Swipe Lab

Swipe Lab — Carousel Content Factory

Your personal carousel factory. Scrape creators for inspiration → generate carousels in YOUR brand and voice.

First Run — Onboarding

If no brand file exists at ~/.claude/skills/swipe-lab/brands/, this is a new user. Run onboarding:

Step 1: Get their brand

Ask: "What's your Instagram handle?"

Then scrape their profile to extract:

• Bio, name, handle
• Visual style (photo vs graphic ratio, color patterns, vibe)
• Content themes from recent posts
• Voice/tone from captions

Ask them to confirm or adjust:

• "Here's what I found about your brand. Does this look right?"
• Show: colors, voice description, content pillars, audience
• Let them tweak anything

Step 2: Pick creators to study

Ask: "Drop 1-3 Instagram handles of creators whose carousels you love."

Scrape each creator's carousels:

• Screenshot every slide
• Extract topics, hooks, structures, content patterns
• Store in ~/.claude/skills/swipe-lab/knowledge/{handle}/

Step 3: Generate brand config

Create ~/.claude/skills/swipe-lab/brands/{handle}.json using this structure:

{
  "handle": "@theirhandle",
  "name": "Their Name",
  "businesses": ["Business 1", "Business 2"],
  "tagline": "Extracted or provided tagline",
  "niche": "Their niche in one sentence",
  "audience": "Who they serve",

  "brand_colors": {
    "primary_accent": "#hexcolor",
    "background_dark": "#0a0a0a",
    "background_light": "#ffffff",
    "text_dark": "#111111",
    "text_light": "#ffffff",
    "text_muted_dark": "#888888",
    "text_muted_light": "#666666",
    "bottom_bar": "#hexcolor"
  },

  "typography": {
    "font_family": "Inter",
    "heading_weight": "900",
    "body_weight": "400",
    "heading_size": "56px",
    "body_size": "28px"
  },

  "visual_style": {
    "aesthetic": "Description of their visual vibe",
    "layout": "How they lay out content",
    "bottom_bar": "Bottom bar style description"
  },

  "voice": {
    "tone": "Description of how they write",
    "reading_level": "Grade level",
    "phrases_to_use": ["phrase 1", "phrase 2"],
    "phrases_to_avoid": ["phrase 1", "phrase 2"],
    "content_pillars": ["pillar 1", "pillar 2", "pillar 3"]
  },

  "cta_patterns": {
    "dm_keywords": ["KEYWORD1", "KEYWORD2"],
    "format": "DM me \"{keyword}\" to get started.",
    "closing_line": "@handle | Business Name"
  },

  "carousel_defaults": {
    "ratio": "4:5 (1080x1350)",
    "accent_color": "#hexcolor",
    "slide_alternation": "How slides alternate (dark/white, all dark, etc)",
    "bottom_bar_height": "55px"
  },

  "inspiration_creators": ["@creator1", "@creator2"]
}

Step 4: Confirm and go

"You're set up. From now on, just say 'Swipe Lab' and tell me what you want."

Returning User — Normal Flow

If a brand file exists at ~/.claude/skills/swipe-lab/brands/, load it and go.

If MULTIPLE brand files exist, ask: "Which brand? [list handles]" — or auto-detect from the current working directory / conversation context.

Quick Start Examples

• "Make me a carousel about [topic]" → generates in their brand
• "Swipe Lab — scrape @someone and make carousels from their topics" → scrapes + generates
• "Give me 5 carousels I can post this week" → picks topics from knowledge base + generates
• "[Theme name] carousel about [topic]" → generates in specific theme
• "Update my brand" → re-scrape their profile, update brand.json
• "Add @newcreator to my inspiration" → scrape and add to knowledge base

The 7 Themes

Theme selection: Auto-pick based on content type unless the user specifies. Override the theme's default accent color with the user's brand accent color.

