Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 79 600 files from 1 763 authors, of which 61 947 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
A component that renders nothing when its feature is unavailable cannot remove the container someone else wrapped it in — gate the slot on the same availability predicate, publish that predicate beside the component, and branch the slot's clickable and non-clickable forms properly. Use when an empty cell, gap or stray divider appears where an optional control should be, when a group's rounded end caps land on the wrong item, or when a wrapper swallows the taps meant for the control inside it.
Build a radar/fingerprint chart whose axes are normalised 0..1 from your own data — no external corpus, one guarded denominator per axis, and a second polygon (the previous period) because a lone shape on self-normalised axes says nothing. Use when a "usage personality" or "year in review" chart needs a reference it cannot get, when a thin period leaves one axis at a guarded zero beside four real readings, or when one axis goes NaN and the whole polygon disappears.
Hold the colours that have no Material role — a liked-state pink, an active-line highlight, shimmer tones, overlays that sit on artwork — in an @Immutable token class provided through staticCompositionLocalOf, with the bytecode-level reason static is the right choice for theme values and a rule for what belongs in the token class versus in the colour scheme. Use when hex literals are spreading through composables, when a colour has to differ between light and dark but is not a scheme role, or when you are choosing between staticCompositionLocalOf and compositionLocalOf.
A relay that owns the shared state stamps its own default onto fields your command did not set — most painfully forcing "not running" onto every item change — so a follower that obeys the message verbatim stops the thing it just loaded. Carry the previous intent across the change, and publish the missing field as a second command. Use when followers in a synchronised room go silent on every next/previous/end-of-item while the source plays on.
A self-measuring shimmer modifier plus skeleton composables that stand in for a list while it loads — how the modifier learns its own size, why the base colour under the sweep is load-bearing, why the modifier order between clip and background changes what gets rounded, and why the skeleton's lazy lists must have scrolling switched off. Use when building loading placeholders, or when a shimmer renders as a flat block, has square corners under a rounded design, or pauses visibly between passes.
A selection dialog that maps the chosen localized label back to a stored value breaks the day two labels translate identically — the write is skipped or lands on the wrong option, with no error; covers carrying the id instead of the text, making the miss loud, and the sibling hazard of a default declared both in the store and as the collector's initial value. Use when a settings choice does not stick in one language only, when a picker writes a neighbouring option, or when a screen renders the wrong variant for a moment on entry.
The moment an app shell paints its content panels in anything other than colorScheme.background, every gradient, scrim and fade whose tail converges on colorScheme.background ends on a hard seam — the decoration is still correct, its destination colour is just no longer on screen. Covers resolving one page-background value and handing it to every decoration, the two valid answers (aim at the shell colour, or paint your own ground), and sizing a decoration with matchParentSize over the content it decorates rather than a constant measured on one form factor. Use when a gradient stops mid-screen with a visible edge, when a fade looks right on a phone and wrong in a desktop window, or when adding a window chrome breaks screens that were never touched.
Build a scrim that melts artwork into the page background without a visible seam — smoothstep easing so the ramp is flat at both ends, colour stops interpolated in Kotlin rather than left to the renderer, and a transparent stop that carries your own RGB instead of Color.Transparent. Use when a gradient overlay shows a hard line where it starts or ends, when the middle of a fade turns muddy grey or darker than either end, or when a fade that looks right on one platform bands into stripes on another.
Four small helpers worth carrying in a shared module, each with the one way it misleads — a symmetric set difference for diffing two id sets, a position index for constant-time lookups, a tolerant parse/serialize pair for timestamped tokens that returns null instead of throwing, and a translator that rewrites an external link into your own scheme. Use when a diff reports every item as changed, when a position lookup returns the wrong index for a repeated element, when one malformed line takes down a whole screen, or when pasting a link into a search box searches for the link.
A vertical slide transition defaults to HALF the element's height, so it appears already halfway through its own movement and the first part is missing — which reads as a pop, not a slide. Covers passing the full height, the sign that decides which edge it comes from, pairing a shorter fade with a longer slide so the element is opaque before it settles, making exit quicker than enter, and publishing the pair as shared values so every screen matches. Use when a bar or panel seems to snap into place instead of sliding, when enter and exit feel mismatched, or when the same control animates differently on two screens.
