Content Repurposing Engine

Transform any content into platform-optimized outputs across 17+ channels.

Quick Reference

Step 1: Input Detection

Detect the input type and extract raw content accordingly.

Detection patterns (check in order):

Extraction rules:

• For YouTube: extract transcript, video title, channel name, description, and duration
• For blog URLs: extract title, author, publication date, body text, and any embedded media descriptions
• For audio: transcribe to text, note duration and any speaker labels
• For local files: read full contents; for PDF use the Read tool with page ranges if large
• For pasted text: use verbatim; ask the user for a title/topic if not obvious

If extraction fails, report the error clearly and ask the user to paste the content directly.

Step 2: Content Atomization

Break the extracted content into 5-15 reusable "atoms." Each atom is a self-contained content unit.

Atom types:

Atomization rules:

1. Extract between 5 and 15 atoms (aim for 8-12 for most content)
2. Always include at least one tldr, one insight, and one quote or contrarian
3. Label each atom with its type
4. Rate each atom's standalone impact from 1-5 (5 = works perfectly as a standalone post)
5. Identify the main argument, target audience, and primary topic
6. Preserve the original author's voice in quoted atoms
7. For longer content (>2000 words), aim for 12-15 atoms; for shorter, 5-8

Atomization output format:

## Content Atoms

**Main Argument:** [one sentence]
**Target Audience:** [who benefits from this content]
**Primary Topic:** [category/niche]

| # | Type | Atom | Impact |
|---|------|------|--------|
| 1 | tldr | ... | 5 |
| 2 | insight | ... | 4 |
| ... | ... | ... | ... |

Step 3: Voice Detection and Adaptation

Default behavior: Detect the brand voice from the source content. Analyze formality, humor level, directness, emotion, and jargon to infer the author's natural voice.

Override flags:

• --voice casual — friendlier, more emoji, conversational, contractions
• --voice professional — formal, data-driven, authoritative, measured
• --voice witty — personality-forward, clever wordplay, punchy, unexpected angles

Core rule: The brand voice stays consistent across all platforms. Only the TONE adapts per platform. Load references/voice-adaptation.md for the full platform-tone matrix.

Step 4: Orchestration — Spawn 6 Parallel Agents

After atomization, spawn 6 subagents in parallel. Each agent receives:

• The full list of content atoms (with types and impact ratings)
• The main argument, target audience, and primary topic
• The detected or overridden brand voice
• The --brief flag status (if true, produce 1 post per platform instead of full sets)
• The --images flag status

Agent 1: repurpose-social

Platforms: Twitter/X, Threads, LinkedIn, Facebook Receives: All atoms, voice profile, platform specs Produces:

• Twitter: 1 standalone tweet + 1 thread (8-12 tweets) + 1 poll + image prompt
• Threads: 1 thread (5-10 posts) + 3-5 standalone posts + 1 image post concept
• LinkedIn: 1 text post + 1 carousel outline (8-10 slides) + 1 poll
• Facebook: 1 text post + 1 question post + 1 poll Sub-skills invoked: repurpose-twitter, repurpose-threads, repurpose-linkedin, repurpose-facebook

Agent 2: repurpose-visual

Platforms: Instagram, TikTok, Pinterest, Snapchat, Quote cards, Image prompts Receives: All atoms (especially quotes, stats, insights), voice profile Produces:

• Instagram: 1 carousel (7-10 slides with copy) + 1 reel script (30-60s) + caption + hashtags
• TikTok: 1 video script (15-60s) + 1 carousel (2-10 slides) + 1 stitch/duet concept
• Pinterest: 3-5 pin descriptions + 1 idea pin script (5-10 slides) + board suggestions
• Snapchat: 1 story script (3-5 frames) + 1 Spotlight script (60s) + AR lens concept
• Quote cards: 3-5 text overlays for image generation
• Image prompts: hero image + carousel cover + 3 quote card prompts Sub-skills invoked: repurpose-instagram, repurpose-tiktok, repurpose-pinterest, repurpose-snapchat, repurpose-quotes

Agent 3: repurpose-longform

Platforms: Newsletter, Email sequence, Reddit, Quora Receives: All atoms, full original content reference, voice profile Produces:

• Newsletter: subject line + preview text + 200-500 word body
• Email sequence: 3-email drip (day 0, day 2, day 4) with subject lines and CTAs
• Reddit: title + post body adapted to subreddit norms
• Quora: 1 answer (300-1000 words) + 1 Space post + question suggestions Sub-skills invoked: repurpose-newsletter, repurpose-reddit, repurpose-quora

