本文へ移動
cccskills
無料GitHub で公開

cinematic-website-prime-directive

One-line-prompt → cinematic, gorgeous, functional, well-tested, deployed website. Pre-hydrated SPA + full PWA kit + JSON-LD rich snippets + third-party integrations. React 19+Vite default (Angular optional). 100 concrete improvements grouped into 10 categories that EVERY single-prompt site build must satisfy before being marked done. Trigger whenever the user says 'make a website for X', 'build a site for X', 'rebuild X.com', or any equivalent one-liner.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md14.4 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Cinematic Website Prime Directive

One-line prompt → satisfy all 100 build-breaking rules across 10 categories before DONE.

Start here: website-build-manifest is the cross-linked acceptance index for the whole one-prompt build (phase order, per-page/per-site gates, architecture, quality bar, EMDASH_* config). Load it first, then this skill's 100 rules + the detail rules it links.

1. Cinematic Visual Doctrine (1-10)

  1. Dark-first (#060610 base) + bold accent (#00E5FF / #7C3AED / #50AAE3) — never neutral grey
  2. Fluid clamp() type — body 16-19px, display 64-128px, ratio 1.25-1.333
  3. Sora + Space Grotesk + JetBrains Mono (variable woff2 subsets preloaded)
  4. Layered surfaces: opaque base + semi-transparent panels + glassmorph cards via backdrop-filter
  5. OKLCH palette w/ color-mix(in oklch, ...) derived shades — no static rgb tints
  6. text-wrap: balance headings, text-wrap: pretty body — no orphans
  7. One signature moment per page — motion (scroll-driven parallax, view-transition swap, popover anchor), a bespoke composition, or a typographic hero. Medium follows concept; "cinematic" is not one visual recipe. The canvas nebula (#114) is a reference impl, never a mandate. See reference/aesthetic-quality-lexicon.md.
  8. Hero MUST use AI-generated brand-aligned image, never stock placeholder
  9. Asymmetric hero grids, never centered-stack default
  10. Every section pair has visual rhythm: dense → breath → dense

2. Pre-Hydrated SPA + Routing (11-20)

  1. React 19 + Vite + Tailwind v4 + TanStack Router (default)
  2. Astro 5 + React islands for marketing-static-heavy
  3. Next.js 15 App Router only when SSR/ISR adds real value
  4. Angular 22 + Ionic + Spartan UI only when user says "Angular" or signal-heavy enterprise
  5. View Transitions API (@view-transition { navigation: auto; }) for SPA page swaps
  6. Selective hydration per island (Astro) or RSC boundaries (Next)
  7. Prefetch on link hover via TanStack Router or Astro prefetch="hover"
  8. Speculation Rules <script type="speculationrules"> prerender eagerness:"moderate"
  9. Route code-split via React.lazy + Vite manualChunks — no single chunk >250KB gz
  10. Static HTML shell prerendered w/ H1 + meta tags — never script-injected

3. PWA Kit (21-30)

  1. site.webmanifest w/ name, short_name, description, theme_color, background_color, display:"standalone"
  2. screenshots[] ≥3 entries w/ form_factor:"wide" + form_factor:"narrow"
  3. shortcuts[] for top 3 actions, share_target, file_handlers, protocol_handlers
  4. Maskable icons 192/512 + monochrome notification icon
  5. Workbox v7 SW: StaleWhileRevalidate HTML/JS, CacheFirst + ExpirationPlugin images
  6. offline.html shell w/ branded copy + link to cached pages
  7. A2HS install banner via beforeinstallprompt + iOS separate UI
  8. iOS splash via apple-touch-icon 180×180 + apple-touch-startup-image
  9. iOS Safari 18.4 Declarative Web Push opt-in (PWA-installed only)
  10. Kill-switch SW (unregister + clear caches) as fallback