Why `x NOT IN (subquery)` matches zero rows and still reports success whenever a NULL is actually present in the subquery result — a standing risk for any nullable column — how to guard every such subquery, and how to make a silently-inert statement detectable instead of invisible. Reach for it when a DELETE or SELECT with a NOT IN filter returns nothing on data you can see with your own eyes, when a cleanup pass "succeeds" and frees nothing, or before writing any NOT IN over a nullable column.
Inset consumption travels to a composable's descendants and never to its siblings, so two inset-aware bars stacked in one column each reserve the system bar and open a band of dead space exactly one bar tall. Covers parameterising a bar's `windowInsets` with the framework default, deciding once who consumes, and why the same component must keep the default at its overlay call sites. Use when a strip of empty space appears between two bars, when it only shows in one mode of a screen, or when a bar's leading icon is clipped in landscape after you zeroed its insets.
A timestamp column written through an ORM type converter can hold the local wall clock encoded as if it were UTC — an exact round trip that is only correct through the converter, and the converter is chosen by the field's TARGET TYPE, so declaring a projection field as a raw number silently opts out and applies the offset a second time. Covers why every total still adds up, why the error is exactly zero on some machines, and why the fix is asking for the type the converter understands rather than picking a time zone. Use when an hour-of-day or day-of-week breakdown is shifted by your own offset while every count and sum is right, when a chart says people are most active at 3am, or before typing a stored time column as a number in a hand-written projection.
The multiplatform resource formatter substitutes plain positional placeholders and nothing else — no flags, no width, no escaped percent — so padding, rounding, units and symbols belong in code and the resource only ever joins already-formatted pieces. Covers the same omission in its other two shapes: a raw stored number printed straight to screen, and a date-time library's month names that are constants rather than locale lookups. Use when a format specifier renders verbatim on screen, when a label appears in English regardless of language, or when a screen prints a number in the unit the database happens to store.
Reading a response whose shape drifts — classify each field by the marker the payload itself declares rather than by its position, treat a filtering map as data loss and count what it drops, refuse to substitute a placeholder for a failed parse, and parse composite strings from their stable end. Use when a parser works in one locale and not another, when a list arrives shorter than the source shows, or when a made-up value turns up somewhere it was never entered.
Ship a build with crash reporting and a build with none from one codebase, by swapping a module that exposes three top-level functions instead of an interface, so call sites are byte-identical and the no-tracking build provably contains no reporting code. Also covers a desktop crash dialog built on the older widget toolkit, because the modern UI may be exactly what just died, and how to marshal it onto that toolkit's event thread. Use when a privacy build must contain no reporting dependency at all, when a swapped implementation is drifting from its counterpart, or when the app dies with no visible error and no way for a user to send you the details.
A conflating state holder keeps one slot, so a callback that writes it several times per event makes write ORDER the correctness question — collectors see only the last write. Use when a spinner sits over content that is already loaded, when a loading flag clears at the wrong moment, or when a screen shows the state that was true one step ago.
A story-style reel — a pager that advances itself on a per-card timer, with a segmented progress bar, tap zones to skip forward or back, and a press-and-hold that pauses it — where every card's data is computed once before the first frame instead of card by card, and the segmented bar's count comes from a card list some years never fill completely. Covers a frame-delta timer that a long hold cannot bank progress against, why an empty onLongPress callback is what makes hold-then-release resume instead of navigate, reading the pager's target page rather than its current one inside a tap handler, and pinning a captured card's colour scheme so it renders the same regardless of the viewer's own theme. Use when a reel's progress bar jumps to the wrong segment, when releasing a paused hold immediately skips a card instead of resuming it, when a loading spinner appears mid-story instead of only before the first card, or when a shared card looks different depending on which theme the device was in.
One list row carrying three gestures at once — tap, long-press-to-select, and swipe-sideways-for-an-action — plus a mode flag that changes what the tap means. Covers the pointerInput key that decides whether the swipe detector sees the current mode or the one captured at composition, translating the row in the layout phase, latching the commit threshold, and leaving the opposite drag direction to the parent. Use when a row keeps swiping after multi-select has started, when selection only begins working after the row happens to recompose, or when a swipe fights the pager or list underneath.
