feat: デモシフトボードとイベント駆動ツアーの追加

.claude/skills/demo-ux/SKILL.md

@@ -0,0 +1,138 @@
+---
+name: demo-ux
+description: Best practices for designing, reviewing, or refining "try-it-now" interactive product demos, in-app product tours, onboarding walkthroughs, and sandbox experiences. Use when building or critiquing a demo page, tour component, tooltip copy, sandbox, aha-moment flow, reset/skip behavior, or any try-before-signup experience. Triggers on "demo UX", "ツアー設計", "ウォークスルー", "オンボーディング設計", "チュートリアル", "product tour", "walkthrough", "try it now", "サンドボックス設計", "デモのレビュー", or when reviewing files named `*Tour*`, `*Demo*`, `*Onboarding*`, `*Walkthrough*`.
+---
+
+# Demo UX Best Practices
+
+Knowledge base for designing try-it-now demos, product tours, and onboarding flows that convert. Grounded in 2024–2026 benchmarks (Navattic, Arcade, Chameleon, Userpilot, Appcues) rather than first principles — because most "design intuitions" in this space are wrong in ways only data reveals.
+
+## When to use
+
+- Designing a `/demo` or `/try` page from scratch
+- Reviewing or refining an existing tour / tooltip / sandbox
+- Deciding whether a tour should auto-start
+- Writing tooltip microcopy
+- Debating step count, skip behavior, or reset behavior
+- Choosing between guided tour, free sandbox, or hybrid
+
+## Core decisions (checklist)
+
+Work through these in order. Each has a default answer backed by data; deviate only with a specific reason.
+
+### 1. Step count — **default: 3 steps**
+
+- **3 steps hits 72% completion. 4 steps → 45%. 7 steps → 16%.** The cliff between 3 and 4 is real.
+- If the guided flow needs more than 3 steps, split it: 3-step tour + progressive disclosure (contextual tooltips fired later when the feature is actually used).
+- A tour that explains features *before* the user tries them is almost always wrong. Fold explanation into the moment the user is about to use the feature (event-driven tooltips).
+
+### 2. Start trigger — **default: user-initiated, not auto-start**
+
+- **User-initiated tours complete 2–3× more often** than auto-start.
+- Modal-on-arrival increases bounce. Don't cover the product on first paint.
+- Instead: land in the sandbox, show a visible-but-dismissible CTA inline ("最初の1日を整えてみる →" / "Start the tour").
+- If you must nudge: delay 30–60s or trigger on idle / scroll.
+
+### 3. Initial state — **default: pre-broken, not empty**
+
+- Empty states push time-to-value past the user's attention budget.
+- **Pre-broken > pre-populated > empty.** Grammarly (text with mistakes), Linear (seeded issues), Notion (getting-started workspace), Figma (community files) — none start empty.
+- The pre-broken state *is* the aha setup: the user fixes it, the aha lands.
+- Design the broken state so the problem is **visually obvious** — long bars, missing entries, visible conflicts. Don't require explanation.
+
+### 4. Progression — **default: event-driven, not click-Next**
+
+- "Click Next" tours let users advance without performing the value-creating action — the single biggest conversion killer in tour design.
+- Fire the next tooltip **when the user completes the action** (drag finished, item deleted, form submitted).
+- For non-action steps (welcome / summary), click-Next is fine.
+
+### 5. Tooltip copy — **default: 1 sentence, ≤150 chars, verb-first**
+
+- **≤150 chars, ~20–30 words, one idea per tooltip.**
+- **Verb-first, second person, imperative**: "Add your first shift" beats "You can add shifts here."
+- **Instructional > narrative.** Narrative fits marketing videos, not tooltips.
+- Point at the next action, not at the feature.
+- See `references/copy-patterns.md` for patterns and Japanese-specific nuances.
+
+### 6. Skip — **default: always visible, persistent, preserves state**
+
+- Skip control must be visible on every step. Forcing a tour == churn.