4. Structured Data + Rich Snippets (31-40)

  1. 5+ JSON-LD blocks per route minimum
  2. Always: WebSite + Organization + WebPage + BreadcrumbList + FAQPage
  3. Page-type add-ons: LocalBusiness, Product, BlogPosting, HowTo, Person
  4. Author schema (Person + sameAs) + dateModified on every content page (EEAT)
  5. JSON-LD facts MUST also appear as visible HTML body text
  6. Validate every route against Google Rich Results Test API before deploy
  7. Schema.org citation: CreativeWork[] array for every quantitative claim
  8. Itemprop attributes mirror JSON-LD on key elements
  9. No duplicate @id across blocks; use stable IRIs
  10. inLanguage on every block; isAccessibleForFree:true on public content

5. SEO + AI Search / GEO (41-50)

  1. Title 50-60 chars HARD, meta desc 120-156 chars HARD, keyphrase 0.5-3%
  2. OG card 1200×630 ≤100KB BRANDED (not scraped photo)
  3. Twitter Card meta on every route
  4. sitemap.xml w/ <lastmod> on every URL
  5. robots.txt triple-state: Allow OR Disallow per GPTBot, ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot, Google-Extended, CCBot, Bytespider — never default
  6. humans.txt, .well-known/security.txt, browserconfig.xml, llms.txt (OPTIONAL)
  7. Quotable answer blocks 40-60 words per page (LLM citation magnet)
  8. FAQPage w/ 5+ Q&A blocks (highest AI citation rate)
  9. Lead paragraph answers query directly in <40 words
  10. Per-route canonical URL (custom hostname when primary_hostname set)

6. Third-Party Integrations (51-60)

  1. GA4 + GTM snippet (head + noscript fallback)
  2. PostHog snippet w/ persistence:'memory' (cookie-free), autocapture on
  3. Sentry @sentry/cloudflare v11 (stream trace lifecycle default since 11.0) + withSentry + OIDC DSN
  4. Workers Tracing [observability] enabled = true + OTLP export to Axiom
  5. AI Gateway binding for every LLM call
  6. Turnstile invisible (data-appearance="interaction-only") on every form — NEVER visible
  7. Stripe checkout / Resend transactional / Clerk auth (M2M JWT) stubs where appropriate
  8. Calendly / Cal.com inline booking when applicable
  9. Google Maps embed via Maps Embed API, lazy-loaded
  10. Live chat (Crisp / Plain) only when support is stated goal. Cmd+K opens AI chat or command palette AND focuses text input same frame — autofocus + requestAnimationFrame(() => inputRef.current?.focus({preventScroll:true})) after open-state flip, re-press = re-focus + select(), Esc returns focus to trigger. Playwright build gate: await page.keyboard.press('Meta+K'); await expect(input).toBeFocused();

7. Performance + Core Web Vitals (61-70)

  1. LCP ≤2.0s (target ≤1.5s)
  2. CLS ≤0.05 — every image, iframe, ad slot has explicit dimensions
  3. INP ≤100ms (target — 200ms is fail)
  4. fetchpriority="high" on LCP <img> AND its <link rel="preload">
  5. AVIF primary + WebP fallback + JPEG legacy — no JPEG XL
  6. Font woff2 subset w/ unicode-range per script
  7. JS budget ≤200KB gz / route, no single chunk >250KB gz
  8. CSS ≤50KB gz, fonts ≤100KB woff2 preload
  9. Largest image ≤200KB. Hero image ≤150KB.
  10. Long Animation Frames API instrumented for INP debugging

8. Accessibility WCAG 2.2 AA (71-80)

  1. axe-core 0 violations at all 6 bp (375, 390, 768, 1024, 1280, 1920)
  2. Contrast ≥4.5:1 body, ≥3:1 large text + UI components
  3. Target size ≥24px (2.5.8 — axe auto-tests)
  4. Focus appearance (2.4.11) visible at every interactive — manual review
  5. Focus not obscured (2.4.12/13) — sticky headers must not clip focused
  6. Dragging alternative (2.5.7) — every drag has click/tap equivalent
  7. Consistent help (3.2.6) — contact/help link same location every page
  8. Redundant entry (3.3.7) — never re-ask info given
  9. Accessible auth (3.3.8/9) — no cognitive-test CAPTCHA; offer passkeys
  10. Lighthouse a11y ≥95