One bridge joins a local component to a shared room, with the direction of travel decided by role — the source publishes what it does, the follower applies what arrived and publishes nothing — plus a flag held across the apply so a locally-observed side effect of a remote command is not fed straight back. Use when two clients in a synchronised session ping-pong each other, when applying a remote pause immediately republishes a pause, or when a follower's own reactions fire on state it did not cause.
Sweep a travelling highlight through a label by putting a moving gradient on the TextStyle itself — the glyphs are painted by the brush, so there is no overlay, no clip and no measured width to keep in sync. Covers declaring the infinite transition unconditionally so the sweep does not restart every time the label appears, why the sweep head must be a pure high-contrast colour rather than the label's own, why the gradient stops are pixels and must travel past both ends, and that a brush replaces the text colour outright. Use when a shimmering label jumps back to the start whenever it reappears, when the gleam is invisible against the label's own grey, or when setting a brush makes a carefully chosen text colour vanish.
A see-through selection chip — a drop shadow under a transparent shape shows THROUGH it as a dark ring, and zeroing the resting elevation does NOT remove it, because each interaction state carries its own independent token default. Covers dropping the selection-only leading icon that widens the chip by 26dp and reflows a scrolling row under the finger that just tapped it, switching the outline off when the fill arrives, and pairing a role colour with its own "on" token. Use when a transparent chip has a dark halo that comes back on press or hover, when tapping a chip shifts its neighbours sideways, or when a selected chip's label is unreadable on one of the two themes.
Soften an app-wide touch ripple by alpha alone and give it the right bounds — the colour derives from the local content colour and stays correct inside a forced-scheme subtree, while pinning one paints darker than a near-black surface. Covers why the theme's configuration reaches a bare Modifier.clickable, why .clip(shape) must come before .clickable, why a card and its clip need one shape value, why two clickable modifiers must never stack, and the deprecated constructor that is the only way to set alpha. Use when a tap on a rounded item flashes a square, when the ripple reads as a sooty smudge on a dark theme, or when a long-pressable item ripples twice.
A small keyed cache for values that drift — each entry carries the moment it was fetched and answers an isStale check against a time-to-live constant, the whole map is stored as one JSON string in key-value preferences, and decoding is lenient plus wrapped so a schema change degrades to a cache miss instead of destroying every entry. Use when caching resolved covers, lookups or per-key results without a database table, or when a cache stopped working entirely after a model field was added.
Handling a transitive dependency whose strict version constraint overrides the version you chose, dragging a shared lower-level library up or down for the whole build — how to find who pinned what, when to force a version back versus align everything with the pin, and how to document a pin so nobody upgrades it back into the breakage. Reach for it when adding one unrelated library produces a missing-method failure at runtime, inside a rendering pass, in a component you did not touch.
Two vendors ship the same package name at different versions into different source sets of one multiplatform build, so a member function that one vendor has already turned into a top-level extension resolves on exactly one target — a specific import compiles for Android and fails for desktop, or the reverse. Covers spotting the duplicate coordinate, why a wildcard import is the correct fix here rather than a smell, and the pinning discipline that keeps the pair readable. Use when shared UI code stops compiling on one target only after a routine dependency bump, when an unresolved-reference error names a symbol you can plainly see in the other target's sources, or when two catalog entries carry the same artifact name.
Lay out a type-safe Navigation-Compose graph so it stays readable as the destination list grows — one serializable route object per file grouped by area, area graphs as extension functions on the graph builder, a single transition set on the host, and per-route theming applied by wrapping the screen inside its own entry. Use when a navigation file has grown to hundreds of lines, when route arguments start needing custom types, or when transitions or theming differ between destinations for no stated reason.
A parse-failure fallback must be a sentinel outside the legal domain, or expressed in the type — never a value the success path can also produce. Expose "not known" as its own question. Use when a field means two different things depending on where it came from, when a placeholder reaches the screen, or when a consumer cannot tell absent from measured.
The order a WebSocket session has to be brought up and torn down — reader started before the first message because the answer comes back through it, the handshake settled on a deferred with a timeout, the close frame sent under a non-cancellable context, and an event buffer that suspends rather than drops. Use when a socket connects but the session never becomes usable, when a deliberate disconnect leaves the peer thinking you are still there, or when clients drift out of sync after a burst of traffic.
