
HTML tags, so "7.7\n kWh" becomes "7.7 kWh" on a single line.
* Address review feedback on show_values
Removes the HTML-tag-stripping regex from the label formatter (an
incomplete-sanitization pattern flagged by CodeQL); the value
formatter never returns HTML and the label renders as plain ECharts
canvas text, so no stripping is needed. Splits the vertical
label-sizing measurement on all whitespace instead of literal spaces,
so the newline-separated value line is measured correctly. Scopes
show_values to the energy and power Sankey card configs instead of
the shared Sankey base, and hides the option in the card editor for
water and water-flow cards, since those cards don't support it.
* Apply batched suggestions from code review
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Format Sankey card templates with Prettier
Restore Prettier wrapping around the show_values template bindings so CI lint:prettier passes.
* Fix vertical Sankey value labels wrapping on thin nodes
Unicode isolates around show_values text were treated as wrap characters, and font sizing ignored ECharts minMargin, so short names like TV with small values split mid-word. Size labels to the inner wrap width and keep LTR with a mark instead of isolates.
* Reserve extra bottom space for vertical Sankey values
When show_values is on, node labels can be three lines (name, number, unit). Increase the vertical series bottom inset from the single-line 25px so the lowest row is not clipped.
* Keep locale grouping spaces in Sankey value labels
Trim formatter output without collapsing internal whitespace, so narrow no-break thousands separators are not turned into wrap points. Measure the longest word on regular spaces and newlines only.
---------
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
Co-authored-by: Cursor Agent <[email protected]>
* Add a Modbus panel
The Modbus integration exposes the connections it keeps open over
`modbus/connections/list`. This surfaces them under Connectivity, mirroring the
Serial panel: one entry per physical device, grouped by whether the link is up,
naming the config entries holding units on it.
Because one device is shared by every integration talking to it, the panel is
where a shared RS-485 bus or gateway becomes visible — the two integrations on
`/dev/ttyUSB1` cannot be seen from either integration's own page.
The Serial panel gains the other half of that: a port carrying a Modbus
connection now points at the Modbus panel. The two are matched on the device
path, which Modbus stores as configured, so both the scanned and the resolved
path are compared. Reaching Modbus is not required for the Serial panel to
load.
Demo fixtures come with it, so the connections and the shared bus can be seen
without a Home Assistant instance. The RS-485 port the Serial demo already
lists gains the two integrations that hold units on it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FjmjZsxhdj9ekGmhFHXg3p
* Modbus: show link state as an indicator, not a verdict
A closed link is not a fault. The library opens one when an integration next
polls, and a device is free to drop an idle one in between, so grouping the
connections into Connected and Disconnected cards behind a warning icon
reported trouble where there was none.
One list now, ordered by address alone, with a per-row indicator. A link
opening or closing no longer moves a row between cards, and the summary is a
count rather than a health verdict.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FjmjZsxhdj9ekGmhFHXg3p
* Modbus: a serial port is open, not connected
Opening a serial port proves a device node exists; nothing on the wire has to
answer. Only a network connection performs a handshake, so "Connected" was
promising more than a serial row can know. Those rows read Open and Closed.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FjmjZsxhdj9ekGmhFHXg3p
---------
Co-authored-by: Claude Opus 5 <[email protected]>
* Add dashboard visibility condition classifier, translator and splitter
Pure-logic foundation for delegating dashboard visibility conditions to
core (#52836).
- Add a VisibilityCondition type spanning the client-only lovelace
conditions (screen, user, view_columns, location, time) and core
automation conditions, alongside the existing lovelace Condition.
- translate.ts: classify conditions as client- or server-evaluated and
translate server ones to core format (entity -> entity_id, state_not
-> not-wrap, numeric bound coercion), preserving lovelace's
not = not(AND) semantics.
- split.ts: split a tree into maximal server subtrees (one subscription
each, sibling-grouped) plus a three-valued client combiner.
* Add reactive condition evaluator controller
Phase B of delegating dashboard visibility conditions to core (#52836).
- ConditionEvaluatorController opens one subscribe_condition per server
subtree, evaluates client leaves locally, observes screen/time
boundaries, and exposes a tri-state visible/hidden/unknown result plus
error. It recomputes on push/listener/hass/context change, debounces
re-subscription when the tree changes, and tears down on disconnect.
- Add observeConditionChanges to listeners.ts (notify-only, decoupled
from checkConditionsMet) and widen extract.ts to the VisibilityCondition
tree, factoring time-boundary scheduling into a shared helper.
* Harden condition translation for incomplete and odd numeric inputs
Follow-up to the adversarial review of the translator (#52836):
- Incomplete/garbage state conditions (no entity, no value, or an empty
object) now translate to an always-false core condition instead of a
schema-invalid `state`, matching checkConditionsMet and avoiding a
broken grouped subscription.
- numeric_state bounds: coerce only finite numeric strings (incl. "" -> 0)
to numbers, pass genuine entity-id references through, and drop junk or
non-finite strings (matching lovelace's "NaN -> ignored" and never
emitting a non-JSON-serializable Infinity).
* Fix condition evaluator controller lifecycle edge cases
Follow-up to the adversarial review of the controller (#52836):
- Key re-subscription on a structural signature of the condition tree
rather than array reference identity, so a host re-deriving the array
each render neither starves the debounce nor churns subscriptions.
- Reset the published result to `unknown` on host disconnect so a
detached/reconnecting host never renders a stale, no-longer-live result.
- Read hass lazily in the time-boundary listeners so timezone changes are
picked up on the next boundary instead of being pinned at subscribe time.
* Evaluate dashboard visibility through the reactive condition evaluator
Rework ConditionalListenerMixin to derive visibility from
ConditionEvaluatorController instead of evaluating checkConditionsMet
synchronously. Stateful conditions (state, numeric_state, template, sun,
zone, device, integration) are delegated to core via subscribe_condition;
client-only conditions (screen, user, view_columns, location, time) stay
local. The mixin re-feeds the evaluator on connect and on hass/config/
column changes, and drives _updateVisibility from its tri-state verdict.
Consumers (hui-card, hui-badge, hui-section, hui-heading-badge,
hui-view-sidebar, hui-conditional-base) now read the mixin's
_conditionsVisible(), which prefers the server-aware verdict and falls back
to an optimistic synchronous seed while a server subtree is pending — exact
for legacy lovelace conditions (no flash for existing dashboards) and hidden
for core-only conditions until the server reports.
- fold the host entity_id context into the evaluator path via
addEntityToCondition, and read core-format entity_id in
checkStateCondition / checkStateNumericCondition so seed and server agree
- addEntityToCondition no longer grafts a context entity onto an
already-core condition that carries its own entity_id
- the conditional card/row now evaluates legacy {entity, state} conditions
- cache the entity-folded array so the evaluator's signature memo holds
across hass updates, and drop the cached verdict when the tree changes by
value so the seed is used for the new tree
* Add server condition types to the dashboard visibility editor
Let the visibility editor add and edit the core-format server condition
types (template, sun, zone, device) by embedding the automation condition
editors, which already speak core format. ha-card-condition-editor
dispatches these types to ha-automation-condition-editor; the existing
lovelace editors and the and/or/not containers are unchanged, and because
the logical editors nest ha-card-condition-editor, mixed trees dispatch
each child correctly.
- extend the add-condition menu with the new types (icons + labels)
- suppress the client-side live-test for server-class conditions (and any
logical tree containing one); checkConditionsMet can't evaluate them, so
the indicator stays neutral instead of showing a misleading failure
Server-backed live-test and read-both/write-new conversion of lovelace
state/numeric_state conditions are follow-ups.
* Evaluate the visibility condition editor live-test through the reactive evaluator
Drive the per-condition live-test indicator with the same
ConditionEvaluatorController the dashboard uses at runtime, so server-class
conditions (template/sun/zone/device and core-format state/numeric_state) get
a real subscribe_condition-backed verdict instead of a neutral indicator.
Client-only conditions stay evaluated locally and mixed logical trees combine
both via three-valued logic.
- Fold the card entity into the observed condition exactly as the runtime
mixin does, and memoize the folded array so the evaluator's signature memo
keeps hitting on hass-only updates.
- Map the evaluator verdict to the indicator: visible -> pass, hidden -> fail,
pending -> unknown, server error -> invalid (raw error shown as the tooltip
detail, localized label kept as the aria-label).
- Keep a client-side validity check for purely client trees so a malformed
client-only config still surfaces as invalid.
- Recurse the no-entity (filter-mode) suppression so nested entity-less
conditions are handled, and report an as-yet-unknown manual test as no
result rather than a failure.
- Drop the now-unused invalid-config alert and its orphaned translation keys.
* Evaluate the visibility status banner server-side
Drive the card-level visibility summary banner with the same
ConditionEvaluatorController used by the per-condition live-test, so a set
containing server-class conditions reports its real visible/hidden verdict
instead of being flagged as an invalid configuration.
- Add a distinct "unknown" banner state for while a server result is still
pending, separate from "invalid" (a genuine configuration error).
- Keep a client-side validity check for purely client trees, and fold the
card entity into the observed conditions, matching the per-condition editor.
- Extract isPureClientCondition (every leaf client-side, as opposed to
isClientCondition's any-leaf) so both consumers share one classifier.
* Edit dashboard state conditions via the core condition editors
Route `state` / `numeric_state` visibility conditions to the core automation
condition editors (outside entity-filter mode), so they share one editing
surface with the server-class types and are authored in core format.
- Read both: existing lovelace-format conditions (`entity`, `state_not`, …)
are translated to core for display; a `state_not` shows as `not(state)`.
- Write new with touch-to-convert: opening a condition leaves it untouched;
editing or adding one persists it in core format (`entity_id`, `state` list).
- Entity-filter mode keeps the lovelace no-entity syntax and editors.
- Register the core `state` / `numeric_state` editors (dynamicElement only
renders a tag, it does not define the element) and widen the editor chain's
condition arrays to the mixed visibility union.
* Remove the superseded client-only condition listeners
The dashboard now evaluates visibility through the reactive condition
evaluator (which uses `observeConditionChanges`), so the old synchronous
client-only listener path has no remaining callers.
- Delete `ConditionListenersController` and the `setupConditionListeners` /
`setupMediaQueryListeners` / `setupTimeListeners` helpers.
- Keep `observeConditionChanges` and the shared time-boundary scheduler, and
re-point their tests onto it so the scheduling edge cases stay covered.
* Pin the condition editor live-test for hidden and invalid configs
The per-row live-test set `invalid` (or hidden) directly but left its
evaluator callback unguarded, so the evaluator's torn-down `unknown`
result — fired ~500ms after observing `undefined` — clobbered the pinned
state until the next hass tick re-pinned it, causing a transient flicker.
Add the same `_override` guard the sibling visibility-status banner
already uses: the hidden / client-invalid branches set the result
directly and pin it, and the evaluator callback ignores results while
pinned.
* Remove the orphaned conditional-listener no-op methods
`addConditionalListener` / `clearConditionalListeners` became no-op
shims once the evaluator took over subscriptions and teardown; their only
callers were the listener-wiring removed in the migration. Drop both
methods, the mixin's now-redundant `disconnectedCallback` override, and
the stale `clearConditionalListeners()` call in hui-conditional-base.
* Accept server-evaluated conditions in validateConditionalConfig
The conditional card/row/element gate their config on
validateConditionalConfig, which only knew the client-side condition
types and rejected the server-evaluated ones (template, sun, zone,
device, and integration-provided conditions). Now that those are
authorable and delegated to core, accept them — core validates them —
so configuring one no longer throws "Invalid configuration".
* Evaluate conditional picture element visibility server-side
The picture-elements `conditional` element was the one visibility
consumer still evaluating fully client-side, so its stateful conditions
were not delegated to core (and server-class types it could not evaluate
fell through to a permanently-hidden result). Convert it to a
ReactiveElement driven by ConditionalListenerMixin, exactly like its
sibling hui-conditional-base, so the evaluator delegates stateful
conditions through subscribe_condition and evaluates the client-only
ones locally.
* Keep conditional cards mounted while hidden for server conditions
A conditional card gated on a server-evaluated condition (template, sun,
zone, device) never reappeared: hui-card removes a hidden child card from
the DOM, tearing down the evaluator, and the synchronous seed can revive a
client condition but not a server one, so the subscription was never
(re)opened and the server result that would show the card never arrived.
Set connectedWhileHidden so the conditional card/row stay mounted while
hidden and keep their subscriptions alive, like the other cards that must
keep working while hidden. The inner element is still unmounted when
hidden, so there is no extra render cost.