9. Testing + Verification (81-90)

  1. Playwright E2E 6bp covering homepage → key flow → conversion
  2. Tests start at homepage, navigate by clicking nav (never page.goto for internal)
  3. Vitest 5 unit tests for utilities + Zod schemas
  4. Lighthouse CI: Perf ≥90, A11y ≥95, BP ≥95, SEO ≥95 (Lighthouse 13.x — since 13.3 an experimental Agentic Browsing category scores AI-agent navigation; perf weights TBT 30 / LCP 25 / CLS 25 / FCP 10 / SI 10)
  5. AI vision QA ≥9/10 per route (visual-qa agent w/ Opus 4.8)
  6. Percy AI Visual Review for full-page regression
  7. Yoast GREEN on every content page
  8. Build validator: every internal asset resolves, every hyperlink valid, every quantitative claim cited
  9. CSP report-uri capturing violations during E2E
  10. wrangler tail clean during smoke tests (no Worker errors)

10. Deploy + Observability + Done (91-100)

  1. Atomic deploy via Workers Builds (preferred) or wrangler deploy + version metadata
  2. Gradual deployment (1% → 10% → 100%) w/ auto-rollback on error spike
  3. CDN purge paired w/ every deploy
  4. /health endpoint returns {status, version, timestamp}
  5. D1 backups: Time Travel 30-day PIT + R2 long-term archive
  6. secrets.required validated at deploy
  7. Self-Verify Statement per route (URL + smoke test result + 3 screenshots)
  8. Sentry / PostHog / Workers Tracing / AI Gateway / GA4 all firing — verify in dashboard
  9. Browser console: zero errors, zero CSP violations, zero failed resources
  10. DONE = deployed at real URL + every Hard Gate green + Self-Verify per route + announced

Anything less than all 100 — plus always-on rules 111-118 (§12) — is in-progress.

11. Source-Site Enhancement (101-110 — fires when prompt names existing domain)

  1. Phrase detect — rebuild | optimize | enhance | modernize | recreate | clone | better version of {domain} OR bare-domain prompt where domain resolves 200 = enhancement mode. Output = union(SOURCE_URLS, STANDARD_PAGE_SET[org_type], DEMOGRAPHIC_LOCALES, JEWELS) − CRUFT_URLS. See rules/source-site-enhancement.md.

  2. Crawl FIRST — sitemap.xml → robots.txt → HTML BFS → Wayback fallback → CMS index endpoints (Squarespace /config/pages, WP wp-sitemap-posts-*.xml, Wix _api/v1/sitemap.xml) → _url_inventory.json. Every URL classified keep | merge | 301 | drop.

  3. Org-type page-set floor per 15-site-generation/page-set-expansion.md:

  • nonprofit: 14 standard (home, about, mission, programs, services, team, volunteer, donate, contact, blog, faq, news, financials, annual-report) + 10 jewels
  • local-business: ~12 standard + service × city pSEO
  • saas: home, product, pricing, features, customers, blog, docs, changelog, security, status, integrations, roadmap, api-docs
  • portfolio: home, about, work, blog, contact + per-project pages
  1. Demographic-i18n auto-trigger via ACS B16001 lookup on service area — every language ≥10% community share gets /{locale}/* full route mirror with hreflang cross-refs. Newark NJ ⇒ en+es+pt automatic.

  2. Jewel discovery (org-type-specific high-value pages absent from source):

  • nonprofit: /parish-toolkit|/planned-giving|/financials|/annual-report|/donate/refurbish|/ways-to-give|/transcript|/alumni|/press|/testimonials|/partners
  • saas: /changelog|/security|/status|/integrations|/customers|/roadmap|/api-docs
  • local: /specials|/gallery|/team|/reviews|/insurance|/financing
  • portfolio: /now|/uses|/colophon|/reading-list
  1. IA normalize — /home → /, /our-mission-1 | /blog-1 | /testpage | /new-page-* → drop or 301, flat /health-clinic → /services/health-clinic nested + index when 9+ services. CRUFT_PATTERNS auto-301 to canonical.

  2. Squarespace random-ID slugs → semantic + 301. Same for Wix _compiler/page-data dupes + WP ?p=123 permalinks.

  3. Jewel content authority via Form 990 / Charity Navigator / Candid / Wayback / Newspapers.com / LinkedIn — never stub, every quantitative claim APA-cited per rules/citations.md.