When two looks of one screen each need a fit-exactly-one-screen measurement — measure the fixed blocks, split the remainder into equal gaps, floor it at a minimum — keep a copy per look instead of hoisting one; covers the effect keys the block needs, the invisible spacer that must mirror the ratio actually drawn, and which spacer may animate. Use when the gap above or below a hero element keeps last track's size, when content that should end at the fold overflows or leaves a band of dead space, or before extracting "the same" layout maths from two screens into one helper.
Ship a self-signed MSIX (the modern Windows application package format) that end users can actually install without a hosted update site — bundle an install script plus the signing certificate beside the package, script the trust-then-sideload steps, and keep the signing key stable across releases; reach for it when double-clicking the packager's output fails, when its wrapper installer dies fetching a URL that returns 404, or when a new build refuses to install over the previous one.
A weighted child occupies its whole slot even when its content measures narrower, which pins the sibling beside it to the far edge; `weight(1f, fill = false)` releases the unused width back to the row's arrangement so child, gap and sibling read as one cluster. Covers why the tell is the sibling rather than the weighted child, why the fix is two edits and not one, and why dropping the weight instead is a different layout. Use when a bar looks centred at one content size and lopsided at another, when a button clings to the screen edge with a hole beside it, or when a trailing label sits far from the text it belongs to.
Detect that a desktop app is running inside a virtual machine on Windows after the classic command-line management query tool was removed in Windows 11 — query the management layer through PowerShell, probe both the manufacturer and the model field, and pick the fail direction deliberately; reach for it when a transparent or undecorated window renders nothing on a VM while the process keeps running, or when a detection probe that worked for years suddenly reports empty on every modern host.
A per-word karaoke wipe driven straight off a ticking time source looks stepped instead of smooth, a word that was already fully sung stays lit after the user seeks backward past it, or skinning an existing line-level lyrics renderer for word-level highlighting leaves an unsynced sheet glowing white end to end. Use when building or debugging word-by-word lyric highlighting, a synced-transcript view, or any left-to-right text "fill" effect driven by a playback clock.
Sequence a large XML-to-Compose migration — hardest screen first, expect the real work to be consolidating scattered state rather than swapping widgets, and plan for navigation to drag a route-serialization migration in with it. Use when planning a multi-release UI migration, when the layout file count refuses to go down despite screens "being migrated", or when deciding which screen to convert next.
Write an agent skill that survives an adversarial review — a description carrying both the trigger and the error symptom the reader is staring at, a short orientation, a Traps section that dominates the file, and verification commands you have actually run. Use when authoring or reviewing a SKILL.md, when a skill reads like documentation instead of hard-won advice, or when review keeps finding claims the source repository does not support.
A top-N query is right for a list and wrong for a share-of-the-whole — the cut tail shrinks the denominator and the entropy normaliser, inflating concentration and diversity alike — so the same grouped data needs two queries with different bounds. Covers why the truncation is invisible in the result, why the unbounded query's ORDER BY becomes load-bearing, and when unbounded is actually safe. Use when a "top 5 share" or diversity score reads implausibly high, when one grouped query is feeding both a leaderboard and a statistic, or before reusing a capped DAO method for anything that divides by a total.
Use when one or more photographs must become adaptive photo-plus-abstraction editorial compositions while source facts, spatial relationships, and strict photo preservation remain controlled.
> Audits and improves GitHub repository quality — README structure, community health files, .github directory setup, issue/PR templates, metadata, releases, and branch hygiene. Use this skill whenever the user asks to improve, audit, review, or set up a GitHub repository, or when they mention things like "make my repo look professional", "add contributing guidelines", "set up issue templates", "improve my README", "clean up my repo", "prepare my repo for open source", or "make my GitHub project look good". Also trigger when working inside a git repository and the user asks about best practices for documentation, community files, or repository structure. If the user is in a repo directory and mentions README, CONTRIBUTING, SECURITY, CODE_OF_CONDUCT, LICENSE, .github, templates, or release tags, use this skill.
|
> Runs a readability audit on a blog post draft or other multi-paragraph prose, calibrated for readers who read English as a second language. Checks ten categories — overall structure and topic order, paragraph structure, opening paragraph strength, tiered sentence length, passive voice, difficult words, filler and hedging, transitions, variation, and heading hierarchy — and reports a Flesch Reading Ease score with a per-category status. Use when the user asks to check readability, run a readability pass, or asks "is this readable", or proactively as a second pass after a substantial draft is complete. Also invoked by the github-repo, github-profile, and wp-readme-optimizer skills on their generated prose. For short strings (titles, meta descriptions, taglines, bios), use the `metadata-check` skill instead — Flesch and paragraph-level checks don't apply to them.