* Load config translations for the reused condition editors
The dashboard visibility editor reuses the automation condition editors
(state, numeric_state, template, sun, zone, device), which label their
form fields from the `config` translation fragment. The lovelace panel
never loads that fragment, so those labels rendered blank — e.g. the
numeric_state limit-type selectors and the above/below fields.
Load the `config` fragment when the conditions editor first renders, so
the embedded editors resolve their labels.
* Fold the card entity into entity-less conditions on read
A card with a host entity can carry entity-less state / numeric_state
visibility conditions that implicitly target that entity (folded in at
runtime). The editor translated them with an empty entity_id, so the
reused automation editor showed an empty, invalid entity field.
Fold the current-mode context entity into the displayed condition, and
recompute it when the entity context arrives (it can follow the
condition). Opening still does not rewrite the stored config; only
editing converts it to explicit core format.
* Demote the unused state/numeric_state condition editors to base classes
With the automation condition editors handling state / numeric_state in
the dashboard editor, the lovelace `ha-card-condition-state` and
`ha-card-condition-numeric_state` elements are no longer rendered. Their
only remaining role is as the base class for the entity-filter
`-no_entity` variants, so drop the now-dead custom-element registrations
(and the redundant side-effect imports) while keeping the shared logic.
* Format condition editors
* Fix condition editor indentation
* Invalidate the condition evaluator result as soon as the tree changes
When the observed tree changed by value, the controller kept the old
split and subscriptions alive until the debounced re-subscribe fired, so
it kept publishing (and the editor's Test action reported) the previous
tree's verdict, and a host that had already reset to unknown never got a
notification when the new tree settled on the same result.
Tear the old tree down when the signature changes and publish unknown
until the new subscription reports. Trees with nothing to subscribe to
re-evaluate immediately rather than waiting out the debounce.
Track a pending re-subscribe with its own flag: undefined is a valid
signature (observation cleared), so using it as the "nothing pending"
marker kept rescheduling the timer on every hass update and could leave
the previous subscription open indefinitely.
* Translate bound-less numeric_state conditions to a schema-valid core condition
Dropping a junk or non-finite bound could leave a numeric_state condition
with neither above nor below, which core's schema rejects; because sibling
server conditions are grouped into one subscription, that single invalid
leaf failed the whole group and hid the content. Lovelace merely ignored
such bounds and only required the value to be numeric, so express that
residual check as a template condition instead. An entity-less
numeric_state resolves to always-false, matching checkConditionsMet.
* Validate every visibility condition client-side and handle unsupported automation editor configs
validateConditionalConfig already accepts the server-evaluated types, so
gating it behind "every condition is client-only" only served to skip
validation as soon as a server condition was present; a malformed screen
condition next to a template condition was reported as hidden rather than
invalid. Run it for the whole set.
The embedded automation condition editors fire ui-mode-not-available when
an existing core config fails their UI struct. Fall back to YAML with the
struct warnings, as the automation condition row does, instead of
rendering a blank editor while still claiming UI support.
* Type dashboard visibility contracts as VisibilityCondition
Card, badge, section, view sidebar and heading badge visibility, plus the
conditional card, row and element conditions, still declared the
lovelace-only Condition type, so the newly supported template, sun, zone,
device and integration conditions were invalid at the config boundary and
needed casts throughout. Declare them as VisibilityCondition, widen
validateConditionalConfig accordingly and drop the casts that made
redundant. Entity-filter conditions stay lovelace-only.
* Subscribe when hass arrives after the conditions were observed
The subscribed signature was recorded before checking that hass was
available. A host that connected with its config set but received hass
later therefore never subscribed: the later observe saw the same
signature, found no split and stayed unknown for good. Only record the
signature once a subscription is actually opened.
* Seed pending visibility with a three-valued local evaluation
While a server result was pending, the mixin ran the whole tree through
the legacy client evaluator. Core-only leaves default to false there, so
not: [template] was inverted to visible until core replied, and core
state / numeric_state options such as for, match and value_template were
ignored, both contradicting the no-flash guarantee.
Evaluate locally only the leaves the legacy evaluator reproduces exactly
(client-only types, lovelace state / numeric_state, and core ones without
those options), keep every other leaf unknown through and / or / not, and
hide while the outcome still depends on an unknown leaf.
* Hide server-only condition types in entity-filter mode
The map and entity-filter cards still evaluate their conditions locally
with checkConditionsMet, where template / sun / zone / device fall through
to a failing state check and filter out every entity. Don't offer them in
the add menu for that mode.
* Leave conditions carrying enabled to the server in the local seed
Core treats a disabled condition as neutral (and enabled may be a
template), while the legacy evaluator ignores enabled altogether. A
pending not: [disabled failing state, ...] was therefore seeded visible
although core reports hidden. Treat any node carrying enabled as unknown.
* Reject filter-incompatible pasted conditions in entity-filter mode
The condition clipboard is shared across editors in the session, so a
template / sun / zone / device condition, or a core-format leaf pinned to
its own entity_id, could be pasted into a map or entity-filter editor,
where the local evaluator would filter out every entity or evaluate the
wrong one. Hide the paste entry and ignore the action for such clipboard
contents while in filter mode.
* Re-feed the section visibility evaluator after a strategy config resolves
_config is not reactive, and for a strategy section it is assigned after
the Lit update that last fed the evaluator, so _conditionsVisible kept
reading the verdict for the raw or previous config until the next hass
update. Re-run setupConditionalListeners right after assigning it.
* Hide content whenever a server condition subtree errors
An errored subtree was recorded as false and fed through the combinators,
so a client-side not inverted it and an invalid server condition could
show content; the runtime mixin only looks at the result, not the error.
Force the result to hidden while any subtree is in error.
* Honor enabled: false on visibility conditions
Core skips a disabled condition inside a compound: it neither passes nor
fails. Rebuilding logical conditions during translation dropped enabled
(and the other row metadata), so a disabled server-side group was
evaluated, and the mixed-tree split and the local seed ignored it as well.
Preserve the row metadata on translated logical conditions, skip disabled
nodes when splitting and when seeding, and leave a template-valued
enabled on a seeded node unknown since only core can render it.
* Type the conditional card conditions as VisibilityCondition
ConditionalCardConfig, the contract HuiConditionalBase actually consumes,
still declared the lovelace-only union, unlike the row and element
contracts.
* Keep a client-side node with a template-valued enabled unknown
Only core can render enabled: "{{ ... }}", and core never sees a mixed or
client-only node, so the split silently treated such a node as enabled
and let its children decide visibility. Keep it unknown instead, erring
toward hiding and matching the optimistic seed.
* Whitelist the condition types usable as entity filters
The filter-compatibility check only excluded the four built-in server
types, so an integration-provided condition copied from a visibility
editor could still be pasted into an entity-filter editor, where the
local evaluator cannot evaluate it and filters out every entity. Check
against the types checkConditionsMet supports instead.
* Fold the host entity into legacy entity-less conditions as well
A legacy condition such as { state: "on" } has no condition key, so
addEntityToCondition never stamped the card entity on it. That used to be
harmless because checkConditionsMet fell back to the evaluation context,
but the server translation has no context and turned the condition into
always-false once core reported. Treat the legacy shape as the state
condition it is, and make the helper generic over the visibility union so
callers no longer need casts.
* Keep the lovelace outcome for infinite numeric bounds
Number("1e400") is Infinity, which lovelace does not ignore: above +inf
or below -inf can never pass, while above -inf or below +inf always do.
Dropping such a bound could flip an untouched condition from hidden to
visible. Resolve the unsatisfiable cases to always-false and drop only the
vacuous ones. Infinity is not JSON-serializable, so neither is ever sent.
* Only leave a template-valued enabled unknown in the local seed
The check also matched enabled: true, so an explicitly enabled condition
was seeded unknown and the card hidden until core responded, exactly the
flash the seed is meant to avoid.
* Show no test chip while the condition evaluator reports an error
An invalid server condition evaluates to hidden plus an error; reading
only the result turned Test into a transient "not met" chip that
contradicted the invalid live indicator.
* Seed only core conditions the client compares identically
The optimistic seed still admitted core-format shapes the legacy evaluator
handles differently: an attribute comparison (core compares the raw value,
the client stringifies it) and an entity-valued numeric bound (core errors
when the entity is missing, the client ignores the bound). Leave those
unknown until core reports.
* Accept a single condition as the children of a logical condition
Core's LogicalCondition allows conditions to be one condition or a list,
but every traversal here assumed a list and crashed on the single shape.
Widen the type accordingly and route all traversals through one helper.
* Keep lovelace leaves with legacy-only semantics client-side
Core compares raw attribute values, only dereferences input_* comparison
values and errors on a missing bound entity, where lovelace stringifies
the attribute, resolves any existing entity and ignores the bound.
Delegating such existing conditions changed their visibility silently.
Classify a lovelace-format state / numeric_state leaf that relies on any
of those as client-side so it keeps its exact legacy evaluation; once the
user edits it, it is saved in core format and evaluated by core.
* Drop the pinned editor indicators when leaving an override branch
The empty and client-invalid branches pin the indicator and clear the
evaluator, which is then already unknown; when the condition becomes
valid the pending observation is unknown too, so no notification arrives
and the stale pinned state stayed visible until core first responded.
* Reject core's single-child logical shorthand as an entity filter
Filter consumers and the filter-mode editor expect the lovelace list
shape for logical children, so a pasted core condition using the
single-child shorthand caused a runtime error there. Require a list in
the filter-compatibility check and traverse defensively when looking for
no-entity conditions.
* Seed only core leaves whose entity is present and, for numeric_state, bounded
Core errors on a missing entity and rejects a bound-less numeric_state,
while the client evaluates a missing entity as unknown and passes any
numeric value, so both shapes could flash visible before the subscription
error arrived. Leave them unknown until core reports.
* Re-subscribe when the hass connection object is replaced
Subscriptions are bound to the connection they were opened on. An
unchanged tree observed with a hass carrying a different connection kept
the stale subscriptions; re-open them in that case while still not
churning on ordinary hass updates.
* Keep non-finite bounds distinguishable in the condition signature
JSON.stringify serializes Infinity, -Infinity and NaN alike as null, so
flipping a YAML .inf bound produced the same signature and the old
subscription stayed active.
* Seed lovelace leaves only while their target entity exists
Core reports an error for a condition on a missing entity, which hides the
tree, while the legacy evaluator compares against the literal unknown.
Seeding such a leaf optimistically could therefore flash the content
before the error arrived; leave it unknown instead.
* Do not seed a core state condition that compares against another entity
checkConditionsMet dereferences any existing entity used as a comparison
value, while core only dereferences input_* entities, so such a core leaf
could flash visible before the subscription reported false. Apply the
same entity-reference guard the legacy classification uses.
* Preserve row metadata when rebuilding lovelace leaves
Both the evaluation translation and the editor's display conversion
rebuilt lovelace state / numeric_state leaves without enabled, alias and
note, so editing a disabled legacy condition silently re-enabled it and a
template-valued enabled was dropped before reaching core.
* Surface template rendering errors from condition subscriptions
subscribe_condition events carry template_errors next to a result when a
template inside the condition failed to render. Forward them through the
evaluator's error so the editor flags the condition, without overriding
core's result the way a hard error does.
* Treat a null numeric bound as absent
Lovelace ignores a null bound (== null), but the translator only checked
for undefined and Number(null) turned it into an unintended 0. The editor
conversion now skips null bounds as well.
* Keep the and wrapper for a single-child not and row metadata on fallbacks
Core skips a disabled child, so a bare not over one child evaluates to
true where lovelace's not (¬ of the AND of its children) gives false; the
and wrapper restores that for any arity. The always-false and numeric
value fallbacks also dropped enabled, alias and note, so a template-disabled
leaf would have participated in its parent regardless.
* Treat an empty legacy entity as absent everywhere
checkStateCondition falls back to the host entity with entity || context,
so an existing condition with entity: "" targeted the card entity. The
entity fold, the optimistic seed, the editor conversion and the
translation used nullish checks instead, which kept the empty string and
turned such a condition into an error or an always-false leaf. Use the
same truthy fallback throughout.
* Append non-finite values to the condition signature instead of substituting
A marker string standing in for Infinity could in principle collide with
an equal user string. Keep plain JSON and append the ordered non-finite
values, which is unambiguous.
* Remove a stray brace rendered under every condition editor
A leftover interpolation terminator outside the template expression
rendered a literal "}" below the condition content. Also reject pasting a
condition pinned to an explicit entity into an entity filter, where the
fold-in would keep it and judge every candidate by the copied entity.