Agent 4: repurpose-community

Platforms: YouTube Community, Skool, Discord Receives: All atoms (especially questions, polls, insights), voice profile Produces:

• YouTube Community: 1 text post + 1 poll (5 options) + 1 image post concept
• Skool: 1 discussion post + 1 challenge/action post + 1 poll
• Discord: 1 announcement post + 1 discussion thread prompt + 1 embed message Sub-skills invoked: repurpose-youtube, repurpose-skool, repurpose-discord

Agent 5: repurpose-seo

Platforms: Cross-platform SEO metadata Receives: All atoms, main argument, target audience, topic Produces:

• Primary keywords (3-5) and secondary keywords (5-10)
• Hashtag sets per platform
• SEO title and meta description (for blog/newsletter)
• Alt text suggestions for all generated images
• Schema markup suggestions (Article, VideoObject, FAQPage) Sub-skills invoked: repurpose-seo

Agent 6: repurpose-broadcast

Platforms: WhatsApp Channels, Telegram Channels, Medium Receives: All atoms, full original content reference, voice profile Produces:

• WhatsApp: 1 channel update (100-300 chars) + 1 poll + 1 content teaser
• Telegram: 1 channel post (500-1000 chars) + 1 deep dive (1000-2000 chars) + 1 poll
• Medium: 1 article (1500-3000 words) + title/subtitle + 5 tags + publication suggestions Sub-skills invoked: repurpose-whatsapp, repurpose-telegram, repurpose-medium

Platform filtering

If --platforms is specified, only spawn the agents that cover the requested platforms:

• twitter, threads, linkedin, facebook → Agent 1
• instagram, tiktok, pinterest, snapchat, quotes, images → Agent 2
• newsletter, email, reddit, quora → Agent 3
• youtube, skool, discord → Agent 4
• seo → Agent 5
• whatsapp, telegram, medium → Agent 6

If a single platform is requested, spawn only the relevant agent.

Step 5: Collect Outputs and Generate Summary

After all agents complete:

1. Compile all outputs into the output directory structure (see below)
2. Generate images — ALWAYS attempt image generation when /banana is available (do NOT wait for --images flag). If /banana is unavailable, save prompts to quotes/banana-prompts.md for manual use later
3. Generate a summary table showing what was produced per platform
4. Generate a 7-day publishing calendar (unless --brief was used)
5. Generate HTML viewer — Run python3 scripts/generate_html.py <output-dir> to create index.html (dark-themed viewer with Copy buttons per content piece) and all-content.md (single consolidated file with all platform outputs)
6. Report to user — Show summary and link to index.html for easy browsing and copying

Output Directory Structure

./repurposed/<YYYY-MM-DD_HHMMSS>/
  summary.md              # Overview of all outputs + publishing calendar
  atoms.md                # Content atomization results
  twitter/
    standalone-tweet.md
    thread.md
    poll.md
  linkedin/
    post.md
    carousel.md
    poll.md
  instagram/
    carousel.md
    reel-script.md
    caption.md
  threads/
    thread.md
    standalone-posts.md
    image-post.md
  facebook/
    post.md
    question.md
    poll.md
  pinterest/
    pins.md
    idea-pin.md
    boards.md
  snapchat/
    story-script.md
    spotlight-script.md
    ar-concept.md
  youtube-community/
    text-post.md
    poll.md
    image-post.md
  skool/
    discussion.md
    challenge.md
    poll.md
  discord/
    announcement.md
    thread-prompt.md
    embed.md
  tiktok/
    video-script.md
    carousel.md
    stitch-duet.md
  reddit/
    post.md
  quora/
    answer.md
    space-post.md
    questions.md
  newsletter/
    newsletter.md
    email-sequence.md
  medium/
    article.md
    tags-publications.md
    crosspost-note.md
  whatsapp/
    update.md
    poll.md
    teaser.md
  telegram/
    post.md
    deep-dive.md
    poll.md
  seo/
    keywords.md
    hashtags.md
    metadata.md
  quotes/
    quotes.md             # 5 quotable moments
    banana-prompts.md     # /banana prompts (always generated)
  images/                 # Generated images (auto when /banana available)
    quote-card-*.png
    carousel-cover.*
    hero.*
  seo-metadata.md         # Cross-platform SEO metadata
  all-content.md          # MANDATORY: Single consolidated markdown (all platforms)
  index.html              # MANDATORY: HTML viewer with Copy buttons per content piece
  calendar.md             # 7-day publishing calendar

MANDATORY OUTPUT: Every /repurpose run MUST produce all-content.md (single file with everything) and index.html (viewer with Copy buttons). Run python3 scripts/generate_html.py <output-dir> as the final step.