> intent fit, focus keyphrase placement, E-E-A-T signals (experience, expertise, authoritativeness, trustworthiness), helpfulness and originality, content freshness, and internal linking. Use when the user asks "will this rank", "SEO check this post", "content SEO audit", or "optimize this post for search" — or proactively after a readability pass on a post that's meant to rank. For how the prose reads, use the `readability-check` skill; for titles and meta descriptions, use `metadata-check`; for site-wide technical SEO (sitemaps, structured data, canonicals), use `astro-seo` or `static-seo`.
> Reviews short high-value strings — page titles, meta descriptions, schema description fields, FAQ answers, GitHub repo taglines, profile bios, social-card copy, and other metadata where Flesch and paragraph-level readability checks don't apply. Checks front-loading, concreteness, filler, active voice, title/description duplication, difficult words, SERP-truncation fit, and one-idea-per-field. Use when the user asks to review metadata, a tagline, a bio, or SEO strings, or proactively after generating any short string with audience-facing impact. Chained into by astro-seo, wp-readme-optimizer, github-repo, and github-profile for their metadata outputs. For multi-paragraph prose, use the `readability-check` skill instead.
> Audits and improves SEO for Astro sites. Use when the user asks to audit, set up, or improve SEO on an Astro site, or mentions head metadata, structured data, JSON-LD, sitemaps, IndexNow, Open Graph images, schema endpoints, NLWeb, hreflang, or search engine indexing in an Astro project. Produces drop-in code routed through `@jdevalk/astro-seo-graph` and chains into `metadata-check` for generated SEO strings.
> Audits and optimizes GitHub profile pages — profile README, metadata fields, pinned repositories, stats widgets, and contribution visibility. Use this skill whenever the user asks to improve, create, review, or optimize their GitHub profile, or mentions "profile README", "GitHub bio", "pinned repos", "GitHub stats", "contribution graph", or "GitHub presence". Also trigger when someone says things like "make my GitHub look good", "I want a better GitHub profile", "help me stand out on GitHub", "set up my GitHub page", or "optimize my developer profile". Works for both personal profiles and organization profile pages.
> Audits and improves SEO for static HTML sites. Use when the user asks to audit, set up, or improve SEO on a static site (Hugo, Jekyll, 11ty, Gatsby, Next.js static export, hand-rolled HTML, or `wp-static-clone` output), or mentions head metadata, structured data, JSON-LD, sitemaps, IndexNow, Open Graph images, schema endpoints, NLWeb, hreflang, or search engine indexing for a static site. For Astro projects, use `astro-seo` instead — its recipes produce less hand-rolled boilerplate by routing through `@jdevalk/astro-seo-graph`.
|
> Clones a live WordPress (or other CMS-driven) site into a static HTML site deployable on any static host (Cloudflare Pages, Netlify, Vercel, S3+CloudFront, plain Apache/nginx). Use when the user wants to "scrape", "freeze", "archive", "static-ify", or "move to [host]" a WordPress site, or asks to turn a sitemap into deployable static HTML. Pulls every URL from sitemap_index.xml, fetches all assets, rewrites paths to be root-relative, strips WP runtime markup, and outputs a flat directory ready to deploy with no build command.
> Reviews and rewrites WordPress.org plugin readme.txt files for maximum quality. Use this skill whenever a user pastes, uploads, or references a WordPress plugin readme.txt, or asks to improve, audit, review, score, or optimize a plugin's WordPress.org listing page. Also trigger when the user says things like "make my plugin page better", "optimize my readme", "help me rank higher on WordPress.org", or "review my plugin listing". Always run the full audit + rewrite workflow unless the user explicitly asks for only one part.
|
書籍「おい、とりあえず終わらせろ」の5ステップをもとに、仕事・学習・家事・個人制作などで止まった場面を責めずに省察し、見えていなかった前提と次の一歩を見つける対話型Skill。先延ばし、完璧主義、着手や提出への怖さ、試した後の停滞を整理したいときに使う。単純な実行、説明、軽微な修正、網羅的な計画づくりには使わない。
A lightweight, risk-scaled workflow for AI-assisted embedded firmware development. Use when onboarding an embedded project, investigating or implementing firmware changes, working around generated code, handling shared interfaces, protocols, Flash/NVM, boot, DMA, ISR/RTOS concurrency, hardware bring-up, recovery, or target verification. Supports minimal user input and structured Issue-style tasks.