* Tidy the optimistic seed's leaf check
Condense its doc comment, take the states map rather than the whole hass
object, and simplify the bound checks.
* Tighten comments on dashboard condition evaluation.
---------
Co-authored-by: Aidan Timson <[email protected]>
* Exit scene live mode when leaving the editor
Leaving the page previously only unsubscribed, so live-activated
device states stayed behind. Restore stored states unless the tab
is hidden for panel suspend.
* Drop extra unit tests for scene live-mode disconnect.
* Restore scene live states only on editor disconnect.
Back and delete already unmount the editor, which now exits live
mode, so a second applyScene was redundant.
* Resubscribe to scene live edits after hidden-tab suspend.
Existing scenes stayed in live mode when the panel was reattached,
but the state_changed subscription was not restored.
* Keep the scene live subscription across hidden-tab suspend.
Unsubscribing and resubscribing on reattach races the closed
websocket; leaving the subscription in place lets the library
restore it after reconnect.
* Tear down scene live mode on hidden-tab route changes.
document.hidden is not enough: saving a new scene remounts a
new editor. Skip teardown only when an ancestor was detached
(panel suspend). Reuse an existing live subscription on
reattach so new scenes do not get duplicate listeners.
Co-authored-by: Petar Petrov <[email protected]>
---------
Co-authored-by: Cursor Agent <[email protected]>
* Serial: send consumers to their own config panel
A port used by Zigbee or Z-Wave led to that integration's entry on the
integration page, which is a device list — not where the radio on that port is
managed. Point those consumers at the integration's own panel instead.
Which panel that is was already decided in `ha-config-entry-row`, from the
panels an integration registers at runtime falling back to
`integrationsWithPanel`. That resolution moves to `getConfigPanelPath` so both
callers stay in step, and so a port picks up any panel the integration page
would already link to.
A consumer that is not running keeps its integration-page link: its panel is
not loaded to receive it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FjmjZsxhdj9ekGmhFHXg3p
* Serial: send the Thread radio to the Thread panel
A Thread radio is held by the Open Thread Border Router app rather than by a
config entry, so it never reached the panel resolution. Map the app to the
integration behind it and look the panel up from there.
The panel is only offered when its integration is loaded, which an app being
started does not imply.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01FjmjZsxhdj9ekGmhFHXg3p
---------
Co-authored-by: Claude Opus 5 <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add Thread to the demo
Load the thread and otbr integrations and mock router discovery, dataset
listing and border router info, so the Thread panel renders in the demo.
The preferred network holds Home Assistant's own border router alongside a
HomePod mini and a Nest Hub; an Amazon Echo forms the one other network.
Widens ThreadRouter's brand union with "amazon", which the backend
already returns. The union is still not exhaustive.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Mock the Thread actions and match the dataset casing
The extended PAN ID was lowercase while the TLV string carries it uppercase.
The dataset dialog looks for one inside the other with a case-sensitive
`includes`, so the preferred network's info showed neither the border
router's URL nor its active dataset.
The panel also offers adding a dataset, choosing the preferred network and
border router, resetting the border router and changing its channel, none of
which were mocked, so each rejected as unimplemented. They now change the
mocked data, and `otbr/set_channel` answers with the delay the panel reads
back.
`ThreadRouter["brand"]` is no longer a union. It only reaches `brandsUrl`,
and the backend resolves it from the border agent's vendor name, so every
new vendor meant another frontend change for no typing benefit.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Push Thread network changes to the discovery stream
The panel groups the routers it draws from `thread/discover_routers`, not
from `otbr/info`, so changing the network only in the info left the border
router drawn under its old network for good. The mock now keeps its
subscribers and announces the router again when it moves.
A reset also has to put the border router on a network of its own instead of
reusing the existing extended PAN ID, which the panel groups by, and it adds
that network rather than taking over the preferred one, so the way back onto
the original network stays available.
`ThreadRouter["brand"]` is null for a vendor the backend does not know, and
the panel now leaves out the icon in that case rather than requesting one
that cannot exist.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Build the Thread datasets from real TLVs
The pasted dataset TLV was malformed: an odd number of hex digits, a network
name field declaring ten bytes for nine, and a network key that ran into the
next type, so a decoder derailed partway through. Datasets are now assembled
from type/length/value triplets, which keeps the declared lengths honest and
each dataset's own extended PAN ID and network name inside its TLV.
Each dataset carries its own, so `thread/get_dataset_tlv` answers for the
dataset it was asked about and rejects an unknown one, and moving the border
router carries the active dataset with it rather than leaving the previous
network's behind, which would have hidden the URL again on the new network.
An imported dataset takes its extended PAN ID from the type 0x02 field
instead of the first sixteen characters, which were the timestamp, and a
reset generates eight hexadecimal bytes rather than a value starting
"RESET".
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Import Thread datasets from the credentials given
The import invented the channel, network name and PAN ID and stored a
rebuilt TLV, so importing real credentials showed a different network than
the one submitted and read back different credentials. It now takes all four
fields from the TLV and keeps that TLV verbatim.
Anything that does not decode is refused rather than falling back to a
generated ID, so the panel shows its error path for malformed credentials.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Treat a Thread import as a revision of the network it names
Datasets sharing an extended PAN ID are revisions of one network, so
importing credentials for a network that is already known updates it rather
than adding a second card for it, and only a newer active timestamp wins.
Re-importing the credentials the demo itself shows used to duplicate the
network.
The active timestamp is now required, as the backend requires it, and a
reset leaves the default router unset, since nominating one is its own
action in the panel.
A router without a brand no longer reserves the avatar column, so its name
is not left indented against an empty graphic.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep the Thread dataset, credentials and channel in step
Deleting a dataset left its credentials behind, so reading them back returned
a network the panel had already dropped. They go with it now.
Changing the channel only moved the border router's reported channel, leaving
the dataset and its active TLV on the old one, so the channel prompt and the
dataset dialog disagreed on the next read. A reset had the mirror problem,
installing a channel 15 network without moving the reported channel. Both
carry the dataset, its TLV and the channel together.
Network names are decoded as UTF-8 rather than one character per byte, so a
non-ASCII name reads correctly and a sequence the backend would refuse is
refused here.
Adds a demo test for the panel: the two networks it starts with, the reset
and rejoin that only show up if the discovery subscription is told, and
credentials being per dataset and going away with one. Checked that it bites
by dropping the announcement, which fails it on the card count after a reset.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Refuse repeated tags and a zero channel in a Thread dataset
A repeated type is malformed and the backend's parser refuses it, rather
than letting the later one win, and a channel field carrying zero is not a
channel it accepts. An absent channel is still fine.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Change the Thread channel without rewriting the credentials
Rebuilding the dataset TLV from its summary replaced the network key, PSKc,
mesh-local prefix, security policy and anything else this mock does not
model with fixture constants, so changing the channel on an imported network
silently swapped its credentials for someone else's. Only the channel
triplet is rewritten now, leaving the rest of the TLV as it was.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Drop the remaining comments from the Thread mock
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add Zigbee to the demo
Load the zha integration and mock its devices, groups, configuration and
network settings, so the Zigbee panel, its options, groups, network info
and network map render in the demo. Neighbor tables give the map a mesh of
routers and end devices.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Give the Zigbee temperature sensor a unique entity ID
sensor.bedroom_temperature already exists in the home demo config, whose
entities are added last and overwrote the display registry entry,
detaching it from its Zigbee device and undercounting the panel's
entities on that demo.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Label the Zigbee options and take the offline plug unavailable
The options page localizes every field through the backend translations and
falls back to the raw identifier, so the pages showed names like
alarm_master_code. Seed a key per schema field, for the global options as
well as the alarm ones.
The panel reports the TV plug as offline, so its entity has to agree.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Mock the Zigbee panel's write actions
Saving either options page, downloading a backup, changing the channel and
adding or removing a group or its members all rejected as unimplemented, so
every one of those visible controls failed. They now change the mocked state
that the panel refetches.
Saving merges into the stored configuration rather than replacing it, the
way the backend does, so saving one section leaves the rest alone.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Seed the ZHA unavailability timeouts and copy the configuration
The options page falls back to two hours when these keys are absent, so the
battery timeout showed 2 hours where ZHA's default is 6. Both are seeded
now.
The configuration response handed out the backing object, and both editors
mutate the fetched data as controls change, so an edit the user never saved
survived leaving the page. It is copied on the way out; only the update
command writes to the stored configuration.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* List the groupable Zigbee devices
The create group page and the add members dialog build their pickers from
this command, so returning nothing left both permanently empty and made the
group commands unreachable from the panel. The lights and plugs are listed
now.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep the Zigbee pairing page, backups and group rows honest
The add device page subscribes to the permit command as soon as it opens, so
with nothing registered it sat on its spinner for the full permit duration
behind a rejected subscription. A subscription that simply stays open is
enough, since nothing pairs in the demo.
A backup shared its network and node objects with the live settings, so
changing the channel afterwards rewrote the stored backup, which is the one
thing a backup must not do. They are copied now.
Group endpoints carried no entities, so both pickers labelled every light
and plug "No entities". Each now carries the entity the fixtures give it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Wire up the ZHA device pages in the demo
The ZHA info card and device actions look up the device's `zigbee`
connection and render nothing without it, so the demo devices only ever
showed the generic device page. Add the connection, and mock the two
commands that become reachable through it: the manage page's cluster
list (with each cluster's attributes and commands, so the attribute and
command tabs are usable), and the reconfiguration subscription, which
now walks the device's clusters and terminates instead of rejecting.
Written attributes are remembered, so reading one back agrees with what
the write button reported.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Carry the registry area on the ZHA device payloads
The group pickers read the area straight off the device payload rather
than resolving the device registry, so the plugs listed no area. The
backend fills it in from the registry; derive it from the registry
fixtures here for the same reason, so the two cannot drift apart.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Make the Zigbee bindings tab work
Both bind buttons ended in their error state: `zha/groups/bind` and
`zha/groups/unbind` were never registered. Registering them is not
enough on its own — group binding lists only a device's client side, so
with every mocked cluster on the server side the table read "No data"
and the button could never be enabled. The remotes now carry the `out`
clusters they would have on real hardware, and `zha/devices/bindable`
returns the routers, so the device half of the tab is no longer hidden.
Also drops the `any` from the configuration update payload, which
already has an exact type.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add Z-Wave to the demo
Load the zwave_js integration and mock its network status, provisioning
entries and node and controller statistics, so the Z-Wave panel, its
options, statistics, network info, provisioned devices and network map
render in the demo. Last working routes give the map a mesh with
repeaters, asleep and dead nodes.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Give the Z-Wave motion sensor a unique entity ID
binary_sensor.kitchen_motion already exists in the home demo config, whose
entities are added last and overwrote the display registry entry,
detaching it from its Z-Wave device and undercounting the panel's entities
on that demo.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add MQTT to the demo
Load the mqtt integration and mock topic subscriptions and device debug
info, so the MQTT panel renders in the demo and its listen card streams
messages on the topic you subscribe to.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Give the MQTT sensor entities unique entity IDs
sensor.kitchen_temperature already exists in the home demo config, whose
entities are added last and overwrote the display registry entry,
detaching it from its MQTT device and undercounting the panel's entities
on that demo.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Key the MQTT debug info by device and send real payloads
The debug info ignored the requested device, so the garage door's MQTT info
showed the fridge sensor's entity. Return the entities of the device that
was asked for, and nothing for one that is not MQTT.
The discovery payload is the config object, not a JSON string:
`mqtt-discovery-payload` dumps it as YAML, so a string rendered as one
quoted scalar. The frontend type said string while the component already
required an object; it now matches what the backend sends.
A subscription takes a topic filter, so echoing it back as the message topic
invented topics no broker could publish. Wildcard levels now resolve to a
concrete matching topic.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Allow an empty MQTT discovery payload
An entity with debug traffic but nothing discovered has no config, and the
backend sends an empty string, so narrowing the field to an object alone was
wrong in the other direction.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
Co-authored-by: Simon Lamon <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add Infrared to the demo
Load the infrared integration and add the emitter and receiver proxy
entities it reads, so the Infrared panel and its device list render in the
demo. The panel is entity driven, so it needs no WebSocket mock.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Add Matter to the demo
Load the matter integration and mock its network topology, so the Matter
panel and its network map render in the demo, with a Thread and Wi-Fi
topology that includes a border router and an offline node.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Key Matter node diagnostics by device
The diagnostics and ping mocks ignored the requested device, so every
Matter device reported node 1 on Thread and available. The device page
gates its actions on that: the offline garden sensor offered actions that
need a live node, and the Wi-Fi plug offered a Thread network link.
