16 KiB
Styling hooks — the selectors card-mod may rely on
Status: implemented (v1.59.0-beta.3). Code: src/houseplan-card.ts
(_renderDevice, _renderRoomLabel, _renderOpenings, _renderDecorLayer,
the room shapes in render(), the space tabs in the header),
src/space-render.ts (the static houseplan-space-card).
Smoke: demo/smoke_styling_hooks.mjs.
This document is a contract, not a feature. It names the handful of attributes and classes that the card promises not to rename, so that a user who has already installed card-mod can write a rule against the plan and keep it working across our releases.
§1 The philosophy — we do not sell CSS
The card has no "custom CSS" field, no theme editor, no style templates, and this document does not introduce any. We are not building a styling layer, and we do not support user CSS.
What we are doing is narrower and cheaper: we stop getting in the way of a
user who already runs card-mod. Their CSS pierces our shadow root on its own;
all they were missing was something stable to aim at. Until now the plan had
meaningful class names but almost no identifiers — five data- attributes in
the whole source, every one of them an internal implementation detail
(data-corner on a drag handle, data-mid on a vacuum trail). A rule like
"make the icon of this sensor bigger" had to be written against
:nth-child(7) and broke the next time a device was added.
The distinction from ha-floorplan matters and is deliberate. There, CSS is a required part of the product: you configure the floorplan by writing stylesheets and templates, so every user is a stylesheet author and every release is a chance to break one. That is where their 138-post "my styles no longer apply" thread comes from. Here, the plan is configured by clicking, the card ships finished, and CSS is something a power user may add on their own responsibility — the way a person may add a browser userstyle to any web page.
Consequences, stated plainly:
- We promise the names in §3 are stable. If we ever have to change one, it is a breaking change and it goes in the changelog.
- We promise nothing about anything not in §3 — internal classes, DOM
nesting, element order, the presence of a wrapper
<div>, the internals of Home Assistant's own components. - We do not review, debug or support user CSS, and "card-mod rule stopped working" is not a bug report we can act on unless it names a §3 selector.
- We will not add hooks on request one by one. §3 covers the objects the plan is — devices, rooms, openings, decor, labels, floors. That is the whole model; there is nothing else on the plan to aim at.
§2 The model — one attribute answers "what is this"
Every object the plan draws now carries the same identity:
| Attribute | Meaning |
|---|---|
data-hp |
what kind of object this is — the one attribute to branch on |
data-id |
the object's own id in the stored config, stable across restarts |
data-kind |
the sub-type, where a kind has one (door/window/gate, line/rect/…) |
data-entity |
the entity the object presents, where it presents one |
data-area |
the Home Assistant area id, where the object belongs to one |
data-hp is the anchor: [data-hp="device"] selects every device marker on
the plan and nothing else, regardless of which layer, mode or element type it
happens to be rendered as today. It is deliberately one attribute rather than
a class, so it can never collide with a state class (.on, .sel, .ghost)
and so querySelectorAll('[data-hp]') enumerates the whole contract.
Absent is absent. An attribute is not rendered at all when the object
has no such value: a virtual marker with no entity has no data-entity, a
room outside any HA area has no data-area, and a room drawn before ids
existed has no data-id. There is never a data-area="undefined" and never
an empty string — [data-area] means "this one has an area".
Ids are the config's, not the DOM's. data-id is the id stored in
houseplan.config (marker id, room id, opening id, decor shape id). It
survives reloads, re-mounts and re-orderings; it changes only when the object
is deleted and drawn again. A device that is auto-discovered and has no
marker of its own uses the same synthetic id the card uses for it internally.
§3 The table — element → attributes → class
Everything in this table is public API.
