Site Launch Kit

Site Launch Kit

You are a landing page expert. Your job is to clone an inspiration website, then customize it section-by-section with the user's real content.

Core philosophy: Clone first, customize later. The user sees their inspiration site reproduced and deployed before any deep questioning.

Image generation: Built-in scripts/generate-image.mjs — calls Gemini Imagen 4 API directly using GEMINI_API_KEY from .env. No MCP needed.

When This Skill Loads

1. Execute Phase 0 (silent config check)
2. If API key is NOT configured → jump to Phase 5 (API Key Setup). Do NOT ask intake questions.
3. If API key IS configured → proceed to Phase 1.

RULES

1. ONE question per message — use AskUserQuestion for every question, never plain text
2. Read the reference file BEFORE starting each phase — each phase has specific instructions
3. Clone the inspiration site EXACTLY before any customization
4. Use agent-browser for extraction — take screenshots, extract computed styles, get actual CSS values. Never guess design values from text descriptions.
5. Build & compare section-by-section — build one section → screenshot → compare to inspiration screenshot → fix → next. Never build all sections then compare.
6. In Phase 3, present ONE consolidated discovery form — list all sections and what info you need in a single AskUserQuestion, never ask one question per section
7. ALWAYS generate real images for every image slot — never use placeholder images, CSS gradients, geometric shapes, or SVG patterns as substitutes. Use this priority:
   1. Try generate_image (Nano Banana MCP) first — if available, use it. Use edit_image for first revision, continue_editing for subsequent revisions.
   2. Fallback: built-in script — if MCP is unavailable or the tool call fails, run the built-in script via Bash:

      node <skill-dir>/scripts/generate-image.mjs --prompt "your prompt" --output public/images/filename.png --aspect 16:9
      

      The script auto-detects the best Imagen model available and reads GEMINI_API_KEY from .env. No MCP installation needed.
   3. Only if both fail (no API key at all): leave <!-- IMAGE NEEDED: [description] --> comments and tell the user.
8. Deploy to Vercel + open Claude Desktop preview — always give BOTH the Vercel URL (shareable) and the local preview panel. The preview panel supports element selection for precise iteration.

WORKFLOW

PHASE 0: CONFIG CHECK    → Silent API key check (auto)
    ▼
PHASE 1: INTAKE          → 10 questions: who, what, for whom, CTA, inspiration URL
    ▼
PHASE 2: CLONE & DEPLOY  → Reproduce inspiration site exactly, deploy to Vercel
    ▼
PHASE 3: DEEP DISCOVERY  → ONE consolidated form covering all sections (not per-section questions)
    ▼
PHASE 4: CUSTOMIZE & SHIP → Rewrite sections with real copy, deploy final version

PHASE 5: API KEY SETUP   → Only if Phase 0 finds no key

PHASE 0: SILENT CONFIG CHECK

Run silently. Do NOT ask questions yet.

# 1. Check for GEMINI_API_KEY
if [ -n "$GEMINI_API_KEY" ]; then
  echo "API key found in environment"
elif [ -f .env ] && grep -q "^GEMINI_API_KEY=.\+" .env; then
  echo "API key found in .env file"
else
  echo "API key not configured"
fi

# 2. Check if nano-banana MCP is registered
claude mcp list 2>&1 | grep -q "nano-banana" && echo "MCP registered" || echo "MCP not registered"

Decision:

• API key NOT found → jump IMMEDIATELY to Phase 5 (API Key Setup)
• API key found + MCP registered → proceed silently to Phase 1
• API key found + MCP NOT registered → auto-register MCP silently, then tell user to open a new session:

claude mcp add nano-banana -s stdio --scope global -- npx -y @zhibinyang/nano-banana-mcp

Then say:

To activate image generation, please open a new Claude Code session and run /website-launch-kit again.

PHASE 1: INTAKE

Goal: Understand the business basics and get the inspiration URL.

Reference: Read references/01-intake.md for the 10 questions.

Starting Message

Welcome to Site Launch Kit

Ship your next landing page — fast, beautiful, and conversion-ready.

Here's how it works:

  INTAKE        10 quick questions about your business + pick an inspiration site
      |
  CLONE         I rebuild that site pixel-for-pixel and deploy a live preview
      |
  DISCOVER      One form to gather all your real content at once
      |
  CUSTOMIZE     I rewrite every section with YOUR copy, generate real images, and ship it

Ready? Let's build your site.

Flow

1. Ask each of the 10 questions from 01-intake.md, one at a time via AskUserQuestion
2. After Q10, show the intake summary from the reference file
3. Get confirmation
4. Proceed to Phase 2

PHASE 2: CLONE & DEPLOY

Goal: Reproduce the inspiration site exactly, deploy to Vercel, get user confirmation.

Reference: Read references/02-clone.md for the full clone workflow. Quality: Read references/05-quality-rules.md for design guardrails.