Turn the questions and instructions a user keeps typing into a project into short slash-command skills, mined from their real Claude Code and Codex session history and git log. The shortcuts load in Claude Code, Codex and Cursor. Use when the user says they ask the same things every day, wants an FAQ or shortcuts or abbreviations for a project, wants to "stop writing the whole command", or asks what they repeat most.
| Write algorithm-optimized captions/descriptions for social media posts and reels. Supports Instagram (Reels, Feed, Carousel), TikTok, Threads, Facebook, and YouTube Shorts. The user sends material (text, image, script, competitor post, screenshot) and gets ready-to-use captions for ALL platforms at once, in English. Use when asked to "write a caption", "write a description", "caption for instagram", "описание к посту", "подпись для рилса", "описание для тикток", "caption for reels", "social media post", or any caption/description writing task. Also triggers when user sends an image, screenshot, or text and says something like "сделай описание", "напиши подпись", "caption this", "описание для этого".
> Reference guide for connecting popular APIs to Home Assistant via Node-RED, YAML, or custom energy APIs (Tibber, Nordpool), weather (SMHI, OpenWeatherMap, yr.no), transport (SL, Trafikverket, Resrobot), smart home clouds (Shelly, Tuya, Philips Hue, IKEA), and global APIs (OpenAI, Spotify, Google Calendar, Telegram, GitHub). Use this skill whenever the user mentions a specific external service, API, or data source they want to connect to Home Assistant - even if they don't say "API".
> Node-RED visual automation flows for Home Assistant. Covers visual automation, flow JSON, trigger-state, api-call-service, function nodes, context storage, timer patterns, error handling, and node-red-contrib-home-assistant-websocket nodes. Use when user mentions "Node-RED", "flow", "visual automation", or "node-redflöde".
> ESPHome device configuration, firmware, and IoT product development. Covers ESP32, ESP32-S3, ESP32-C3, ESP32-C6, ESP32-H2, ESP32-P4, ESP8266, RP2040, RP2350, nRF52, LibreTiny, Shelly, Sonoff, Tuya, BLE proxy, Matter firmware, Thread, Zigbee, GPIO, sensor YAML, LVGL displays, LED strips, voice assistant hardware, device flashing, Arduino conversion, alarm_control_panel, lock, valve, media_player, microphone, speaker, audio DAC, event entities, datetime entities, Z-Wave proxy, MIPI DSI displays, and DLMS smart meters. PCB design (KiCad), enclosures, 3D printing, CE/FCC certification, BOM optimization, and manufacturing from prototype to production scale.
> Home Assistant YAML configuration and automation. The most common skill for general HA questions. Covers automations, blueprints, scripts, scenes, template sensors, Lovelace dashboards, Mushroom cards, packages, helpers, presence detection, voice Assist, calendar automation, Jinja2 templates, notification patterns, energy monitoring, Matter devices, automation debugging and traces, dashboard sections view, and trigger-based template entities.
> Aurora Smart Home orchestrator — routing layer for all smart home skills. Use this skill when the user asks ANY smart home question and you need to decide which skill to invoke, or when a task spans multiple skills (e.g., "build a sensor that shows on a dashboard and triggers automations"). Invoke aurora FIRST before reaching for a specific skill — it will route to the right specialist(s) and recommend the correct Claude model to keep automation, IoT, dashboard, ESP32, Node-RED, or any request about controlling or monitoring devices at home.
> Beautiful, copy-paste-ready Home Assistant dashboard designs with complete CSS themes, Dark Minimal, Material You, Nordic, Neon/Cyberpunk, Warm Home, Soft Pastel, Luxury Gold, and Retro Terminal. Use this skill whenever the user wants to make their dashboard look good, asks about card styling, CSS for Home Assistant, card-mod, button-card templates, Lovelace themes, dashboard aesthetics, or wants a specific visual style - even if they don't use the words "design" or "CSS".
Claude Skills — questions
Answers built from the skills we actually parsed.