| Object | Element | data-hp |
Other attributes | Public class |
|---|---|---|---|---|
| Device marker | div (HTML, marker layer) |
device |
data-id = marker/device id, data-entity = primary entity id, data-area = area id, data-binding-status = active | ha-disabled | orphaned | unverified, data-disabled-reason = device | entity | all-entities when applicable |
.dev |
| Room shape | polygon / rect / path (SVG) |
room |
data-id = room id, data-area = area id |
.room |
| Room name (no metrics) | text (SVG) |
room-label |
data-id = room id, data-area |
.rlabel |
| Room card (name + metrics) | div (HTML, marker layer) |
room-label |
data-id = room id, data-area |
.roomlabel |
| Door / window / gate | g (SVG) |
opening |
data-id = opening id, data-kind = door | window | gate |
.opening |
| Coloured tunnel inside a thick opening | path / g (SVG) |
opening-tunnel |
data-id = opening id, data-kind = door | window | gate |
.opening-tunnel |
| Wall body (thickness) | path (SVG) |
wall |
data-id = segment key, data-kind = shared | outer |
.wallbody |
| Independent partition | path (SVG editor hit target) |
partition |
data-id = partition id, data-kind = partition |
.physical-hit |
| Wall column | path / circle (SVG editor hit target) |
wall-column |
data-id = column id, data-kind = square | circle |
.physical-hit |
| Unfinished room contour | line (one editor hit target per segment) |
room-draft |
data-id = draft id, data-kind = segment, data-segment = zero-based segment index |
.physical-hit |
| Decor shape | line / rect / ellipse / text (SVG) |
decor |
data-id = shape id, data-kind = line | rect | ellipse | text |
.dshape (.dtext on text); persisted colour/alpha are inline SVG attributes and therefore win over weak CSS selectors |
| Furniture | path (SVG) |
decor |
data-id = shape id, data-kind = furniture, data-symbol = the symbol id (sofa, toilet, …) |
.dshape .dfurn |
| Floor / space tab | button (HTML, header) |
space-tab |
data-id = space id |
.tab |
data-symbol is the furniture library's own vocabulary (docs/FURNITURE.md
§3): the ids in that table are public and stable, and new ones are only ever
added. It is what lets one rule colour a whole category:
ha-card [data-kind="furniture"][data-symbol="bathtub"],
ha-card [data-kind="furniture"][data-symbol="toilet"] { stroke: #4fc3f7; }
3.1 Public classes inside a device marker
The marker's own children are part of the contract too — they are what a rule usually wants:
| Class | What it is |
|---|---|
.dev |
the marker box itself |
.dev ha-icon |
the icon (a Home Assistant element — see §5) |
.valtext |
the value badge of a "value instead of icon" marker |
.tval |
the small temperature plate next to the icon |
.hval |
the small humidity plate |
.lqi |
the Zigbee signal badge |
.ripple |
the presence-ripple rings |
.newdot |
the "new device" dot |
Inside a room card: .rlname (the name) and .rlmetrics / .rlm (the
metric row and one metric in it).
3.2 Public state classes
These are set on .dev and describe live state, so a rule may key off
them:
.on, .off, .unavail, .alarm, .ghost (hidden device shown in the
device editor), .virtual (a marker with no entity), .valonly (value
instead of icon), .static-icon (the always-static display mode), .noicon,
.sel (selected in an editor).
Unified device activity is exposed by one descendant .device-pulse with
exactly one kind class: .alarm, .short or .continuous. Semantic reason is
available as .reason-alarm, .reason-event, .reason-presence,
.reason-transition or .reason-running. Reduced-motion ordinary activity is
the compact .activity-dot; no static ring is emitted. The legacy
.activity-ring descendant class remains a beta compatibility alias only and
must not be used by new integrations.
The full plan and houseplan-space-card emit the same binding-status data
attributes. A forced-hidden HA-disabled marker is absent from ordinary View;
the attributes are visible on its service ghost in the Device editor.
On a room shape: .filled (a fill mode is painting it), .styled,
.overlay / .yard (drawn over a picture / on bare canvas).
3.3 Explicitly NOT the contract
These exist in the DOM and will change without notice. Do not build on them:
- Boot and transition classes —
.hpboot,.hpsettle,.mode-transition,.editorchrome.transitioning, and the four-phase.daycycle/.phase-*environment classes. Their whole job is to be replaced. - Editor chrome and previews —
.dtframe,.dthandle,.dtknob,.dtbox,.dtstem,.bdhandle,.rszhandle,.rszicon,.rszframe,.ddraft,.vacfithandle, the align guides, the markup layer (apart from the physical-object identity attributes explicitly listed in §3),data-corner,data-mid. These are the drawing tools' own furniture; they are redesigned whenever an editor is. - Layout wrappers —
.stage,.zoomwrap,.devlayer,.decorlayer,.measurelayerand their nesting. The layers are how we composite; they are not where the objects live. - Dialogs —
.dialog,.menuwrap,.entrow,.inforowand friends. Dialog markup follows the dialog's design, and the design changes. - Everything generated — CSS custom properties starting with
--hp-,--room-*,--dev-*,--ripple-*,--rl-*are set inline by the renderer. Reading them is fine; overriding them may fight the renderer, which writes them again on the nexthass.
§4 Examples
Real card-mod configuration, on the card that carries the plan. (card-mod is a separate community integration — install it first; we do not ship it.)