Both are now derived from the topology the map already renders, so a
device's node ID, transport, node type and availability match it, and an
unknown device is rejected as the backend would.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Give Matter fixtures unique entity IDs and mock its device actions
lock.front_door and sensor.garden_temperature already exist in the home
demo config, whose entities are added last and overwrote the display
registry entries, detaching them from their Matter devices. The panel
counted 3 entities instead of 5 on that demo. Renamed to a side door lock
and a patio sensor, which no demo config uses.
The device page offers commissioning, fabric and credential actions for an
available node, none of which were mocked, so each failed with
command_not_mocked.
Adds a demo E2E test for the panel and its map. Emptying the topology
subscription fails it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Mock the Matter lock services and fix the commissioning codes
The lock device exposes "Manage lock", whose dialog reads back the response
of matter.get_lock_info and matter.get_lock_users; neither was mocked, so
it always showed its load-failed alert.
setup_pin_code carried the manual pairing code. The three commissioning
codes now agree on the Matter test payload for passcode 20202021.
Pinging an unknown device rejects like the diagnostics command, matching
the backend, which resolves the device before acting either way.
The E2E test now reads the diagnostics of a Thread, a Wi-Fi and an offline
device, so a regression back to one shared response fails it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Return a credential result and gate the Matter Thread action
`setMatterLockCredential` reads `user_index` off the per-entity response, so
the demo's empty object made saving a code throw. Return the indices the
dialog reads back, and drop the mock for a `clear_lock_credential` service
the frontend never calls.
The device action linking to the Thread panel did not check that the
integration is loaded, unlike the same link on the Matter dashboard, so it
could navigate to a panel that is not there.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep the mocked Matter lock users across changes
The manage dialog reloads the list after every add, edit and delete, so the
static response made each change look like it was reverted. Keep the users
per lock entity and mutate them, and hand back a copy so the dialog's
reactive state sees a new list.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Add Bluetooth to the demo
Load the bluetooth integration and mock its scanner, advertisement and
connection subscriptions, so the Bluetooth panel, its adapter info,
monitors and network map render in the demo.
Also adds the scaffolding the other Settings > Connectivity panels build
on: a per-integration fixtures contract, registry builders, and the two
registries that collect components, WS command prefixes and registry data
from each integration. `addEntities` now carries `device_id` and
`platform` into the mocked display entity registry, which the panels use
to count entities per device and per integration.
Mocked subscriptions emit their first message from a timeout. Subscribing
is synchronous in the mock, so an immediate callback lands before the
subscriber is ready: `createCollection` overwrote it with its empty
initial fetch, and pages that ignore messages received before their first
render dropped it.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Serve integration manifests for the demo's config entries
The integration page reads back the manifest it fetches without guarding,
so opening one for a domain the demo had no manifest for threw. Bluetooth
now ships its manifest through the connectivity fixtures, and manifest/get
falls back to a generated manifest rather than answering with undefined,
which also covers the entries that already lacked one.
Adds a demo E2E test that opens the Bluetooth panel and asserts its counts
are non-zero. The connectivity mocks reach the panel through lazy mock
registration and deferred first emissions, neither of which the demo suite
covered; without them the panel renders but stays empty.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Cover the Bluetooth scanner details subscription in the demo E2E test
The dashboard's adapter count comes from the config entries, so the test
passed even with the scanner details subscription broken, while the
adapter page and network map stayed incomplete.
The adapter page renders a settings button only for adapters that are not
remote scanners, which it knows from the scanner type in those details, so
asserting one button across the three adapters covers the subscription
without asserting on copy.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Keep connectivity entities when switching demos
Switching demos replaces the whole state map, and only the energy entities
were added back, so every connectivity state disappeared until a reload.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
* Add Serial to the demo
Load the usb integration and mock the serial port listing, so the Serial
panel renders in the demo with connected, available and disconnected
ports.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01C6PpwxkMaHCWkEFA27YnLE
---------
Co-authored-by: Claude <[email protected]>
* Clarify what uninstalling an app deletes
The uninstall dialog offered a single switch labeled "Also remove app
data". That named the wrong thing twice over: the switch sends
remove_config, which Supervisor applies to the app's configuration
folder, while the app's data folder is deleted on uninstall either way,
independent of the switch.
Relabel the switch after the folder it actually removes, and state in the
dialog that uninstalling permanently deletes the app's stored data, so
the unconditional part of the operation is no longer implied to be
optional.
Rename the local variable and the uninstallHassioAddon parameter to
removeConfig to match the API field they carry.
* Move the app name out of the uninstall dialog title
Dialog guidelines ask to keep user generated content out of titles, as
names can get long enough to be unreadable there. The name now carries
the sentence in the body, which already describes what is deleted.
* Scope the uninstall warning to the app's private data folder
Uninstalling deletes the app's data folder and, when the app maps it,
the public config folder. Files the app wrote through the share, media
or Home Assistant config mappings stay where they are, so promising
that all of its data goes away is wrong for those apps.
Name the folder the deletion is limited to, and say the configuration
folder is only deleted if the app uses one, matching how the Supervisor
API documents the field. Also switch the label to "delete" for
consistency with the sentence above it.
* Name what an app actually keeps in its data folder
Checked the 30 most installed apps for what they write there. It holds
databases (MariaDB, InfluxDB, Grafana, UniFi, Nginx Proxy Manager,
AdGuard), credentials and network identity (Matter fabrics, the Thread
network key, Tailscale node state, Z-Wave JS settings, Mosquitto
accounts) and internal state such as caches and SSH host keys.
"Any other files it created" reaches past all of that. ESPHome keeps
device configuration in the Home Assistant config folder, Node-RED sets
its user directory to /config, WireGuard writes its keys to /ssl, and
those survive an uninstall.
* Fix punctuation in uninstall dialog text
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
---------
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Add select-slider card feature for select/input_select tile control
* Modified select-slider card feature to become just a style of the preexistent select-options
* Modified name icons to segmented and removed suggestion
* Fixed issues with punctuation and spaces
* Renamed segmented to buttons
* Wait for connectivity settings readiness
* Wait for serial settings readiness
* Wait for storage settings readiness
* Simplify delayed serial scenario
* Test delayed connectivity readiness
* Fit the picker popover to its list and align it with the select menu
* Use styleMap for the popover custom properties
* Fix combo box list not rendering for direct embedders
* Rename AI tasks settings to AI and add Model Context Protocol card
The AI settings page now shows a second card for the Model Context
Protocol server. When no mcp_server config entry exists, a centered
button starts the config flow. When enabled, the card shows the server
URL, an alphabetical list of registered LLM APIs with copy buttons for
their URLs, a configure button that opens the options flow, and a
disable action that deletes the config entry.
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Mock MCP server config entry and LLM APIs in demo
Lets the demo show the enabled state of the Model Context Protocol
card on the AI settings page.
Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Rename MCP server URL row and mock MCP in the demo
The row heading did not say what makes this endpoint different from the
per API endpoints below it. Call it "Your MCP URL" and show the config
entry title, so it is clear which APIs it serves.
Mock an MCP server config entry and the registered LLM APIs in the demo.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Simplify MCP card URLs to labeled copy buttons
Build the URLs from the origin Home Assistant is browsed on, so the
copied URL is reachable for whoever copies it.
Drop the URL text from the card and label the copy action, which was an
unlabeled icon that did not say it copies a URL. Name the API section
Individual MCP APIs so it is clear those are MCP endpoints too.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Move MCP disable action into a header overflow menu
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Align MCP card header actions and drop the disable icon
Put the help and overflow buttons in the header row so they line up with
each other and the title, instead of positioning them absolutely.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Rename MCP card row to Your MCP API
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Drop trailing word from MCP card description
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Put configure before copy and drop Calendar from the demo
Ordering the configure button first lines the copy buttons up across all
rows. Calendar is not an LLM API integrations register, so mock only
Assist and Music Assistant.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Always show the MCP configure button
The options flow is now in core, so the entry always supports options.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Tell users where to add the MCP URL
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Link the MCP docs from the card description
Point at the integration documentation for agent-specific instructions,
and call them AI agents.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Drop the MCP card help icon
The description links the documentation, so the header icon is a second
link to the same page.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Show MCP URLs in outlined rows
Give each URL its own outlined row with the URL under the name, and copy
with an icon button, so the URLs are visible before copying.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Use hassUrl for MCP URLs and show load errors
Build the URLs with the standard helper instead of the window location.
Catch a failing load and show an alert. Without it a failed load left
the card with neither the URLs nor the enable button.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
* Show a spinner while the MCP card loads
Without it the card showed only its description until the config entries
arrived.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01RQUrBPMXN35UQ19ZYPRpEZ
---------
Co-authored-by: Claude <[email protected]>
* Remove web components polyfills and ES5 custom elements adapter
All supported browsers (legacy floor: Chrome 59, Safari/iOS 12, Firefox 94)
have native shadow DOM and custom elements, so the webcomponents bundle,
ShadyCSS branch, and lit polyfill-support are unreachable. The legacy build
now emits ES2017 classes, so custom-elements-es5-adapter and the
window.loadES5Adapter hook (no known third-party consumers) are removed.
Also stops shipping the never-loaded dialog-polyfill css and drops both
now-unused dependencies.
* Remove keyed-es5 Terser workaround
The custom keyed directive existed because Terser with ecma: 5 miscompiled
the destructured update() parameters (#28732). The legacy build now minifies
with ecma: 2017, so the stock lit-html keyed directive works in both builds.
* Remove old-browser JS shims and stale ES5 build references
Drops the IE-only navigator.msMaxTouchPoints check, replaces the
toggleAttribute helper with the native method (polyfilled automatically for
Chrome < 69 in the legacy build), and removes babel excludes for the
uninstalled proxy-polyfill and unfetch packages. Updates comments that
still described the legacy build as ES5.
* Remove vendor prefixes for no-longer-supported browsers
Deletes -ms- prefixes (IE/EdgeHTML only) and -webkit-/-moz- prefixed
declarations that every supported browser understands unprefixed, or that
Lightning CSS re-adds automatically from the unprefixed property in
production builds. Blocks that only had prefixed user-select now use the
standard property (previously Firefox got no user-select there at all).
Converts the four -webkit-linear-gradient() declarations - the sole
gradient syntax on those sliders - to standard linear-gradient().
* Remove dead html_url custom panel support
html_url pointed to an HTML Import, a Polymer-era feature removed from
Chrome in 2019 and never shipped elsewhere. The loader has not handled the
html type for years (it fell through to a rejection), and core's
panel_custom integration no longer accepts the option, so the branch was
unreachable. Also drops the ha-panel-${name} legacy tag naming that was
keyed on html_url.
* Repair list-plugins-and-polyfills script for Babel 8
The audit script died at startup since the Babel 8 update: preset-env no
longer exposes lib/debug.js (logPlugin is now inlined locally, built on the
public getInclusionReasons helper) and babel-plugin-polyfill-corejs3 v1 no
longer ships lib/shipped-proposals.js (list inlined).
Instead of invoking the preset with a hand-mocked plugin API - the part
that kept drifting - the plugin listing now runs a real transform of an
empty file with preset-env in debug mode, declaring the same caller
capabilities as babel-loader. The polyfill listing now passes the
configured core-js version to core-js-compat, mirroring the provider's own
filtering so the report cannot list modules the installed core-js lacks.
Output is byte-identical to the direct-invocation approach. Also documents
the script in the build-scripts README.
* Correct macOS floor for Safari 26 in companion app UA regex
Safari 26 ships for macOS 14 Sonoma and 15 Sequoia, not only macOS 26, so
the SAFARI_TO_MACOS entry breaking the minimum-supported-macOS pattern
would have sent updated macOS 14/15 companion apps to the legacy build
once the modern floor reaches Safari 26.
* Remove orphaned values left behind by vendor prefix removal
* Require macOS 14.6 for Safari 26 in companion app UA regex
Replace the wizard's own registration and sign-in screens with the cloud
panel's, extracted into cloud-register-card, so signing up mid-wizard gets
email-confirmation auto login instead of the legacy register plus login
polling. The wizard's copy of the login error ladder goes with it.
Drive the registration state from cloud/status rather than from the cloud
events, which carry nothing it does not already hold. The waiting view now
clears when the pending registration does, so a restart no longer leaves it
spinning on a registration the backend has forgotten.
Depends on the auto-login state fix in home-assistant/core#180423: without
it a retry loop that gives up before the register view stores its controller
leaves cloud/status reporting a pending registration with no reason, which
this no longer reads from the event instead.