Reference Files

Load these references as needed during the pipeline:

Sub-Skills

Image Sourcing Pipeline — 3-Tier System

Images are sourced automatically for every platform. The pipeline follows a 3-tier priority chain.

Tier 1: Website Images (if input is a URL)

When the input is a blog post or article URL, extract_article.py extracts images from the page.

• Use scripts/fetch_images.py --article-images to filter: skip logos, icons, avatars, ads, SVGs, < 400px
• Rank by keyword overlap with content atoms
• These are the MOST relevant — they came from the source content itself

Tier 2: Stock Photos (Pixabay → Unsplash → Pexels)

Search for topic-relevant images via WebSearch:

• site:pixabay.com [topic keywords] wide professional (preferred, no attribution required)
• site:unsplash.com [topic keywords] professional (resize params supported)
• site:pexels.com [topic keywords] high quality (fallback)
• Extract direct CDN URLs from results
• Verify each URL with HEAD request before using
• Target: 3-5 relevant images

Tier 3: AI-Generated (Gemini via /banana)

• ALWAYS for quote cards (text overlay requires custom design)
• ALWAYS for carousel covers with title text
• FALLBACK when Tiers 1-2 produce < 3 suitable images
• Uses 6-Component Brief: Subject → Action → Context → Composition → Lighting → Style
• Load references/image-sourcing.md for templates and platform dimensions

Execution Order

1. Check if input URL has images → extract and filter (Tier 1)
2. Generate stock photo queries → use WebSearch to find Pixabay/Unsplash/Pexels images (Tier 2)
3. Count total images found. If < 3 usable, generate more via /banana (Tier 3)
4. ALWAYS generate 5 quote cards via /banana (or save prompts if unavailable)
5. Save all images to ./repurposed/<timestamp>/images/
6. Include image URLs/paths in platform output files

/banana Detection

Check for gemini_generate_image MCP tool OR ~/.claude/skills/banana/SKILL.md OR GOOGLE_API_KEY env var.

• Available: generate images automatically
• Not available: save all prompts to quotes/banana-prompts.md for manual generation later

Error Handling

Subcommand: analyze

Usage: /repurpose analyze <url-or-file>

Runs only Steps 1-2 (input detection + atomization). Does NOT generate platform outputs.

Output:

• Content atoms table with types and impact ratings
• Main argument, target audience, primary topic
• Recommended platforms based on content type
• Suggested voice profile
• Atom count and quality assessment

Use this to preview what the engine extracts before committing to full repurposing.

Subcommand: calendar

Usage: /repurpose calendar <url-or-file>

Runs Steps 1-2 (extraction + atomization), then generates a 7-day publishing calendar.

Calendar rules:

• Monday: Long-form (newsletter or Reddit post)
• Tuesday: Thread (Twitter) + Carousel (LinkedIn)
• Wednesday: Quote cards + Instagram carousel
• Thursday: Polls across all platforms
• Friday: Newsletter send or email drip start
• Saturday: YouTube Community + Skool posts
• Sunday: Stories + light engagement posts (Facebook question, Skool discussion)

Calendar output includes:

• Day-by-day schedule with platform, content type, and which atom(s) to use
• Suggested posting times per platform (from platform-specs.md)
• Dependencies (e.g., "Tuesday carousel requires Monday's images")
• Total pieces count

Output: Saved to ./repurposed/<timestamp>/calendar.md and printed to console.

Execution Flow Summary

User input
    |
    v
[1] Detect input type (URL/file/text)
    |
    v
[2] Extract raw content (script or Read)
    |
    v
[3] Atomize into 5-15 labeled atoms
    |
    v
[4] Detect or apply voice profile
    |
    v
[5] Spawn 6 agents in parallel:
    |-- repurpose-social (Twitter + Threads + LinkedIn + Facebook)
    |-- repurpose-visual (Instagram + TikTok + Pinterest + Snapchat + Quotes)
    |-- repurpose-longform (Newsletter + Email + Reddit + Quora)
    |-- repurpose-community (YouTube Community + Skool + Discord)
    |-- repurpose-seo (Keywords + Hashtags + Metadata)
    |-- repurpose-broadcast (WhatsApp + Telegram + Medium)
    |
    v
[6] Collect all outputs → write to ./repurposed/<timestamp>/
    |
    v
[7] Generate summary table + 7-day calendar
    |
    v
[8] Report results to user