+- Show step counter next to skip so the user can choose "3 more → done" vs. "skip".
+- **Skipping should NOT reset the sandbox.** Let the user keep poking the state they've built.
+- B2B tour completion averages ~5%. Skip is the default behavior, not the exception.
+
+### 7. Reset — **default: reload-resets, plus a visible "Start over" button**
+
+- Reload should restore initial state (simplest, no surprises).
+- Additionally expose a visible "Reset demo" / "最初からやり直す" button in the page chrome.
+- Single-step confirmation ("Your demo progress will be cleared"). No typed confirmation — this is a sandbox.
+- Label as "Reset" / "Start over", not "Delete".
+
+### 8. End state — **default: inline CTA in the last step, not a full-screen modal**
+
+- Inline CTA in the last tooltip outperforms full-screen celebration for SMB flows.
+- Show 1–2 differentiated CTAs: low-commitment + high-commitment ("Keep exploring" + "Start free").
+- **"Learn more" CTRs ~63%; "Book a demo" ~41%.** High CTR ≠ high qualified conversion — match wording to intent.
+- Consider a persistent top-bar CTA for users who hit aha mid-flow.
+
+### 9. Mobile — **default: desktop-first with explicit fallback**
+
+- Interactive tours are desktop-first. That's the state of the industry.
+- **52% of top-1% demos ship an optimized mobile experience** — the rest redirect or email-capture.
+- Fallbacks: (a) simplified mobile-only demo, (b) looping video, (c) "best on desktop — email me the link".
+- If the demo uses drag/drop/precision hover, a mobile fallback is mandatory.
+
+### 10. Analytics — **instrument these 5 events minimum**
+
+1. `demo_loaded` — page opened
+2. `tour_started` — start CTA clicked (distinguishes auto vs. user-triggered)
+3. `step_completed` — per-step with step id. Finds the one step killing the funnel.
+4. `aha_reached` — custom event when the user completes the value-creating action
+5. `cta_clicked` — final signup / upgrade CTA
+
+Top-1% demos: final CTR 67%, demo-to-signup 8× median. If your numbers are far from these, the fix is usually step-count or progression style — not copy.
+
+## The persona lens
+
+Two meta-personas matter:
+
+| Persona | Prefers | Why |
+|---|---|---|
+| **Hands-on SMB / operator** (shop owners, ops managers, ICs) | Sandbox > guided. Short tours. Try-it-now > sales demo | SMB buyers "tolerate lighter, shorter tours" (SmartCue). Self-serve converts. |
+| **Exec / enterprise evaluator** | 90-second narrative video > interactive. Prefers demo-with-a-human | "Watch-to-evaluate" — Arcade-style video demos convert this persona better. |
+
+Building for both? Ship a 60–90s video demo alongside the interactive demo. Let the user choose.
+
+## Common anti-patterns (do not do)
+
+- 7+ step tours that front-load feature explanation
+- Auto-triggered modal on first page load
+- "Click Next" progression for action steps
+- Empty initial state ("Add your first X to get started")
+- Skip button hidden in an overflow menu
+- Multi-step reset confirmation
+- Full-screen celebration modal at end (over-ceremonious)
+- Narrative / marketing copy in tooltips ("Welcome to the future of X…")
+- Mobile "coming soon" dead-ends
+- No aha-reached event instrumented
+
+See `references/anti-patterns.md` for extended list with symptom → cause → fix.
+
+## When to reference deeper files
+
+- **`references/benchmarks.md`** — specific numbers to justify a design decision (completion rates, conversion benchmarks, TTV, CTR by wording). Full citations.
+- **`references/copy-patterns.md`** — when writing or reviewing tooltip / CTA / banner text, especially in Japanese. Includes verb-first patterns, 3 I's (Inform/Influence/Interact), Japanese copy nuances.
+- **`references/anti-patterns.md`** — when reviewing an existing demo for problems. Organized as symptom → cause → fix.