* Group the connectivity panels behind a Connectivity settings page
Move the connectivity panels (Matter, Zigbee, Z-Wave, KNX, MQTT, Thread,
Bluetooth, serial, infrared, radio frequency, Insteon, and tags) behind a new
Connectivity page. It sits with Voice assistants in the second settings group,
where those panels used to be listed individually.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01Y7Yy1BcbkMjCZEtkTEqyKm
* Update config panel e2e coverage for the Connectivity page
The connectivity panels moved off the settings root, so split the link smoke
cases and assert them on the Connectivity page instead. Add a route smoke case
for /config/connectivity.
Co-Authored-By: Claude Opus 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01Y7Yy1BcbkMjCZEtkTEqyKm
---------
Co-authored-by: Claude <[email protected]>
`ha-camera-stream` declares a `load` event and both of its video players fire it through `fireEvent(this, "load")`, but the MJPEG and snapshot fallback `<img>` never does. Every camera without a working stream therefore goes through the one path that stays silent.
`hui-image` listens for that event to record `_lastImageHeight`, and that value is what drops the `.ratio` class so the container can size to its content. Without it the container keeps the 16:9 padding box it falls back to before anything is measured, while the picture is drawn at its own ratio and overflows. Measured against a 600x410 camera in a 600px wide card: the container stays 337.5px tall while the picture is drawn 410px tall, and it is still so ten seconds after the card is built.
Firing the event from the fallback image settles it: the container becomes 410px, matching the picture, and the 16:9 fallback goes back to being a placeholder for the first frame instead of a permanent state. Cameras that do have a stream are unaffected — they already fire `load` from their player.
* Add timer data helpers for formatter-based display and duration serialization
Refactor computeDisplayTimer to take formatEntityState instead of hass so
context-migrated components can reuse it, add finishes_at to TimerEntity,
and add durationDataToTimerString for serializing duration input values.
* Modernize timer more-info dialog with state header and live countdown
Use ha-more-info-state-header with a ticking remaining-time display and
the standard more-info control layout. Adds timer to
DOMAINS_WITH_NEW_MORE_INFO, replacing the legacy state-card row.
* Add timer-actions and timer-presets card features
timer-actions shows start/pause/cancel/finish buttons (finish opt-in)
with state-aware disabling; start becomes restart while active.
timer-presets shows configurable one-tap durations that call timer.start.
Timer tile suggestions default to the timer-actions feature.
* Pulse timer red when it finishes
When a timer runs out or timer.finish is called, the tile icon and the
more-info countdown pulse red twice. Cancelling does not pulse: the new
timerJustFinished helper distinguishes the transitions via the
last_transition attribute. Honors prefers-reduced-motion.
* Apply suggestions from code review
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Send normalized duration for timer presets
Presets are parsed leniently by createDurationData like other duration
inputs, but the raw config value was sent to timer.start. A malformed
preset such as 1:nope:00 rendered as 1 h yet was rejected by core. Store
the normalized seconds so the label and the service call always match.
* Store timer presets on the entity instead of the card config
Review feedback: presets belong to the timer entity, like cover favorite
positions and light favorite colors. They now live in the entity registry
options and are edited in the more info dialog through the shared favorites
UI, which also brings reset and copy to other timers. The tile feature keeps
its style option but reads the presets from the registry.
* Reject a zero duration when adding or editing a timer preset
A zero duration passed the form's required check but was then treated as a
cancel, so saving the prefilled 0:00:00 silently did nothing. Form dialogs
now take an optional submit-time validate hook that blocks the submit and
shows a field error, and the preset dialog uses it to require at least one
second.
---------
Co-authored-by: Copilot Autofix powered by AI <[email protected]>
* Support multiple for media selector
* Update src/components/ha-selector/ha-selector-media.ts
Co-authored-by: Petar Petrov <[email protected]>
* add to UI
* merge fix
* add helper
* Modify media item picker for clearable and image upload
---------
Co-authored-by: Petar Petrov <[email protected]>
The secondary line of system log entries was made inline-block in #53575
to apply text direction. An inline-block is an atomic inline, so the
parent's text-overflow: ellipsis hid the whole line whenever it did not
fit, leaving only "…" on mobile. Use unicode-bidi: isolate instead so
direction still applies while the text truncates normally.
Fixes#53879
* refactor(profile): split general section into subpages with dashboard navigation
Replace the two-tab profile layout (general/security) with a dashboard-first
navigation structure. The general section is split into three focused subpages
(preferences, localization, browser), each wrapping the original row components
unchanged. A new dashboard page provides a user card (name, owner badge, logout)
and a navigation list linking to all four subpages.
Add ha-navigation-list as a reusable component for page navigation lists.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor: remove orphan translation keys after general section split
Drop tabs.general and current_user, which were only used in the now-deleted
ha-profile-section-general.ts and the old profileSections tab config.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): align dashboard layout with Bluetooth panel pattern
Replace .content with .container (padding on wrapper, max-width/margin on
cards) to match the layout convention used in protocol panels like Bluetooth.
Always show the chevron in ha-navigation-list, removing the narrow condition.
Drop the unused narrow prop from ha-navigation-list.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor: move ha-config-navigation-list to src/components
Makes the component available outside the config panel. Update import
paths in the two existing config consumers and use it in the profile
dashboard. Remove the now-unused ha-navigation-list (which relied on
ha-md-list/ha-md-list-item).
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* fix: security subpage width and card spacing
Use .container > * instead of ha-card to apply max-width and margin,
since the security cards are custom elements whose inner ha-card lives
in their own shadow DOM and is unreachable from the parent stylesheet.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): move language setting to localization subpage
Language is a regional/cultural preference, consistent with the other
settings in localization (time zone, date and time formats).
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): rename profile subpage labels
- "User preferences" -> "Appearance"
- "Browser settings" -> "This browser"
- "Mobile app settings" -> "This mobile app"
* feat(profile): surface theme picker on dashboard page
Move ha-pick-theme-row from the Appearance subpage to the profile
dashboard, between the user card and the navigation list. Theme and
dark mode are now accessible directly without navigating to a subpage.
Update Appearance description to reflect its new scope.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): restructure theme row layout and labels
Split dark mode and color pickers into separate ha-settings-row elements
with headings. Move reset button outside the color-row to keep the label
vertically stable when the button appears. Add padding to reset row.
Rename "Reset" to "Reset colors" for clarity.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* add title to theme settings card
* refactor(profile): apply theme row layout improvements to ha-theme-settings
Migrate the dark mode and color pickers layout from ha-pick-theme-row into
ha-theme-settings: use separate ha-settings-row elements with headings for
dark mode and custom colors, move reset button outside as a sibling element.
Add colors label to ThemeSettingsLabels interface.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): migrate theme settings to ha-list-item-base layout
Replace ha-settings-row elements in ha-theme-settings with ha-list-item-base
inside ha-list-base for a more compact appearance. Split color pickers into
individual rows (one per color). Reduce vertical padding to --ha-space-2.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* refactor(profile): combine color pickers into single row with label
Merge primary and accent color pickers into one ha-list-item-base row
under a "Custom colors" headline. Restore labels on individual pickers
and increase their min-width to 150px.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* fix(profile): stack theme settings rows vertically in narrow mode
Reflect narrow prop on host and use ::part(base/end) to stack all
ha-list-item-base rows in column layout on narrow screens, preventing
headline truncation from oversized end slot controls.
Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
* Fix test:e2e:demo
* fix broken links
* Add my links
* Use s instead of smal for ha-button size
Co-authored-by: Maarten Lakerveld <[email protected]>
---------
Co-authored-by: Claude Sonnet 4.6 <[email protected]>
Co-authored-by: Maarten Lakerveld <[email protected]>
The iOS workaround from #52925 only applied when the selector had an
explicit negative min. The numeric threshold selector used by the power
triggers passes a number selector without a min, so the digit-only keypad
(without a minus key) was still shown. Treat a missing min as allowing
negatives too.
Also leave inputmode unset instead of forcing "text": on a number input
iOS then shows the Numbers and Punctuation keyboard, which has a minus
key and fits numeric entry better than the full QWERTY keyboard.
Fixes#53747
* Load MapLibre RTL text plugin so Hebrew and Arabic map labels read correctly
The vector base map never registered MapLibre's RTL text plugin, so
right-to-left scripts were shaped left to right and every label came out
reversed. Ship @mapbox/mapbox-gl-rtl-text from /static/map/ alongside the
glyphs (no third-party host) and register it once, lazily, before the
first vector layer is created.
Fixes#53851
* Add the English name to map labels in non-Latin scripts
Before the move to vector tiles the CARTO basemap showed English names
everywhere. The OSM style shows local names, which most users cannot
read in Cyrillic, Arabic, Hebrew or CJK regions. Keep the local name and
add the English one from the tiles under it - in parentheses for street
labels, which cannot break lines. Latin-script names are left alone, so
Köln stays Köln.
* Respect gas display precision in gas graph
The gas graph previously formatted the total gas consumption using the generic number formatter, which could ignore the display precision configured on the selected gas entity.
Use the configured gas entity's display precision for both the visible total chip and the total consumption tooltip.
This keeps the gas graph consistent with the entity configuration. For example, a gas entity configured with 3 decimal places will display 0.009 m³ instead of being rounded to 0.01 m³.
* Use configured gas entity display precision in energy distribution card
The energy distribution card previously formatted gas usage with the generic
energy formatter, which could ignore the display precision configured on the
selected gas entity.
Use the display precision of the configured gas consumption entity when
formatting the gas value in the energy distribution card.
This keeps the displayed value consistent with the entity configuration. For
example, a gas entity configured with 3 decimal places will now display
0.009 m³ instead of being rounded to 0.01 m³.
* Respect gas display precision in energy sources table
The energy sources table previously formatted gas values using the generic number formatter, which could ignore the display precision configured on the selected gas entity.
Use the configured gas entity's display precision for both individual gas source rows and the gas total row.
This keeps gas values consistent with the entity configuration. For example, a gas entity configured with 3 decimal places will display 0.009 m³ instead of being rounded to 0.01 m³.
* Use highest gas display precision for energy distribution
The energy distribution card previously used the display precision of the first configured gas source for the aggregated gas value.
Use the highest configured display precision across all gas sources instead of special-casing the first source.
This avoids making source 0 authoritative for an aggregate value and preserves the greatest configured precision when multiple gas sources are present.
* Use highest gas display precision for energy distribution
The energy distribution card previously used the display precision of the first configured gas source for the aggregated gas value.
Use the highest configured display precision across all gas sources instead of special-casing the first source.
This avoids making source 0 authoritative for an aggregate value and preserves the greatest configured precision when multiple gas sources are present.
* Use highest gas display precision in gas graph
The gas graph previously used the display precision of the first configured gas source for its aggregated total.
Use the highest configured display precision across all gas sources instead of special-casing source 0.
This keeps the graph total and tooltip consistent with the greatest configured precision when multiple gas sources are present.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Use display_precision for entity precision
Replace dp with display_precision when reading precision from hass.entities, matching the EntityRegistryDisplayEntry type used by the frontend.
* Restore consumption formatting in energy distribution card
* Respect gas display precision in energy sources table
* Respect gas display precision in energy graph
* Respect display precision in formatConsumptionShort
* Pass gas display precision to consumption formatter
Add logic to calculate gas display precision based on available sources.
* Use gas display precision for graph y-axis
* Update distribution card on gas precision changes
* Update sources table on gas precision changes
* Update gas graph when display precision changes
Refresh the gas graph when a gas entity's display precision or unit of measurement changes, ensuring the configured precision is applied without waiting for new energy data.
* Update distribution card on gas unit changes
Refactor gas source display precision check and update unit of measurement comparison.
* Derive gas display precision with a getter
* Fix hass change handling in energy sources table
* Fix hass change handling in gas graph
* Fix hass change handling in energy distribution
* Fix gas unit narrowing in energy distribution
* Fix formatting in gas graph card
Refactor gasDisplayPrecisions assignment for better readability.
* Fix formatting in energy sources table
* Fix Prettier formatting in energy gas cards
---------
Co-authored-by: Petar Petrov <[email protected]>
wa-after-hide from a tooltip inside the drawer (e.g. the exact time
tooltip on a persistent notification) bubbled to the wa-drawer listener
and was treated as the drawer closing. Only handle the drawer's own event.
Fixes#52901
Core now proxies both vector and raster tiles, which is what lets them be
requested with an application User-Agent and without a referrer - a
browser can set neither. So the frontend stops talking to OpenStreetMap
and CARTO directly and goes through /api/map_tiles.