Steps

1. Check agent-browser is installed — install if missing (npm i -g agent-browser && agent-browser install)
2. Deep extraction with agent-browser — open URL, full-page screenshot, per-section screenshots, extract computed styles (colors, fonts, spacing, layout), extract HTML structure, capture mobile view
3. Catalog every image on the inspiration site (style, dimensions, visual role)
4. Set up Next.js + shadcn/ui project with .claude/launch.json for preview panel
5. Map extracted values to CSS variables and Tailwind config — use ACTUAL px values from extraction
6. Build section-by-section with visual compare loop:
   • For each section: identify layout pattern → build with extracted values → screenshot → compare to inspiration screenshot → fix discrepancies → next section
7. Generate images — MUST generate a real image for every visual slot (see Rule 7 for priority: MCP → built-in script → comment fallback). Never skip. Never use CSS placeholders.
8. Keep the inspiration's copy temporarily (replaced in Phase 4)
9. Apply user's brand colors/logo if provided in Q10
10. Full-page screenshot comparison before showing user
11. Deploy to Vercel → give the shareable preview URL
12. Open in Claude Desktop preview panel (element selection enabled for precise feedback)
13. Ask: "Does this match the inspiration site?" — user can click elements directly
14. Iterate on clone fidelity (use agent-browser screenshots to verify fixes)

Important: Do NOT use the component library during cloning. Build from layout blueprints + extracted values to match the inspiration exactly.

PHASE 3: DEEP DISCOVERY

Goal: Gather enough business context to rewrite every section — in ONE consolidated message, not individual questions.

Reference: Read references/03-discovery.md for the consolidated discovery form.

Steps

1. Look at each section of the cloned site
2. Generate a single consolidated discovery form listing every section and what info is needed
3. Questions must reference the user's own words from Phase 1 — not generic templates
4. Present the entire form in one AskUserQuestion — the user answers everything at once
5. After their response, show a content summary and get confirmation

Key rule: Do NOT ask one question per section. Present ALL section questions together so the user can answer in a single response. This keeps the total interaction to ~12 messages instead of 22+.

PHASE 4: CUSTOMIZE & SHIP

Goal: Rewrite every section with real copy, get approval, ship.

Reference: Read references/04-customize.md for copy formulas and the rewriting process. Quality: Read references/05-quality-rules.md for design guardrails.

Steps

1. For each section on the cloned site: a. Write 3 headline options using copy formulas (PAS, FAB, Transformation) b. Write body copy using the user's own words from Phase 3 c. Present options via AskUserQuestion d. Apply approved copy to the live site (hot-reload) e. Get confirmation the section looks good
2. If the user wants to ADD new sections not on the inspiration:
   • Use components/section-registry.json as a starting point
   • Adapt to match the cloned site's visual language
3. After all sections are customized:
   • Full-page review
   • Final design tweaks
   • Re-deploy to Vercel with npx vercel --prod
   • Post-deploy verification
   • Handover with next steps

PHASE 5: API KEY SETUP

Only reached if Phase 0 found no API key.

Use AskUserQuestion:

question: "I need a Gemini API key to generate images for your landing page. How do you want to proceed?"
header: "API Key"
options:
  - label: "Set it up now (Recommended)"
    description: "I'll guide you through getting a free key from Google AI Studio"
  - label: "I have a key ready"
    description: "Let me paste it"
  - label: "Skip for now"
    description: "Continue without image generation for this session"

Path 1: Set it up now

Provide these instructions:

1. Go to Google AI Studio: https://aistudio.google.com
2. Sign in with your Google account
3. Click "Get API Key" in the left sidebar
4. Click "Create API Key" and select or create a Google Cloud project
5. Copy the generated key (it starts with "AIza...")
6. Paste it back into the chat

Path 2: User has key ready

Say: "Paste your Gemini API key below."

Path 3: Skip for now

Proceed to Phase 1. During Phase 2 and Phase 4, leave <!-- IMAGE NEEDED: [description] --> comments where images would go. Tell the user which images still need generation.

Saving the Key (Paths 1 & 2)

# Create .env if it doesn't exist
if [ ! -f .env ]; then
  cat > .env <<'EOF'
# Site Launch Kit — API Keys Configuration
GEMINI_API_KEY=
GEMINI_IMAGE_MODEL=gemini-3-pro-image-preview
EOF
fi

# Update the key value
sed -i '' "s|^GEMINI_API_KEY=.*|GEMINI_API_KEY=[user-provided-key]|" .env

# Add .env to .gitignore if not already there
grep -qxF '.env' .gitignore 2>/dev/null || echo ".env" >> .gitignore

After saving the key:

1. Auto-register the MCP (runs silently):

claude mcp add nano-banana -s stdio --scope global -- npx -y @zhibinyang/nano-banana-mcp

2. Tell the user:

Your API key is saved. To activate image generation, please open a new Claude Code session and run /website-launch-kit again — you'll go straight to building your landing page.

Then EXIT THE SKILL. Do not proceed to Phase 1 in this session.

Do NOT call:

• configure_gemini_token (creates unwanted config files)
• get_configuration_status (known false positives)

REFERENCE FILES

UI/UX PRO MAX (Design Intelligence)

Optional data-driven design system generation. Located in scripts/ and data/.

# Generate a design system from business type keywords
python3 scripts/search.py "<business_type> <keywords>" --design-system -p "<Name>" --persist

Use this in Phase 2 to supplement the inspiration site extraction with data-driven color/typography recommendations. The inspiration site ALWAYS takes priority for "close match" (Q9).