について
このスキルは、レストランおよびホスピタリティソフトウェアを構築するための必須のドメイン知識を提供し、メニュー修飾子、在庫管理、財務的制約などの主要概念をカバーします。オンライン注文、予約、ロイヤルティプログラム、スケジューリングなどの中核製品を設計する際に、素朴なデータモデルを回避するために使用されます。レストラン固有の機能に関するアーキテクチャ計画や機能仕様策定の際に適用してください。
クイックインストール
Claude Code
推奨npx skills add avelikiy/great_cto -a claude-code/plugin add https://github.com/avelikiy/great_ctogit clone https://github.com/avelikiy/great_cto.git ~/.claude/skills/vertical-restaurantsこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
Vertical: restaurants & hospitality — don't spec it naive
Restaurants run on razor-thin margins (net 3–6%) with a hostile incumbent
stack. A spec that treats a menu as a flat list of {name, price} or
ignores who already owns the POS will ship something no operator can use.
This skill loads the domain so architect/pm sound like they've worked a
shift.
The 4 products in this vertical:
| Product | Archetype | One-liner |
|---|---|---|
| online-ordering | content | Own menu + checkout for dine-in/pickup/delivery — dodge aggregator fees |
| reservations | booking | Bookings, tables, text-the-waitlist |
| loyalty | crm | Points, offers, win-back |
| shift-scheduling | booking | Rota, open shifts, swaps with coverage rules |
Incumbents to position against: Toast (POS, ~$69–165/mo + hardware + 2.49%+ per swipe), Square (POS/SMB), SevenRooms (reservations/CRM, upmarket), ChowNow (commission-free ordering), DoorDash / Uber Eats / Grubhub (aggregators, 15–30% commission per order).
1. Domain vocabulary (use these words in the spec)
- COGS / food cost % — cost of ingredients ÷ menu price. Target ~28–35%.
- Prime cost — food cost + labor cost; the number operators obsess over (target ≤ ~60% of sales).
- Menu engineering — classifying items by popularity × margin into stars (high/high), plowhorses, puzzles, dogs (low/low). Drives what gets promoted or cut.
- 86'd — an item is out of stock / unavailable ("we're 86 on the salmon"). Must propagate instantly to every ordering channel.
- Modifiers / mods — choices on an item (size, temp, add bacon, no onions, sub fries). Grouped, with required/optional + min/max rules.
- Covers — number of guests served (a "200-cover night").
- Turn time — how long a table is occupied; reservations math depends on it (a 2-top turns in ~75 min).
- FOH / BOH — front of house (servers, host, bar) / back of house (kitchen, prep, dish). Scheduling and tips differ between them.
- Tip pooling — pooled tips split by rule (hours, role, points). Legally constrained — see §6.
- Comps / voids — comp = item given free (manager discretion); void = item removed before it's made. Both need audit trails.
- Ticket times — elapsed time from order fired to served; the kitchen's core SLA.
- Third-party aggregator commission — the 15–30% DoorDash/Uber Eats/ Grubhub take. The pain that makes owned ordering a wedge.
- KDS (kitchen display system) — screen in the kitchen that replaces paper tickets; orders route to it by station.
2. Non-obvious domain rules
- The POS is the sticky system of record — don't fight it. Toast/Square own the menu, payments, and floor. Our products integrate with or sit beside the POS; they don't try to replace it. Sync the menu, don't fork it.
- Aggregator commission is the wound; owned online-ordering is the wedge. A restaurant paying 25% to DoorDash on a $40 order keeps $30. Commission- free direct ordering is the single clearest ROI pitch — lead with it.
- Menus have deep modifier hierarchies, not flat prices. "Burger" → size group (required, choose 1) → temp group (required) → add-ons (optional, 0–5) → side (required, choose 1, sub upcharges). Price = base + mods.
- 86'd / out-of-stock is real-time and must sync everywhere. When the kitchen 86's an item it must vanish from online ordering, KDS, and the POS simultaneously, or you sell what you can't make.
- Tips have legal handling. Pooling rules, who can share (FLSA bars managers/owners from tip pools), tip credit, and service-charge vs tip distinction are labor-law constrained, not free-form.
- Margins are razor-thin. A feature that adds 30¢/order of cost can erase the margin on that order. Cost-consciousness is a feature, not a nicety.
- Reservations + waitlist are SMS-driven. "Your table's ready" is a text, not an email. Waitlist quote times and ready-pings are the product.
3. What a naive build gets wrong
- Flat menu, no modifier hierarchy.
{name, price}can't express "medium, well-done, add bacon, sub fries (+$2)". Model modifier groups with required/optional + min/max from day one. - Ignoring 86'd / out-of-stock. Selling a sold-out item online is a refund, an angry guest, and a chargeback. Stock state is first-class.
