Content Repurpose System
Turn one finished video into blog and social drafts, clip plans, vertical reels with hooks you can test, and carousels. Then get one folder that's ready to post, with a plan for where and when.
This Claude Code plugin has ten skills. Onboard once to set your channel and voice. Then use the skills to plan each video and reuse the finished edit. Title ideas use your channel performance data.
The running example in this README uses the channel @SaminYasar_ — Claude Code tutorials for stock-trading automation. Swap in your own handle during onboarding and everything below works the same.
The mental model
┌─────────────┐
│ /onboard │ ← run once: channel + Voice DNA + keys
└──────┬──────┘
│
BEFORE you film ────────┼──────── AFTER you publish
│
/ideation → /yt-titles │ /content-cascade → blog + Twitter + LinkedIn (+guide)
↓ ↓ │ ↓
/hooks → /outlines │ /short-form → 3–5 clip plans
│ │ ↓
│ │ /reels → plan → film hooks → render 9:16 variants
│ │ ↓
│ │ /carousels → tweet-style slides + LinkedIn PDF
│ │ ↓
└──── film the video ────┘ /publish-pack → finished/ folder + PUBLISH PLAN.md
You publish one video. The system turns it into a week of content across six channels. Every run ends in one finished/ folder: numbered sections, captions per platform, and a plan that says what to post where, day by day.
Install
Option A — as a plugin (recommended)
In Claude Code:
/plugin marketplace add AI-Answer/ContentRepurposeSystem
/plugin install content-repurpose-system
Restart Claude Code so the skills load, then run /onboard.
Option B — clone and install the skills directly
git clone https://github.com/AI-Answer/ContentRepurposeSystem.git
cd ContentRepurposeSystem
cp -R skills/* ~/.claude/skills/ # note: no trailing slash on skills/*
Restart Claude Code, then run /onboard.
Prerequisites
| Tool | Why | Install |
|---|
| yt-dlp | reads your upload history + "latest video" | pip install yt-dlp |
| Node.js | transcript fetching + thread posting | nodejs.org |
| Python 3 | channel performance and reel scripts | install Python 3 if missing |
| ffmpeg + ffprobe | reel crops, audio, rendering, and checks | brew install ffmpeg whisper-cpp |
| whisper-cli | local word timestamps for reel cuts and captions | included in brew install ffmpeg whisper-cpp |
| NumPy + Pillow | raw-camera alignment, reel images, and carousel slides | python3 -m pip install numpy Pillow |
youtube-transcript (optional) | free transcript fallback | npm install youtube-transcript |
The /onboard skill checks the base channel tools. /reels also checks the reel tools above. /carousels and /publish-pack need only Python 3 and Pillow. Whisper needs a local model file; point whisper-cli -m at it. For Python packages, use a virtual environment if your Python requires one.
Step-by-step walkthrough
Here's exactly what a session looks like, from zero to a full content drop, using @SaminYasar_ as the example.
Step 1 — /onboard (once)
You: /onboard
Claude walks you through three things:
1. Connect your channel.
Claude: What's your YouTube channel handle or URL?
You: @SaminYasar_
Claude: Connected to @SaminYasar_. I can now read your upload history
and pull your latest video automatically.
This writes your channel into the performance script and resolves your canonical channel ID, so later skills can find your real videos.
2. Capture your Voice DNA. Claude asks ~3 questions and folds the answers into every content prompt:
Claude: In one or two sentences, how would you describe your voice?
You: Warm and confident, analogy-first (poker, thermostats, car
insurance), I always back claims with real numbers, and I close
with a soft "come build with us in the Claude Club."
Claude: Who's your audience and what do you teach them?
You: Non-coders learning to build and automate with Claude — right now,
stock-trading automation.
From now on, your blog/Twitter/LinkedIn output sounds like you, not like a template.
3. Optional keys. Apify (reliable transcripts), Supabase (auto-save drafts), Twitter (auto-post). All skippable — say "skip" and add them later.
4. Carousel profile (optional). The name, handle, badge, and headshot path that show on your carousel slides. It's saved in skills/carousels/profile.json, which git ignores. Only the path to your photo is stored.
At the end, onboarding prints your workflow map and a suggested first command. You only do this once per channel.
Step 2 — /ideation — what should I make?
You: /ideation Claude Code for stock trading
You get 7–10 positioned ideas, each mapped to a core desire, scored for "shock," and checked against competitive gaps. The Tier 1 picks look like:
## I Let Claude Code Copy Every Trade a Senator Makes
- Angle: Turn public Capitol Trades disclosures into an automated
paper-trading bot — nobody has shown the full build.
- Core Desire: Money → Proxy: "trade like someone with inside access,
legally, on autopilot"
- Shock Score: 86/100
- Format: Case Study (build + result reveal)
- Competitive Gap: Plenty of "politicians beat the market" explainers,
zero end-to-end "here's the bot" tutorials.