  4. Hard gate page count — deployed site MUST have keep_count + standard_gap_count + jewel_count + locale_count*(keep+standard+jewel) pages. Nonprofits min: keep + 14 standard + 10 jewels. Build fails when deployed-route-count < expected.

  5. Monitor-fire on first tool-call message — rebuild prompt is multi-faceted (≥7 independent work units: crawl → classify → org-type-infer → demographic-i18n → jewel-content-author → IA-normalize → Squarespace-dedupe → deploy-verify). Sequential = build fail. See rules/source-site-enhancement.md § Parallel-agent playbook.

12. Current-Platform Motion + Asset Discipline (111-118 — always on, every build)

  1. Scroll-driven animation = progressive enhancement — animation-timeline: view() / scroll() ONLY inside @supports (animation-timeline: view()). Firefox: unsupported as of late 2026 — verify at caniuse before relying. Never animation-only state.
  2. JS fallback auto-disables when CSS wins — IntersectionObserver reveal runs only where native support is absent: if (!CSS.supports('animation-timeline: view()')) observe(...). Double-driving one element = double motion = fail.
  3. @starting-style first-paint entrances — transition-based, zero JS; pair transition-behavior: allow-discrete for display/dialog/popover entry+exit. Baseline mid-2024 (Chrome 117 / Safari 17.5 / Firefox 129) — still verify at caniuse.
  4. Canvas nebula/particle hero layer — zero-dep 2D canvas: DPR capped at 2, particle density scaled to viewport area, prefers-reduced-motion → ONE rich static frame (no rAF loop). Reference impl: claude.megabyte.space hero — heroNebula fn in public/index.html (heymegabyte/claude.megabyte.space).
  5. rAF pauses on BOTH signals — document.hidden (visibilitychange) AND element-offscreen (IntersectionObserver on the canvas). Either alone still burns battery + CPU.
  6. Bento asymmetry recipe — auto-fill minmax() grid + 1-2 span-2 anchor cards + accent wash on the anchors. A uniform card wall is slop (extends rule 9's anti-centered-stack mandate).
  7. Kinetic gradient type — background-clip: text + background-size: 200% (or larger) + slow background-position pan; gate on prefers-reduced-motion: no-preference.
  8. Immutable assets MUST be content-versioned — any Cache-Control: immutable / 1yr asset ships as hashed filename or ?v=<version> bumped on EVERY change, INCLUDING any _headers Early-Hints preload of the same URL (CF 103 Early Hints cache also lags a deploy briefly). Reference incident (claude.megabyte.space, 2026-10-04): a stylesheet served immutable, max-age=1yr on an UNVERSIONED URL left returning visitors on stale CSS indefinitely.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Every delivered app — the non-inferable business requirements + exact stack/brand choices every build must satisfy. Loads on every app/site build.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

Cloudflare-first platform selection. Decision trees for Workers, D1, R2, KV, DO, Queues, Vectorize, Containers, Sandboxes, Flagship, Agent Memory, Workflows v2. Default stack, override conditions, auth, data patterns, reliability.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

Extract real brands (Wayback for rebuilds). Copy system, headline/CTA rules, trust surfaces, legal pages, SEO+structured data, anti-AI-slop, microcopy, DESIGN.md, W3C DTCG tokens, pSEO 5 types, GEO/AI search.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

Implements features in vertical slices, always starting with homepage. Enforces anti-placeholder rules — no lorem ipsum, no TODO stubs, no gray boxes. Real content, real images, real interactions. TypeScript strict mode, Zod validation, and structured file organization.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

The consolidated, world-class Cloudflare + integrations reference — CF platform/data/AI-edge patterns, media pipeline, observability/growth, payments, and the site-generation pipeline. The non-inferable choices that build the greatest web apps.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

Copy + discoverability requirements every delivered app must meet — real-brand extraction, anti-slop voice, per-page SEO/JSON-LD, pSEO, GEO/AI-search, truth/citations, competitor floor, i18n, trust surfaces.

日本語の概要は準備中です。原文の説明を表示しています。

heymegabyte/agent-skills242026年10月11日 更新

heymegabyte のスキルをすべて見る

このスキルの問題を報告する