+
+## Output discipline (when reviewing or proposing)
+
+1. **Lead with data, not opinion.** "3-step tours hit 72%, 7-step tours hit 16% — your 7 steps is in the wrong zone" beats "this feels long."
+2. **State the default, then justify the exception** if the design deviates.
+3. **Don't invent tooltip copy** without showing the underlying pattern (verb-first, ≤150 chars, event-driven). Otherwise the user treats it as decoration, not a design artifact.
+4. **Always mention analytics.** "What 5 events are we instrumenting?" If the answer is "we'll figure that out later", the demo isn't done.
+
+## Scope boundaries
+
+- This skill covers **try-it-now demos, tours, and onboarding** — not marketing landing pages, not sales demos, not documentation sites.
+- For full-screen mock design, use `/create-design`.
+- For pinpoint UI polish unrelated to demos, use `/design-ideas`.

.claude/skills/demo-ux/references/anti-patterns.md

@@ -0,0 +1,117 @@
+# Demo UX Anti-patterns
+
+Organized as **symptom → cause → fix**. Use this file when reviewing an existing demo for problems.
+
+## Step count bloat
+
+**Symptom**: Tour has 7+ steps, each explaining a feature before the user tries it.
+**Cause**: Designer assumed "more explanation = better understanding." In practice, front-loaded explanation doesn't stick and users bail.
+**Fix**: Cut to 3 action steps. Move remaining explanations to contextual tooltips that fire when the user is about to use the feature. Treat the 3-step 72% completion benchmark as a hard ceiling.
+
+## Auto-start modal on page load
+
+**Symptom**: Landing on `/demo` immediately pops a modal that covers the product.
+**Cause**: "Immediate engagement" interpretation that ignores user autonomy.
+**Fix**: Render the sandbox first. Add a non-blocking inline CTA ("Start the 30-second tour →") in a dismissible banner or card. User-initiated completion is 2–3× higher.
+
+## Click-Next progression on action steps
+
+**Symptom**: Tooltip says "Try dragging this bar" but has a "Next" button the user can click to advance without ever dragging.
+**Cause**: Default implementation in most tour libraries. Easier to build than event-driven.
+**Fix**: Listen for the actual event (drag-end, click, submit). Advance the tour only when the user performs the action. Disable or hide "Next" on action steps.
+
+## Empty initial state
+
+**Symptom**: Demo opens to "Click + to add your first item" with nothing else on screen.
+**Cause**: Developer-brain: fresh-account experience == demo experience.
+**Fix**: Seed the demo with realistic, pre-broken content. The user should walk in and see a *problem they can fix*, not a blank canvas they need to populate.
+
+## Hidden skip / force-march tours
+
+**Symptom**: No visible skip control; user can only X out of the modal or close the browser tab.
+**Cause**: "Don't let them leave the tour" reasoning.
+**Fix**: Every step has a visible "Skip" / "自由に触る" control, adjacent to the step counter. Skipping preserves the sandbox state. B2B tour completion averages ~5% — designing for skip is designing for reality.
+
+## Skip resets the sandbox
+
+**Symptom**: User skips mid-tour, everything they did is wiped to initial state.
+**Cause**: Conflation of "skip tour" with "reset demo".
+**Fix**: These are two different operations. Skip leaves state alone. Reset (a separate button in page chrome) wipes state. Never collapse them into one.
+
+## Hard reset confirmation
+
+**Symptom**: "Reset demo" button opens a modal asking the user to type "RESET" to confirm.
+**Cause**: Over-application of production-data safety patterns.
+**Fix**: Single-sentence confirmation ("デモの操作内容がクリアされます。よろしいですか？") with a plain Yes button. This is a sandbox — reload already resets.
+
+## Narrative tooltip copy
+
+**Symptom**: Tooltips read like marketing copy: "Welcome to the future of scheduling! Our revolutionary approach lets you…"
+**Cause**: Copy written by a marketer or written before the rest of the tour was designed.
+**Fix**: Verb-first, instructional, ≤30 words, points at the next action. "Drag the bar to change 高橋さん's end time to 16:00."