- Why Now: Capitol Trades API is free + congressional trading is
trending after the latest disclosure cycle.
Pick one and carry it into the next steps.
Step 3 — /yt-titles — title it like your hits
You: /yt-titles a video where I build a bot that copies senator stock trades
This is the skill that makes the system yours: Step 1 pulls your actual top performers (via channel_performance.py), finds the title patterns your audience already rewards, then generates titles in that pattern. Output:
## Your Top Performers (last 3 months)
Your "I Let Claude…" + concrete-outcome format outperforms your how-to titles
~3x on views/day. Numbers and a clear result are doing the work.
## Title Options
### Tier 1 — High Confidence
1. I Built a Bot That Copies Senator Stock Trades (Claude Code)
Based on your "I Built…" pattern — proven, names the tool, concrete.
2. Claude Code vs. Congress: I Copied Their Trades for 30 Days
Your contrarian + timeframe pattern; "vs." adds tension.
### Tier 3 — Swing for the Fences
8. I Gave Claude My Brokerage Account. Here's What Happened.
Pattern-break curiosity gap — higher risk, higher ceiling.
## Thumbnail Text Options
For "I Built a Bot That Copies Senator Stock Trades":
1. "LEGAL?" — tension against the title's premise
2. "+18% / MO" — concrete proof the title withholds
Step 4 — /hooks — nail the first 15 seconds
You: /hooks I Built a Bot That Copies Senator Stock Trades with Claude Code
Uses the Kallaway desire-based framework to produce three fully-aligned hooks (visual + spoken + on-screen text), each scored against the Four Commandments:
### Hook 1 — Case Study + "About Me"
Spoken: "I gave Claude Code a list of every stock a senator bought
last quarter — and let it trade them for me."
Visual: Screen recording: Capitol Trades feed → Claude Code running →
Alpaca paper account ticking up
Text overlay: "COPYING CONGRESS"
Power Words: Subject ✓ Action ✓ Objective ✓ Contrast ✓ Time ✓ (5/6)
Commandments: Alignment ✓ Speed ✓ Clarity ✓ Curiosity ✓ (4/4)
Step 5 — /outlines — a recording-ready script
You: /outlines the senator-copy-trading bot video, ~14 minutes
You get a full outline in recording format: hook, structured body sections with talking points, visual-aid callouts, demo beats, source citations, per-section time budgets, and a production-notes block (competitive landscape, title ideas, thumbnail moments, honest limitations). You sit down and film straight from it.
## Hook (~60s) [the aligned hook from Step 4]
## Section 1 — The Setup (~2.5 min)
Core idea: Congressional trades are public — and weirdly good.
Talking points: • Capitol Trades is free • the disclosure lag
Visual aid: → screen recording of the Capitol Trades feed
Source: capitoltrades.com
## Section 2 — Wiring Claude Code to Alpaca (~5 min) [demo-heavy]
...
## Production Notes → 3 competitor titles, 3 thumbnail moments, limitations
→ Now you film and upload the video. YouTube takes 30–60 min to generate captions; once they're live, continue.
Step 6 — /content-cascade — the long-tail content
You: /content-cascade latest
Claude auto-detects your latest upload, fetches the transcript, classifies the video (tutorial vs. discussion), and generates — in your voice:
- A 1,500–2,500 word blog post with SEO metadata (title, description, slug, keywords, tags, cover image)
- A Twitter/X thread built around the single best angle
- A LinkedIn post mirroring the video's premise
- Because this one is a tutorial, it also generates a LinkedIn lead-magnet post + a Skool-ready HTML guide
Each piece is shown to you for review before saving. With keys configured it can auto-save drafts to Supabase and auto-post the thread; without them, everything saves as local markdown under the video folder's content/, next to the reels and carousels.
Step 7 — /short-form — slice the clips
You: /short-form latest
Scans the transcript for the 3–5 most clip-worthy moments (self-contained insight, shock value, emotional peak, demo reveal) and for each one gives you timestamps, a scroll-stopping hook, a ready-to-paste caption, and hashtags — ranked by best for reach / conversions / engagement.
### Clip 1: "Is this legal?" — Best for reach
Timestamps: 02:14 → 03:01 (~47s)
Hook: Text "IS THIS LEGAL?" | Spoken "Yes, copying a senator's
stock trades is completely legal — here's the catch."
Caption: Congress has to disclose their trades. So I had Claude trade
them for me 👀 Full build on the channel.
Hashtags: #claudecode #ai #stocktrading #investing #fintech
Step 8: /reels: plan, film, render
You: /reels /path/to/finished-edit.mp4
Use this after /short-form, or go straight to /reels when you want rendered files. It takes a finished local video or a YouTube URL. Add the raw camera takes and their matching mic files for the bottom camera panel.
- Plan. Claude picks 3-6 ideas from the word-timed transcript. It writes
work/reels.json, repurpose-plan.md, and film-hooks.md. Each reel gets one footage hook and three film hooks scored with the Kallaway framework.