The proxy is token gated, so `ha-map` fetches one over the WebSocket
before setting up, following the brands token pattern: cached at module
level, refreshed well inside its lifetime so a dashboard left open for
days keeps working. The blocking wait is kept to about a second - a
backend without the proxy must not hold the map hostage - and the
remaining retries run in the background to ride through the window after
a restart where the WebSocket is up but the handler is not registered
yet.
Two things that are not obvious and cost a measurement each:
MapLibre's `transformRequest` has to return absolute URLs. Tiles are
fetched from a worker, which has no document to resolve a relative URL
against, and the TileJSON that core serves has relative `tiles`. Measured
with a relative TileJSON on one origin: with absolute URLs the style
loads and 8,456 features render; without them exactly one request is made
- the style - and nothing else loads, with no error reported anywhere.
Leaflet bakes its URL template at layer creation and throws while
building a tile URL if a template variable is undefined. So the raster
layer takes the token as an option Leaflet substitutes per request, which
also means a refreshed token is picked up without recreating the layer,
and an absent one is empty rather than missing: the tiles 403 and the
markers still draw.
The asset pipeline shrinks to generating two styles. Glyphs and sprites
come from the proxy, so the 48 MB build-time download, the digest
verification, the glyph range filtering and the bold-range guard are all
gone, along with 5.4 MB from the wheel. `localIdeographFontFamily` goes
too: the complete glyph set is reachable now, so CJK renders in Noto
rather than a device font.
Co-Authored-By: Claude Opus 5 <[email protected]>
* Remove no-op User-Agent header from Nominatim requests
`User-Agent` is a forbidden header name for browser fetch/XHR, so the
option passed to both Nominatim calls was silently dropped and the
browser's own User-Agent went out instead. Verified in Chromium against
a local echo server: the header never took effect, the outgoing UA was
the browser default, and zero preflights were received — had it been
applied, the non-safelisted header would have forced an OPTIONS
preflight that Nominatim does not answer.
Identification towards Nominatim is done by the `email` query parameter
both calls already send, which is what the Nominatim usage policy
accepts, so nothing changes on the wire. Drop the misleading dead code
and note in a comment why the header cannot be set from a browser.
`hass` is still needed by both functions for `hass.locale.language`, so
the signatures are unchanged.
Co-Authored-By: Claude Opus 5 <[email protected]>
* Apply suggestions from code review
Co-authored-by: Maarten Lakerveld <[email protected]>
---------
Co-authored-by: Claude Opus 5 <[email protected]>
Co-authored-by: Maarten Lakerveld <[email protected]>
Co-authored-by: Maarten Lakerveld <[email protected]>
* Fix editor state leaking when switching automations, scripts, or scenes
Recreate the page when the item id in the route changes so YAML mode,
undo history, and scene live state cannot carry over to the next item.
* Declare itemId on editor routes instead of guessing from the path
Path-shape matching remounted nested config routers, Lovelace, and
developer tools. Recreate only where edit/show routes opt in.
* Do not restore scene live states when the editor is torn down
A remounted editor starts in review; disconnect only needs to drop
the subscription, not undo device changes from live mode.
* Use the normal create path when remounting itemId pages
Clear the current page instead of calling _createPanel directly so
load and the loading screen still run. Drop the redundant cache delete.
* Fix dialog-form tests leaking focus restore after jsdom teardown
Nested submit/cancel schedules a nextRender callback that ran after
Vitest tore down HTMLElement, failing CI with an unhandled rejection.
A percentage has a real ceiling the way zero is a real floor, but the gap that
keeps series off the plot edges did not know that: a battery reading 20-100%
rounded out to an axis labelled up to 120%, and one sitting flat at 100% — a
device left on the charger — reached 160%.
Recognise the unit at the two line-chart call sites and hold the axis at 100,
mirroring the existing zero clamp. Only while the data stays under it, since
power factor is also reported in % and is signed.
* Show Z-Wave neighbor connections in the network visualization
* Fetch neighbors in bulk and hide them behind a toggle
* Warn that loading neighbor data turns off the adapter
* Guarantee a Y-axis gap on history and statistics charts
ECharts floors the axis minimum and ceils the maximum to a tick multiple,
so quantized states that land exactly on a tick leave the data flush against
the plot edge — which collapses the climate and humidifier action bands to
zero height, since an area fill is based at the axis minimum.
Widen the auto-scaled extent by a fraction of the span so that same rounding
always has something to round away, keeping the axis anchored at zero for
single-signed data and preserving the window ECharts gives a constant series.
* Widen the Y-axis gap threshold to 2% of the span
A nudge only large enough to break an exact tie left the gap unchanged
whenever the data sat just above a tick rather than on it, so a chart could
still render with a hairline of headroom.
Because ECharts rounds out to a whole tick afterwards, this constant does not
set the size of the resulting gap — it decides which axes count as too flat to
leave alone. At 2% the worst case over the tick cycle goes from nothing to
around 3% of the plot, while every measured chart keeps the framing it had.
* Deduplicate in-flight requests
* Use callWS for in-flight request ownership
* Avoid mutating shared request results
* Deep-freeze shared in-flight request results
Centralize immutability in shareInFlightRequest using deep-freeze so
callers do not need readonly types on shared API shapes.
---------
Co-authored-by: Petar Petrov <[email protected]>
The digital clock hardcoded 1.5rem/3rem/4rem, which ignored
--ha-font-size-scale and could not be overridden from a theme. Use
--ha-clock-card-digital-font-size-{small,medium,large} with defaults based
on the font size scale (24px/48px/64px).
Fixes#51602
The CSV header used the timestamps of every statistic in the energy
collection. Since the power sources chart (#27501) that includes the
power statistics, which are fetched at a finer period than the energy
statistics (hourly for a month view, where energy is daily), adding
empty hourly columns between the daily energy values. Collect the rows
first and build the time columns only from the data that is actually
exported.
Fixes#52381
* Prompt for unsaved changes on programmatic navigation
* Tidy unsaved-changes guard tests
* Drop navigations superseded while an unsaved-changes prompt is open
The assistants URL filter can skip creating a row for entities that are
only exposed via Cloud (Alexa/Google) manual config, but the sortable key
was still assigned to the missing row, throwing a TypeError and leaving
the page on an infinite spinner. Only assign the key when the row exists.
Fixes#53689
The chart data modules push a null y value to break the line where an
entity was unavailable. downSampleLineData read it with Number(), and
Number(null) is 0, which is not NaN, so the isNaN guard did not fire.
The marker then competed as a real value of 0 and won its frame's
minimum slot whenever the readings were positive, discarding the
frame's actual minimum and widening the rendered gap.
Keep markers out of the min/max comparisons entirely and hold at most
one per frame in its own slot. It is emitted, after the frame's values,
only when no kept value follows it: a marker followed by a value in its
own frame is a gap that closed within one frame, which is about one
device pixel wide and too narrow to show. That check runs per frame at
emit time, so the per-point path stays as it was. Keeping every marker
instead would blow up the output on series that are mostly null, such
as the climate heating dataset, which went from 823 to 14525 points
before this was bounded.
Skipping markers before the numeric work also makes gapped series
faster: 16% on a series with a few gaps, 27% on one that is mostly
gaps. Both now have benchmark coverage, which the gap path lacked.
Mean mode no longer averages markers in as zero.
* Add sesnor device class selector
* Remove unused constants
* Remove stale test case and import
* Fix CI tests
* Add domain and make selector generic
* Follow up comments
* Use import
* Add security dashboard alerts
* Fix security alerts editor and card defaults
* Use consistent security editor icons
* Fix security dashboard alert state
* Fix lint
* Render security alerts as individual cards
* Use context data in security alerts editor
* Add pulse for alerts
Co-authored-by: Paul Bottein <[email protected]>
* Use explicit cover and lock alert states
* Fix test
* Remove security panel rendering tests
* Use complete states
Co-authored-by: Paul Bottein <[email protected]>
* Remove camera
Co-authored-by: Paul Bottein <[email protected]>
* open state only
Co-authored-by: Paul Bottein <[email protected]>
* Remove security entity filter cache
* Format
* Update security alert visibility tests
---------
Co-authored-by: Paul Bottein <[email protected]>
The option handlers read the value from ev.target, which is whichever
element the click landed on. Real pointer events are retargeted to the
outer role="radio" element by .option .content { pointer-events: none },
but a screen reader's synthetic activation click is not hit-tested, so it
lands on the inner content element, which has no value.
ev.currentTarget is always the element the listener is bound to, so it
always carries the value.
* Focus nested form-dialog fields after a nested level is pushed.
Host focus on ha-form does not pierce selector shadows, so keyboard
focus stayed on the opener instead of the nested controls.
* Simplify nested form-dialog focus and fix CI types.
Wait one render and focus the first real control instead of waiting for
custom-element upgrades or walking a general focusable tree.
* Wait for lazy selector upgrades before nested form focus.
nextRender is not enough on a cold selector chunk; wait for undefined
custom elements so nested focus does not depend on network timing.
The security strategy step rendered ha-form with a data object built
from a @state field that was never assigned, so the form was always
bound to an undefined strategy. ha-form's own optimistic data update
made the radio look selected after a click, but the next re-render of
the step (triggered by any hass update) re-committed the undefined
value and cleared the selection.
Make the step a controlled component: the dialog already owns
_inclusionStrategy and resets it when navigating back, so pass it down
as a property instead of keeping a second, unused copy in the child.
Claude-Session: https://claude.ai/code/session_011MGgDoAUxzf9KBexn6uiPc
Co-authored-by: Claude Opus 5 (1M context) <[email protected]>
_getDetailData caches on stateObj alone while formatting via this.hass,
so results go stale when translation-based format functions reload (e.g.
backend translations finishing after open, or a language switch) until
the entity next changes state. Pass hass.formatEntityAttributeName as an
extra memoize argument so the cache invalidates when format functions
are recreated.
* Fix energy dashboard staying on yesterday after midnight.
* Catch up the energy live day before subscribe fetches.
* Keep energy day math in the server timezone so DST cannot skip a live day.
Browser-local addDays can jump a calendar day on a 23-hour DST fallback.
Assert against tz-internal endOfDay/addDays under Europe/Berlin so UTC CI
catches a regression, and prove the 01:00 timer and catch-up refresh fetch
the live day rather than only updating collection.start.
* Don't follow the live day from midnightRollover alone.
A stored non-today preset would otherwise be discarded on the first subscribe. Drop tautological UTC DST tests that cannot fail.
* Fix editing nested multiple object selectors
* Fix nested form dialog close and focus restoration
Cancel all pending nested form levels when the physical dialog closes, and restore focus to the opener after nested submit or cancel.
Add regression coverage for both behaviors.
* Fix selector object test types
* WIP: Initial attempt
* Arrange into sections
* Clean up UI
* Add descriptions
* Align with Core changes
* Address review feedback and add an info dialog
* Combine websocket commands into one
* Drop unnecessary info dialog rows
* Add a status bar
* Expose `resolved_device` as well
* Show result of action on button feature
* Add button entity
* Add to button entities
* add to tile icon
* Delay for spinner
* Remove result indicator from tile icon
* Add keyboard and screen reader navigation to charts via Chart2Music
* Mark charts as busy while the sonification chunk loads
* Skip chart sonification when there is nothing to navigate
* Drop the direct chart2music dependency
* Exclude legend-hidden series when judging chart sonifiability
* Push statistics chart times as numbers so sonification reads real dates
The greeting sits in the assistant's chat bubble, so render it in the pipeline's language instead of the interface language. Falls back to the interface language when no translation is available (issue #53703).
* fix flowType in config subentry flows/next_flow
* suggested changes
* Repair Flow Next Flow Repair
* Add repair flow support a new repair flow as the next_flow
* Add support for the next_flow being another repair flow
* fix lint errors
* Reviewer response and fix linter error
* Revert to generics to avoid multiple casts
* Move show-dialog-repair-flows to dialogs
* Update src/dialogs/config-flow/dialog-data-entry-flow.ts
* Apply suggestions from code review
* Apply suggestion from @MindFreeze
* format
---------
Co-authored-by: Petar Petrov <[email protected]>
Co-authored-by: Paulus Schoutsen <[email protected]>
* Show per-mount storage usage on the storage page
Each active mount now fetches its own usage from the Supervisor and renders
a bar with a used-of-total line under its address. The requests are not
awaited, so a slow or unreachable server cannot hold up the rows, and a
failed one simply leaves that row without usage.
fetchHostDisksUsage now takes a disk and an optional max_depth. The data
disk callers keep their depth of 3; mounts send no depth at all, since
walking one costs a round trip per directory and the Supervisor already
has a sensible per-target default.