How It Works

ONBOARDING (first time):
  User's IG handle → scrape profile → extract brand
  Creator handles → scrape carousels → extract knowledge
  → brand.json + knowledge base saved

SCRAPE MODE (anytime):
  IG handle → scrape_carousels.py → screenshot every slide
           → Claude reads screenshots → extracts knowledge + topics
           → stored in knowledge/{handle}/

GENERATE MODE (the main thing):
  Topic + Theme → load user's brand.json
                → pull inspiration from knowledge base (optional)
                → write slides in user's voice
                → generate HTML with user's brand colors/fonts
                → Playwright screenshots at 1080x1350
                → ready-to-post PNGs in output folder

File Locations

All files live inside ~/.claude/skills/swipe-lab/:

Voice Rules (MANDATORY)

All carousel copy MUST sound like the user — not generic motivational content.

1. Read their brand.json — use their tone, phrases, reading level
2. Match their content pillars — don't go off-topic
3. Use their phrases — the ones listed in phrases_to_use
4. Avoid their anti-phrases — the ones in phrases_to_avoid
5. Reality check: Would this person actually post this? If it sounds like a template, rewrite it.

Slide Design Rules

1. 4:5 ratio (1080x1350) — takes up more feed space
2. No slide count limits — 3 slides or 15 slides. Content decides the length. Never pad, never cut short.
3. Bottom bar on EVERY slide — swipe teaser + slide counter + @handle
4. Slide 1 stops the scroll — big number, provocative question, or bold visual
5. Use brand colors — accent from brand.json, not theme defaults
6. No empty space — every pixel has a job
7. Swipe teasers — "it gets worse →", "here's the fix →", "my take →"

Managing Multiple Brands

Swipe Lab supports multiple brand profiles. Each is a separate JSON in brands/.

• brands/kclavier.json — Kevin's brand
• brands/itisdrey.json — the user's brand
• brands/someclient.json — A client's brand

When generating, Claude checks:

1. If user specifies a brand: use that one
2. If only one brand exists: use it automatically
3. If multiple exist: ask which one

To add a new brand: "Swipe Lab — set up a new brand for @handle" To switch brands: "Swipe Lab — use @handle's brand" To remove a brand: "Swipe Lab — remove @handle"

Playwright Reference (IMPORTANT — read before scraping or rendering)

Scrape Instagram (with logged-in Chrome profile)

from playwright.async_api import async_playwright
import asyncio, os

async def scrape():
    async with async_playwright() as p:
        browser = await p.chromium.launch_persistent_context(
            user_data_dir=os.path.expanduser("~/Library/Caches/ms-playwright/mcp-chrome-profile"),
            headless=True,
            args=["--disable-blink-features=AutomationControlled"],
            user_agent="Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"
        )
        page = await browser.new_page()
        await page.goto("https://www.instagram.com/someone/", wait_until="domcontentloaded", timeout=30000)
        await page.wait_for_timeout(4000)
        await page.screenshot(path="/tmp/screenshot.png")
        await browser.close()

asyncio.run(scrape())

CRITICAL: Always use wait_until="domcontentloaded" — NEVER networkidle. Instagram never finishes loading network requests, so networkidle will timeout every time.

Render HTML slides to PNGs

from playwright.async_api import async_playwright
import asyncio

async def render():
    async with async_playwright() as p:
        browser = await p.chromium.launch(headless=True)
        page = await browser.new_page(viewport={"width": 1080, "height": 1350})
        await page.goto("file:///path/to/slides.html", wait_until="domcontentloaded")
        await page.wait_for_timeout(2000)
        slides = await page.query_selector_all(".slide")
        for i, slide in enumerate(slides, 1):
            await slide.screenshot(path=f"/tmp/output/slide_{i}.png")
        await browser.close()

asyncio.run(render())

Chrome profile location

~/Library/Caches/ms-playwright/mcp-chrome-profile

If Instagram scraping fails

Session expired. Tell the user to re-login:

npx playwright open --user-data-dir ~/Library/Caches/ms-playwright/mcp-chrome-profile https://instagram.com

Log in, close the browser. Then scraping works again.

Changelog

• [2026-03-17]: Generic Swipe Lab created. Multi-brand support, onboarding flow, 7 themes. Forked from Kevin Clavier's custom build.