- Online ordering that doesn't sync the menu. A second menu that drifts from the POS menu means wrong prices and phantom items. One source of truth, synced.
- Tip handling that breaks labor law. Letting managers into the pool, or mislabeling a service charge as a tip, is an FLSA violation, not a bug.
- No dine-in vs pickup vs delivery distinction. Each channel has different fulfillment, timing, fees, address/table data, and tax. One generic "order" type is wrong.
- Loyalty that's points-only with no win-back. Points without a lapsed- guest re-engagement flow (offers, "we miss you") leaves the highest-ROI CRM lever on the table.
4. Must-model entities
- MenuItem — base price, category, station, tax class, availability state (available / 86'd / scheduled), with one or more ModifierGroups.
- ModifierGroup —
{required: bool, min, max}+ ordered Modifiers (name, price delta, default, in-stock). Hierarchy, not a flat list. - Order — channel (
dine_in | pickup | delivery), status (placed → confirmed → preparing → ready → completed | cancelled), line items with resolved modifiers, computed total, table/address per channel. - Reservation — party size, time, turn-time estimate, table assignment, status; plus Waitlist entry (quoted wait, ready-ping, SMS thread).
- Shift — role (FOH/BOH), start/end, coverage rule (min staff per role/time), open shift + swap request with approval/coverage check.
- LoyaltyMember — identity (phone-first), points balance, earn/redeem ledger, last-visit (for win-back segmentation), consent state.
5. Per-product notes (wedge + the one domain thing)
- online-ordering (content) — Wedge: commission-free direct ordering vs DoorDash's 15–30% and Toast Online Ordering's per-order fee. The one thing: the menu + modifier hierarchy must sync from the POS and honor 86'd state, across dine-in/pickup/delivery, or it's worse than the aggregator it replaces. Menus rank locally → see [[local-seo]].
- reservations (booking) — Wedge: SevenRooms is upmarket/expensive; give SMBs bookings + waitlist without the price tag. The one thing: it's SMS-first — text-the-waitlist and ready-pings are the product; turn time drives table availability. SMS consent → [[lifecycle-messaging]].
- loyalty (crm) — Wedge: most POS loyalty is points-only; we add offers + win-back. The one thing: lapsed-guest re-engagement (segment by last-visit, send an offer) is where the revenue is — design win-back, not just an earn-points counter. Sends → [[lifecycle-messaging]].
- shift-scheduling (booking) — Wedge: rota + open shifts + swaps cheaper/simpler than the incumbents. The one thing: swaps must enforce coverage rules (min staff per role per time band) — an unconstrained swap that leaves the line uncovered is the failure mode.
6. Compliance (light — defer the heavy lifts)
- Tip pooling / labor law (FLSA) — managers/owners may not share in tip pools; keep tip vs service-charge distinct; honor tip-credit rules. State law varies (e.g. CA). Surface as a constraint; get specifics confirmed.
- Food-allergen disclosure — the big 9 US allergens must be declarable on menu items; some jurisdictions require menu labeling. Model allergen tags on MenuItem.
- SMS consent for waitlist/loyalty — TCPA consent, STOP/HELP, quiet hours apply to every text. Defer the mechanics to [[lifecycle-messaging]].
- Payment — PCI scope, SCA, refunds/chargebacks. Defer billing/payment design to the billing/PCI track; don't hand-roll card handling here.
Cross-refs: [[lifecycle-messaging]] (every SMS/email leg — consent + deliver- ability), [[local-seo]] (menus and reservations rank locally; "best tacos near me" is the funnel), [[migration-ready-schema]] (importing an existing menu / guest list from the incumbent POS without losing modifier structure).
GitHub リポジトリ
よくある質問
vertical-restaurants Skillとは何ですか?
vertical-restaurants はavelikiy が作成した Claude Skillです。Skillは、Claudeが必要に応じて読み込む指示とリソースをまとめ、追加の指示なしで vertical-restaurants に関連するタスクを実行できるようにします。
vertical-restaurants をインストールするには?
このページのインストールコマンドを使用してください。vertical-restaurants をプラグインとして Claude Code に追加するか、リポジトリを skills ディレクトリにクローンし、Claudeを再起動してSkillを読み込みます。
vertical-restaurants はどのカテゴリに属しますか?
vertical-restaurants は その他 カテゴリに属します。
vertical-restaurants は無料で利用できますか?
はい。vertical-restaurants は AIMCP に掲載されており、無料でインストールできます。
関連スキル
LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。
このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。
このClaudeスキルは、スポーツベッティング市場(スプレッド、オーバー/アンダー、プロップベットなど)を分析し、過去の傾向や状況統計を検証することでバリューベットを特定します。教育目的のための実践的な提案を構造化されたマークダウン形式で出力します。開発者はスポーツベッティング分析ツールとして本機能を活用できますが、娯楽および教育目的に限定されている点に留意してください。
このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。