* Vertically align ha-bar
---------
Co-authored-by: Simon Lamon <[email protected]>
The brand outline pointed at a token that doesn't exist, so CSS fell back
to the neutral border and brand outlined buttons drew a gray ring.
Disabled outlined buttons only reset their background and label, keeping
the variant-coloured border.
Also add outlined to the gallery: it has always been a valid appearance
but was missing from the documented list the gallery grid is built from,
which is why both bugs went unnoticed.
Co-authored-by: Claude Opus 5 (1M context) <[email protected]>
* Fix filter pane expand animation and use a bottom sheet in narrow mode
* Update src/common/controllers/filter-panel-controller.ts
Co-authored-by: Petar Petrov <[email protected]>
* Apply suggestion from @MindFreeze
---------
Co-authored-by: Petar Petrov <[email protected]>
* Ignore stale History statistics fetches after the selection changes.
A slower fetchStatistics call could finish after a newer sources or date-range request and replace the charts the user is looking at.
* Update src/panels/history/ha-panel-history.ts
---------
Co-authored-by: Paul Bottein <[email protected]>
* Redesign History and Activity filtering into a sources pane
Give both panels the toolbar + left pane layout of the data tables, and
merge target picking and filtering into one "Sources" surface, per the UX
discussion.
- ha-filter-pane: pane on wide screens, bottom sheet on narrow ones,
mirroring the filter pane of hass-tabs-subpage-data-table
- ha-sources-picker: target picker plus domain, device class and
integration filters, shared by both panels
- ha-filter-device-classes: new filter panel, labelled with the backend
device class names
- ha-filter-pane-chip and ha-empty-state: the chip with a filter count
badge and the centered placeholder both panels need
- ha-date-range-nav: the date range picker as one pill with previous and
next steppers, for the toolbar
- History draws timeline names above their bar (inside-labels), which
gives long names the full width
- Activity keeps a floating date header while scrolling, and reset moved
into the overflow menu next to refresh and download
- Drop the now unused compact and add-on-top modes of ha-target-picker,
including ha-target-picker-value-chip
- Right-align the clear button in the domains filter header, matching the
other filter panels
Co-Authored-By: Claude Opus 5 <[email protected]>
* Remove dead code breaking the type check
* Fix stuck loading state when the selection resolves to no entity
* Keep the target picker out of the redesign
* Remember the source filters across visits
* Truncate timeline names that do not fit the plot
* Show device class icons in the sources filter
* Open the sources pane by default on wide screens
* Show the entity count in the sources chip
* Count the same entities in the sources chip and the target picker
* Fix sources resolution, fetching and filter badge in History and Activity
* Update src/panels/history/ha-panel-history.ts
Co-authored-by: Petar Petrov <[email protected]>
---------
Co-authored-by: Claude Opus 5 <[email protected]>
Co-authored-by: Paul Bottein <[email protected]>
Co-authored-by: Petar Petrov <[email protected]>
* Add Matter network topology visualization page
* Add central Home Assistant node to Matter network graph
* Refine Matter graph HA node: theme-aware memoization, drop dead branch
* Render Matter 'unknown' link strength as a present, neutral link
* Pass devices to getDeviceArea so bridged nodes inherit their area
* Distinguish position-unknown Matter graph anchors and name border routers
Anchors to a hubless component only mean the node is reachable, so draw
them dotted instead of reusing the solid line that marks a real path
through a border router or access point.
Also prefer the border router mDNS host name over vendor and model,
which several vendors report identically on every unit.
* Name Matter access points by SSID and show their radio address
Access points were labelled with a BSSID, and the same address was
repeated as the node context. Prefer the SSID for the label and keep the
radio address as context only when it differs, so the radios of one mesh
stay distinguishable without repeating the label.
Falls back to today's BSSID against servers that do not send an SSID.
* Colour Matter graph links by transport and float unknown neighbours
Unknown Thread neighbours are not commissioned on our fabric, so Home
Assistant has no operational path to them. A group made only of unknown
devices no longer draws an anchor to the Home Assistant node and floats
instead of claiming a connection that does not exist.
Link colour now encodes the transport, purple for Thread and pink for
Wi-Fi, with the signal level left on the line width and spelled out in
the tooltip. A dead link keeps the disabled colour, which is the only
thing separating it from a healthy weak one now that both are width 1.
The edge tooltip names the network, since the graph legend describes
nodes only and cannot carry a link entry.
* Float Matter nodes whose route to Home Assistant is unknown
Only border routers and access points get an edge to the Home Assistant
node, because only those are a path we can actually see. A node in a
group with neither was previously anchored to Home Assistant with a
dotted line, which reads as a physical connection that was never
observed.
Removing the anchor makes the component walk that picked a
representative dead code, so it goes too.
* Match the dashboard's line semantics for Matter graph links
Dash an edge whose endpoint is inferred rather than commissioned, or is
offline, which is the same pair of conditions the matter.js dashboard
dashes on. Stop drawing a link whose every direction is dead: the
summary strength is the strongest direction, so none means a stale
neighbour entry, and the dashboard never draws one either.
That also retires the grey dead-link colour, since such links no longer
reach the graph.
* Carry the transport colour onto the Home Assistant edges
An edge from Home Assistant to a hub now takes the colour of the network
behind it, so one transport reads as one colour the whole way back
instead of changing hue at the hub.
Wi-Fi moves from pink to orange: pink sat close to the error red used
for offline nodes, while orange is far from both that red and the Thread
purple.
* Give the Wi-Fi access point node its transport colour
The access point wore indigo while everything else on its network was
orange. Deriving the category colour from the same helper the links use
keeps a hub and its links one hue, and gives the Wi-Fi colour a legend
entry it did not have before.
* Give the border router node its transport colour too
Both hub categories now derive their colour from the same helper as the
links, so a network is one hue from Home Assistant through its hub to
the devices, and each legend swatch keys the links of that transport.
The border router also gains contrast: the shared Thread purple is
legible on both the light and the dark card background, where the deep
purple it replaces was not.
* Add mqtt config dashboard shortcut
* Improve MQTT subscribe panel formatting and add copy button
* Standardize MQTT options flow button to a nav card style
* Refactor MQTT dashboard
* Improve style
* Add link to broker connection settings
* Test the effect of not exporting the SVG data
* Move MQTT svg icon to separate file
* Fix backup progress segment order to match backend execution order
The segmented backup progress bar showed Apps, Media, Home Assistant,
but Supervisor backs up Home Assistant first since 2024 (supervisor#5203),
then add-ons, then folders. The Home Assistant segment activated first
while Apps/Media incorrectly showed as completed, and the bar regressed
once the apps stage started.
Also move the await-addon-restart stages into the last creation group,
since Supervisor awaits add-on restarts after the backup file is
finished, keeping the progress bar monotonic.
Co-Authored-By: Claude Fable 5 <[email protected]>
* Keep await restart stages out of MEDIA_STAGES
Give them their own constant and merge them into the last creation
group only where the ordering constraint applies.
Co-Authored-By: Claude Fable 5 <[email protected]>
* Group backup setup stages with the first progress segment
addon_repositories/app_repositories (and docker_config on older
Supervisors) are emitted while the backup is initialized, before the
home_assistant stage, so grouping them with the apps segment made the
bar regress at the start of a backup.
Co-Authored-By: Claude Fable 5 <[email protected]>
---------
Co-authored-by: Claude Fable 5 <[email protected]>
* Group sankey devices by count as well as by value
The 0.1%-of-total value threshold only catches devices that are a
negligible share of the home total. A breaker panel with dozens of
similar-sized circuit clamps sits entirely above it, so nothing groups and
every node collapses toward the 1px minimum with labels shrinking to match.
Cap the number of named children per parent too, defaulting to 20 (the
readable capacity of the default 400px card). Over-cap devices are grouped
into the same per-parent "Other" node the value threshold already produces,
so the flow arithmetic is unchanged, and the cap is configurable per card
via max_devices.
* Clarify the max_devices label is per upstream device, not per floor or area
* Make max_devices budget named devices, not nodes
A cap of 20 rendered 19 devices because a slot was reserved for the Other
node, and a cap of 1 rendered none. The option now means what its name and
its namesake on the devices graph mean: at most this many devices are shown
by name, with Other as overhead on top.
* Extract the sankey device cap into findDevicesOverCap
Moves the count-cap logic out of buildSankeyDeviceNodes into an exported
function that takes the graph it needs, so it can be unit tested directly,
and trims the commentary to the two non-obvious rules (why the whole subtree
is grouped, and why at least two devices are).
Repair fix flows build their menu from the suggestions the backend
reports. A suggestion this frontend has no translation for yet (e.g. a
newer Supervisor offering a new repair suggestion) rendered as an
empty, unlabeled menu entry. Show the raw option key instead, matching
the fallback the form step header already uses.
Co-authored-by: Claude Fable 5 <[email protected]>
* Support linking a person to an existing user
* Don't delete preexisting user on cancel submit
* retry ci
* Don't show choice when no floating users exist
* collapse to single dialog
---------
Co-authored-by: Simon Lamon <[email protected]>
* Add activity detail dialog
* Apply review fixes to activity detail dialog
* Resolve previous state from history in logbook detail dialog
* Disable text selection on clickable logbook rows
* Document run discovery as a fallback for missing context ids
* Scope the detail chain to the subject's cause path
* Replace clickable logbook rows with explicit detail affordances
* Show the cause badge on automation and integration rows
* Map scheduled and Home Assistant causes to their trigger icons
* Replace logbook detail affordances with full-width rows
* Frame the activity detail dialog with ha-grouped-list
* Add missing import
* Fix milliseconds placement in activity chain times
* Lay out inline features in two columns
* Detect stacked features from the slot
* Rename tile container layout classes
* Rename the below features in the layout helper
* Keep area card features below at full width
* Stretch the area card features below the inline one
* Keep area card features compact outside the compact type
* Revert "Keep area card features compact outside the compact type"
This reverts commit 48268cb4e2.
* Reapply "Keep area card features compact outside the compact type"
This reverts commit e10bb27e2f.
* Keep the inline feature compact on condensed cards
* Add child devices UI
Surface child devices throughout the config UI so they read as first-class
devices nested under their parent:
- Device page: a "Sub-devices" card lists a device's children, and a child's
page shows a "Part of <parent>" link (hardware/model/config-entry are
already inherited from the parent by the registry resolver).
- Integration page: children are nested and indented under their parent
device in the config-entry and subentry device lists.
- Device picker: children are ordered and indented under their parent with a
tree connector, mirroring the area/floor picker.
- Naming: the device picker's secondary label and search now include the
parent device name for children, so they stay identifiable in flat views.
Follow-up to the child devices data layer (#53617).
Co-Authored-By: Claude Opus 4.8 <[email protected]>
* Add child devices to demo and gallery mock data
A power strip parent with two outlet children in the demo device stubs and
the ha-selector gallery demo, so child device rendering (nesting, tree
indentation, parent-context naming, inherited area) can be exercised without
a running backend.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
* Refine child devices UI and target resolution
Follow-up polish so child devices behave like a normal device everywhere:
- Targets & filters: a parent device qualifies for and resolves to its
children's entities (getDevices filter, deviceMeetsFilter,
deviceMeetsTargetSelector, resolveEntityIDs and the target-chip "split into
entities" expand), matching core's server-side target resolution. Selecting a
parent excludes its children from the picker.
- Pickers: keep a parent visible when a child matches the search (device and
target pickers), fix the target picker's nested order (unsorted search +
recomputed last-child flag), and render the sub-device tree in the target
picker's device group.
- Area/naming: the area field shows only the (inherited) area again; the parent
name remains a search term.
- Integration page: correct the tree end connector and align it in narrow mode.
- A parent-disabled child can no longer be enabled from the settings dialog.
- Devices dashboard: show "Part of <parent>" under a sub-device's name and add a
hidden-by-default, searchable and groupable Parent device column that groups a
parent together with its children.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
* Address review on child devices UI
Align target resolution with core and fix picker/table details:
- Only a directly targeted device expands to its child devices. Labels are
never inherited by children (core's target helper is explicit about this) and
areas resolve by effective-area membership, so deviceMeetsFilter and
deviceMeetsTargetSelector evaluate a device's own entities again.
- Add devicesInEffectiveArea, mirroring core's dr.async_entries_for_area: an
area contains its devices plus children that inherit the area, but not a child
with a different explicit area. Used for area expansion, area matching and the
area chip's split action.