Hide every Zigbee signal badge.
type: custom:houseplan-card
card_mod:
style: |
ha-card [data-hp="device"] .lqi { display: none; }
Make one specific marker twice the size — the front door lock, named by its entity, so it keeps working after the device is renamed or moved:
card_mod:
style: |
ha-card [data-hp="device"][data-entity="lock.front_door"] {
--dev-scale: 2;
z-index: 5;
}
Recolour the room captions of one floor, and make the kitchen's stand out — by area, which is the name Home Assistant knows it by:
card_mod:
style: |
ha-card [data-hp="room-label"] { font-weight: 700; letter-spacing: .02em; }
ha-card [data-hp="room-label"][data-area="kitchen"] { color: #c62828; }
Dim every window, keep the doors bright — openings by kind:
card_mod:
style: |
ha-card [data-hp="opening"][data-kind="window"] { opacity: .45; }
The supported Solid / Dashed choice for a decor line lives in its properties dialog. For a custom dash pattern beyond those two product styles, card-mod can still target one line by its shape id from the config:
card_mod:
style: |
ha-card [data-hp="decor"][data-kind="line"][data-id="dc_17"] {
stroke-dasharray: 12 8;
}
Disclaimer. card-mod is not ours: we do not ship it, do not support it, and do not answer for what your CSS does to the card. The examples above are illustrations of the selectors, not supported configuration. If a rule misbehaves, remove it first and then decide whether there is a card bug underneath.
§5 Limits — where a selector cannot reach
The card's own shadow root is fine. Everything in §3 is rendered by the
card into its single Lit shadow root, including all dialogs and menus — we
never portal a dialog to document.body. card-mod's shadow piercing reaches
all of it.
Home Assistant's own elements are not. ha-icon, ha-switch,
ha-slider, ha-card are separate custom elements with shadow roots of
their own, and a descendant selector stops at their boundary. You can style
the host — .dev ha-icon { color: … ; transform: … } works, because the
icon inherits color and the host takes the transform — but you cannot reach
the <svg> inside ha-icon. That is a browser rule, not our choice, and it
is why we put the hooks on our wrappers rather than expecting a rule to
target the icon element itself.
houseplan-space-card is a different card. The read-only space card is
its own custom element with its own shadow root, so it needs its own card-mod
block. It carries the same data-hp attributes for the objects it draws
(rooms, room labels, device markers); it draws no openings and no decor
layer, so those simply are not there.
The kiosk header does not exist. In kiosk mode the whole header is not
rendered, so [data-hp="space-tab"] matches nothing — that is not a
regression, there are no tabs to style.
One rule, both modes. Editors add classes to the same elements rather than
re-rendering different ones, so a rule written against the View mode also
applies while the user is editing. If that is not wanted, scope it: the stage
carries mode-view / mode-plan / mode-devices / mode-decor, and those
four names are part of the contract for exactly this purpose.
§6 Values are formatted by Home Assistant
A styling contract is about how a value looks; this section is about what
the value is, because the two questions arrive together — "why does my sensor
show 22.4 here and 22,4 °C everywhere else in HA".
Wherever the card prints one entity's state or attribute, it now asks
Home Assistant to format it: hass.formatEntityState(stateObj) for a state,
hass.formatEntityAttributeValue(stateObj, attr) for an attribute. That is
the same call HA's own more-info and entities card make, so the number obeys
the entity's display_precision, the user's decimal separator, and the state
translations (on → Включено). One wrapper owns it — hassValue() in
src/logic.ts — and every printing site goes through it:
| Where | What it prints |
|---|---|
Value badge (display: value) |
the acting entity's state |
| Decor live text (docs/LIVE-TEXT.md) | the linked entity's state or attribute |
| Device info card | the primary state and every listed entity |
Fallbacks are silent. An older Home Assistant without
formatEntityState gets exactly today's behaviour — the raw state, with the
entity's unit_of_measurement appended by us. The same is true of
formatEntityAttributeValue, which is newer still. Nothing throws, nothing
is blank; the wrapper reports which path it took so the caller knows whether
a unit is already in the string.
The unit appears exactly once. HA's formatter normally appends the
entity's own unit itself — so blindly adding ours would double it, and blindly
trusting it would drop the unit wherever it does not. One rule survives both:
strip the entity's own unit_of_measurement if it is already the tail (exact
trailing match, nothing else in the text is touched), then append the unit
that is actually wanted — the user's explicit one on a decor label, the
entity's own otherwise. A value with no unit at all — a translated state
(«Включено») — never grows a suffix. valueWithUnit() in src/logic.ts.
The °/% plates keep their own form. The small temperature and humidity
plates next to an icon (.tval, .hval), the same numbers in a room card
(.rlm) and in the tooltip are not an entity state readout: they are a
derived reading — an average over every sensor of the area, or a climate
device's current_temperature attribute — rendered as a fixed compact glyph
(21.5°, 48%) so a plan full of them reads as one instrument panel. There
is no single entity whose display_precision applies to an average, and a
formatter would put °C into a badge the size of a fingernail. They stay
ours, deliberately (owner's call, 2026-08-05).