+
+## Feature-label tooltip copy
+
+**Symptom**: Tooltip title says "Shift Editor" and body says "This is the shift editor."
+**Cause**: Writer explained *what it is*, not *what to do with it*.
+**Fix**: Rewrite as an action prompt. Title: "Edit a shift." Body: "Drag the bar to change the time."
+
+## 擬音語・抽象語の多用 (JP)
+
+**Symptom**: 「ぱっと伝わります」「さらっと見られます」のような曖昧な表現。
+**Cause**: 雰囲気の良い日本語を書こうとして具体性を失う。
+**Fix**: 具体的な動作・数字に置き換える。「スタッフ全員にメール送信されます」「30秒で要点が見られます」。
+
+## 体言止めを本文で多用 (JP)
+
+**Symptom**: 本文が「シフトの編集。バーをドラッグで時間変更。」のように体言止めの連続。
+**Cause**: タイトル向けの指針を本文にも適用した。
+**Fix**: 本文は「バーをドラッグして時間を変えられます。」のように述語で終わる文に。体言止めはタイトル・見出しだけ。
+
+## Full-screen celebration modal at end
+
+**Symptom**: Last tour step fills the screen with "🎉 Tour complete! You've mastered the basics!"
+**Cause**: Over-ceremony; trying to manufacture a "moment".
+**Fix**: Replace with an inline tooltip + 1–2 CTAs on the same spot where the last action happened. Less ceremony, more momentum. Inline CTA outperforms celebration modal for SMB flows.
+
+## Single end-of-tour CTA
+
+**Symptom**: Last step shows only "Sign up now."
+**Cause**: Funnel focus without thinking about lower-intent users.
+**Fix**: Two CTAs — low-commitment + high-commitment. "Keep exploring" next to "Start free." Capture users at whatever intent level they arrive at.
+
+## No mid-tour exit to CTA
+
+**Symptom**: User hits aha at step 2, wants to sign up, has to finish the tour or skip first.
+**Cause**: Tour flow is a sealed box.
+**Fix**: Persistent top-bar CTA throughout the demo. Catches the aha-and-go user (more common than you think).
+
+## Mobile dead-end
+
+**Symptom**: User on phone loads `/demo`, sees "Best viewed on desktop" and a back button.
+**Cause**: Defer-to-desktop assumption.
+**Fix**: At minimum, a looping video walkthrough. Better: a simplified mobile sandbox. Best: email-capture "send me the desktop link." 52% of top-1% demos ship an optimized mobile experience — the industry is moving on.
+
+## No analytics instrumented
+
+**Symptom**: Demo is live but nobody can answer "what % complete the tour? where do they drop off?"
+**Cause**: "We'll add analytics later" that never comes.
+**Fix**: Instrument 5 events on day 1: `demo_loaded`, `tour_started`, `step_completed` (per step), `aha_reached`, `cta_clicked`. Without them, you cannot improve the demo.
+
+## Demo treated as one-shot marketing artifact
+
+**Symptom**: Demo shipped, never iterated on.
+**Cause**: Demo viewed as launch content, not as a product surface.
+**Fix**: Review the demo funnel monthly. The step-with-highest-drop-off is the one to fix — fixing it typically yields +10–15% completion.
+
+## Demo data indistinguishable from real data
+
+**Symptom**: Users confused whether their demo actions are affecting a real account.
+**Cause**: No visual cue that this is a sandbox.
+**Fix**: Subtle but persistent "Demo mode" indicator in the chrome (top bar stripe, corner badge). Don't be so loud that it kills immersion; just enough for the user to know nothing is at stake.
+
+## Using real customer data as demo seed
+
+**Symptom**: Demo shows "田中太郎" from an actual user's account.
+**Cause**: Lazy seeding that reuses production queries.
+**Fix**: Use obviously-fictional names (「居酒屋シフトリ」「田中次郎」) and data shaped for the problem/resolution scenario. Privacy risk and weird authenticity both disappear.