- Device picker: add a searchFn that restores the nested parent/child order
after the fuzzy search and recomputes the last visible child, so a child can
no longer be ranked above its parent and connectors stay correct.
- Devices dashboard: derive the family group name from the family's parent
device with the same fallback for parents and children, so an unnamed parent
cannot end up in a different group than its children.
- Child devices card: pass the device registry so a child shows its inherited
area.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
* Fix formatting in device picker row renderer
Reindent the row renderer template after it gained a block body, so Prettier
is satisfied.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
* Make the child device tree connector span the full row
The connector was a fixed 48px box centred in a taller two-line row, and its
SVG was letterboxed by the default preserveAspectRatio, so the dashed line
stopped short of the row edges and consecutive children never visually
connected.
Let the indicator stretch (preserveAspectRatio="none") and give it the full row
height, so the line runs edge to edge with the elbow at the vertical centre.
non-scaling-stroke keeps the line width and dash pattern identical however far
it is stretched; existing 48x48 usages render unchanged.
Co-Authored-By: Claude Opus 4.8 <[email protected]>
---------
Co-authored-by: Claude Opus 4.8 <[email protected]>
* Show remaining tile card features below when position is inline
Previously the tile card silently dropped every feature after the first
one when "features_position" was set to "inline". The first feature is
now rendered next to the name as before, and any additional features are
stacked underneath, the same way they are in "bottom" position.
The tile container gained a "features-bottom" slot for this, and the
card size and grid options now account for the extra rows.
* Add tests for hui-tile-card size and grid options calculations
* Enhance hui-tile-card to support inline feature layout and improve feature counting logic
* reduce vertical gap for inline features
---------
Co-authored-by: Claude Opus 5 <[email protected]>
* Drop websocket collections created by an iframe panel on teardown
`getCollection()` caches collections (entity registry, label registry, ...) on
the `Connection` object. A custom panel embedded in an iframe shares the
connection with the main window, so a collection the panel is the first to
request is created inside the iframe's realm.
When the user leaves the panel, `ha-panel-custom` removes the iframe and that
realm is destroyed - but the collection stays cached on the connection. Its
store, and the `setTimeout` that hands the cached state to new subscribers, are
gone, so every later subscriber waits forever for a callback that can never
fire. Since the registries moved to `LazyContextProvider` they are subscribed
on demand, which makes an iframe panel the first requester far more often: the
KNX panel's entity table renders empty after leaving and re-entering it, and
the main frontend can inherit the same dead cache.
Clean up from the iframe's own `pagehide` handler, while the realm is still
alive - `disconnectedCallback` on `ha-panel-custom` runs after the browsing
context is discarded, when `contentWindow` is already `null`.
Co-Authored-By: Claude Opus 5 <[email protected]>
* Use unknown instead of any in the CachedCollection shape
The helper never inspects the subscribed state, so the parameter type only
needs to be a placeholder.
Co-Authored-By: Claude Opus 5 <[email protected]>
---------
Co-authored-by: Claude Opus 5 <[email protected]>
* use defaults provided by core for ports in strings, add url to new location after restart
* Apply suggestions from code review
Co-authored-by: Simon Lamon <[email protected]>
---------
Co-authored-by: Simon Lamon <[email protected]>
* Use backend slugify for the entity ID format preview
* Update src/panels/config/core/ha-config-entity-id-format.ts
Co-authored-by: Petar Petrov <[email protected]>
* Use top-level slugify websocket command
* Fix formatting
---------
Co-authored-by: Petar Petrov <[email protected]>
* Search a media player's own library from the media browser
The media browser search only ever asked the media sources, so searching
inside a media player's own library (Music Assistant, Sonos, Squeezebox,
Jellyfin) failed. Ask the entity instead when the current item belongs to
it, the same way browsing already does.
* Update src/data/media_source.ts
* Remove blank line in media_source.ts
Removed unnecessary blank line at the top of media_source.ts
---------
Co-authored-by: Bram Kragten <[email protected]>
Co-authored-by: Simon Lamon <[email protected]>
* Polish the launch screen
- Use the system sans-serif font so the launch screen no longer blocks on
loading Roboto
- Reduce the gap between the logo and the loading text
- Dim the loading text to 66% opacity
- Match the OHF logo variant to the applied theme instead of the system
color scheme
- Make the launch screen text and buttons unselectable
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
* Drop unit test
* Use classMap
* Drop view transition for css transition to avoid flash
---------
Co-authored-by: marcinbauer85 <[email protected]>
Co-authored-by: Claude Opus 5 (1M context) <[email protected]>
description:Home Assistant frontend PR and review guidance. Use when reviewing frontend changes, preparing a PR, checking recurring review issues, or applying the PR template.
description:Home Assistant frontend PR and review guidance, including implementation quality, existing review feedback, established design patterns, and the authority of UI/UX evidence. Use when reviewing frontend changes, preparing a PR, checking recurring review issues, or applying the PR template.
---
# HA Frontend Review
Use this skill when reviewing Home Assistant frontend changes or preparing a pull request.
## Review Preparation
1. Establish the review scope from the diff and the behavior changed by it.
2. Before identifying findings, use the [project skill catalogue](../../../AGENTS.md#project-skills) to load every matching companion's `SKILL.md`. Follow applicable companion references in those skills as well; this review skill supplies the workflow, and companions supply the domain-specific criteria.
3. Apply all loaded guidance within that scope. Report only problems introduced or worsened by the changeset, anchored to changed lines. Read surrounding code to understand affected behavior, not to generate unrelated cleanup or migration requests.
## Pull Request Body
When creating a pull request, use `.github/PULL_REQUEST_TEMPLATE.md` as the body.
@@ -22,7 +28,7 @@ When creating a pull request, use `.github/PULL_REQUEST_TEMPLATE.md` as the body
-`yarn lint` passes when practical for the scope.
-`yarn test` or focused relevant tests are green when practical for the scope.
-Tests are added or updated for new data processing and utilities where applicable.
-Each test added by the change protects real logic, not the look of a component.
- User-facing text is localized and follows `ha-frontend-user-facing-text` guidance.
- Components handle loading, error, unavailable, and missing-entity states.
- Entity existence is checked before property access.
@@ -92,9 +98,27 @@ Configuration and props:
- Keep APIs extensible without adding speculative abstractions.
- Validate configuration before applying changes.
## UI/UX Evidence
For user-facing changes, establish the existing design context as part of frontend review:
- Treat an applicable gallery specification as the authoritative repository source for the documented component or interaction. Inspect its written guidance and demos when present.
- Note relevant production designs, approved designs, and shared `ha-*` components so the implementation can also be compared with established frontend behaviour.
- Require the applicable gallery documentation or demo to change when the implementation intentionally changes behaviour documented there.
- Prefer an established input layout or appropriate shared component over a raw input or ad hoc control.
- Treat explicit UI/UX approval, current repository guidance, and direct human or workflow instructions as stronger evidence than inference.
- Treat linked tasks as evidence of the problem and stated requirements, not automatic UI/UX approval.
- When the direction remains uncertain, search frontend pull requests that carry or previously carried **Needs UX**. Inspect their label history, comments, and reviews rather than relying on the current label alone.
- Judge relevant feedback by its content and surrounding discussion. When GitHub provides `author_association`, prioritise feedback marked `MEMBER`.
- Prefer recent feedback about similar interactions or components. Label removal can show that a workflow gate moved on, but does not prove approval without the surrounding discussion.
- Summarise the applicable guidance instead of maintaining a reviewer list or treating one historical decision as a permanent rule.
## Review Flow
- Before reviewing a pull request, read its existing comments, reviews, and threads, including their status, resolver, and Copilot resolution reason when available.
- Prioritise substantive human feedback, especially from authors marked `MEMBER`, and validate agent-generated feedback against the code and repository guidance.
- Do not duplicate unresolved findings as new inline comments; reference any that still need action in the review summary. Treat resolved feedback as closed only when the resolution reason or surrounding discussion supports that outcome; otherwise validate it against the current code before suppressing it. Respect **Won't fix** and **Incorrect** reasons.
@@ -49,12 +49,13 @@ Do not pass `--help`, `--background`, or `--modern` to `script/build_frontend`;
Managed app, demo, gallery, and E2E app workflows share one lifetime lock, so only one build or development server can run at a time.
## Unit And Utility Tests
## When To Add Tests
-Add or update Vitest tests for data processing, utility code, and behavior that can be tested without a browser.
-Mock WebSocket connections and API calls at boundaries.
-Cover loading, error, unavailable, and missing-entity states where relevant.
-Test accessibility-sensitive behavior when it can be asserted without brittle DOM internals.
-Write tests for code that computes something: data processing, utility functions, config validation, and what happens when the user interacts with a component.
-Do not write tests that check what a component looks like: its text, CSS classes, styles, or slots. Do not write tests that check the default value of an option.
-A component that only takes data from contexts and helpers and puts it in a template does not need a test.
-If you are not sure a test is useful, describe the test and what it would catch, and let the user decide.
- Tests never talk to a real Home Assistant. Replace `callWS`, `callApi`, and the connection with fakes.
description:Home Assistant frontend UI/UX readiness routing for new cards, badges, screens, and interaction patterns. Use it with `ha-frontend-review` to decide whether frontend review is enough or further UI/UX input is needed. Skip it for PRs already carrying Needs UX and for fixes or technical details that preserve an established experience.
---
# HA Frontend UI/UX Readiness
Use the design context and evidence gathered through `ha-frontend-review` to decide whether a change's UI/UX direction is settled and whether further UI/UX review should be recommended or is explicitly required. This skill routes the result; it does not repeat or replace implementation, accessibility, or frontend quality review.
## Early Exit
If the PR already carries the **Needs UX** label, reviewers have already decided to ask designers or UX reviewers for guidance. Do not use this skill to make that recommendation again.
## Triage
1. Apply `ha-frontend-review` and record its UI/UX evidence.
- Include the applicable gallery specification, production or approved designs, shared `ha-*` components, task requirements, historical feedback, and direct instructions.
- Keep the implementation review separate from this readiness decision.
2. Classify the user-facing decision.
- Existing-feature fixes and technical details normally remain with frontend review when they preserve the established experience.
- New cards, badges, screens, and interaction patterns normally need this readiness check.
3. Decide whether the evidence gathered during frontend review settles the direction.
- Frontend maintainer guidance, explicit UI/UX approval, current guidance, an established pattern, or a task that settles the material visual and interaction choices can settle the direction.
- Insufficient or conflicting evidence leaves the direction unresolved.
4. Select exactly one route below and state which frontend-review evidence supports it.
## Routes
- **Frontend review sufficient:** The change preserves the established experience, or the evidence settles the material visual and interaction choices. Cite that evidence and continue with `ha-frontend-review`.
- **UI/UX review recommended:** A new user experience has material design choices that remain unresolved. Identify the decision that needs input and recommend UI/UX review.
- **UI/UX review required:** A workflow rule or direct human instruction explicitly requires UI/UX review. Cite that evidence. A new design, task link, or lack of evidence is not enough on its own.
Frontend maintainer guidance, explicit UI/UX approval, established patterns, and task requirements inform the route. A task can settle requirements without settling the UI/UX. Recommend UI/UX review when important design choices remain unresolved.
@@ -33,10 +33,11 @@ Never run `tsc` or `yarn lint:types` with file arguments. When `tsc` receives fi
- Do not query or manipulate DOM manually when Lit decorators, component refs, or render state are appropriate.
- Scope styles to components, use theme custom properties, and keep layouts mobile-first and RTL-safe.
- All user-facing text must be localized through the translation system.
- Do not write tests just because you changed some code. Write a test when there is real logic that could break without anyone noticing, and explain what the test protects.
## Project Skills
Detailed guidance lives in project skills under `.agents/skills/`. Load the matching skill before detailed implementation or review:
Detailed guidance lives in `.agents/skills/<name>/SKILL.md`. Load every matching skill before detailed implementation or review. For reviews, load `ha-frontend-review` alongside all companions that apply to the changed code or behavior:
-`ha-frontend-contexts`: Lit contexts, `hass` migration, and rerender-sensitive state access.
-`ha-frontend-components`: dialogs, forms, alerts, shortcuts, tooltips, panels, and Lovelace cards.
@@ -46,7 +47,8 @@ Detailed guidance lives in project skills under `.agents/skills/`. Load the matc
"_comment":"Initial JS budget (raw/uncompressed bytes) for the cold-load critical entrypoints. Enforced by build-scripts/check-bundle-size.cjs in CI. Re-seed after an intentional change with `--update --headroom=<percent>`.",
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Maarten Lakerveld