- Film. Record the listed lines. Each is at most 15 words and 4 seconds. Save them as
hooks-filmed/R1-H2.mov, and so on. Say "attach my filmed hooks" to resume.
- Render. The renderer's
--all option makes every reel with each available hook. Each variant keeps the same body. Check the contact sheets, captions against audio, sync, and loudness before posting.
The default is 1080x1920: visuals on top, your raw camera on the bottom, and captions on the seam. Use --panel-crop x,y,w,h and --cam-center x for a different source layout. The reels skill has the commands and plan schema.
Step 9: /carousels: slides people save
You: /carousels the tips video
Claude picks 1-5 ideas that work without the video: a list, prompts to copy, a framework. It writes 6-10 slides per idea in your voice. The cover gets a frame from the video. Other slides get real proof: docs pages, posts, your own screen. Then it renders:
carousels/2 - Steal my 2 prompts/
slides/01.png ... 08.png 1080x1350, light or dark
Steal my 2 prompts - LinkedIn.pdf one page per slide
captions.txt Instagram, LinkedIn, TikTok (photo mode)
config.json edit and re-render any time
The renderer warns about text that runs off a slide, a lone word on a last line, and a cover with no image. Claude opens every slide and fixes what it finds. The carousels skill has the config format and an example.
Step 10: /publish-pack: one folder, ready to post
You: make the finished products folder
Claude gathers every output for the video into <video>/finished/:
finished/
README.md
PUBLISH PLAN.md
1 YouTube long-form/
2 Reels - Instagram, TikTok, YouTube Shorts/ 1 - R1 Don't read the docs.mp4 + captions
3 Hooks to film/
4 Carousels - Instagram, LinkedIn, TikTok/
5 Articles - blog, LinkedIn, Substack/
6 Text posts - X, LinkedIn, Threads, newsletter/
7 Claude Club - Opus 5.5 playbook/
Files are numbered in posting order. Each reel and carousel has a captions file with one section per platform. PUBLISH PLAN.md has four parts: what's here, a matrix of where each piece goes, a day-by-day calendar, and a checklist. A script does the copying (APFS clones on a Mac, so big videos take no extra space) and drafts the plan. Claude writes the captions, the "why that split" notes, and the final calendar.
What you have at the end
After the content commands, the reel filming loop, and /publish-pack, you have one folder of local files and drafts:
| Asset | From | Ready for |
|---|
| 📝 Blog post (SEO-optimized) | content-cascade | your website / newsletter |
| 🐦 Twitter/X thread | content-cascade | X (optionally auto-posted) |
| 💼 LinkedIn post | content-cascade | LinkedIn |
| 🧲 LinkedIn lead-magnet post | content-cascade (tutorials) | LinkedIn → email capture |
| 📚 Skool guide (HTML) | content-cascade (tutorials) | your Skool / Claude Club |
| 🎬 3–5 clip plans | short-form | Timestamps, hooks, and captions for editing |
| 🎬 3-6 reels with available hook variants | reels | Reviewed 9:16 files in out/ |
| 🖼️ 1-5 carousels + LinkedIn PDFs | carousels | Instagram, LinkedIn, TikTok photo mode |
📦 finished/ folder + PUBLISH PLAN.md | publish-pack | Posting, day by day |
Plus, from the pre-production side: a ranked idea, a tested title + thumbnail text, an aligned hook, and a filmable outline — so the next video is faster than the last.
The skills
| Order | Skill | Command | What it does |
|---|
| 0 | onboard | /onboard | One-time setup: channel, Voice DNA, optional keys |
| 1 | ideation | /ideation <topic> | Positioned video ideas with gaps, desire mapping, shock scores |
| 2 | yt-titles | /yt-titles <idea> | Titles ranked against your top performers + thumbnail text |
| 3 | hooks | /hooks <title> | Kallaway-framework hooks (visual + spoken + text), scored |
| 4 | outlines | /outlines <idea> | Recording-ready outline with time budgets + production notes |
| 5 | content-cascade | /content-cascade <url|latest> | Blog + Twitter + LinkedIn (+ lead magnet + guide) |
| 6 | short-form | /short-form <url|latest> | 3–5 clips with hooks, captions, hashtags, timestamps |
| 7 | reels | /reels <video path|url> | Plan 3-6 reels, list hooks to film, and render variants |
| 8 | carousels | /carousels <video folder> | Tweet-style slides, a LinkedIn PDF, and captions per platform |
| 9 | publish-pack | /publish-pack <video folder> | The numbered finished/ folder, captions files, and PUBLISH PLAN.md |
You can also ask in plain English. "Give me title ideas" triggers yt-titles. "Make shorts from my latest video" triggers short-form. "What hooks should I film?" and "attach my filmed hooks" trigger reels. "Make carousels from this video" triggers carousels. "Make the finished products folder", "package everything to publish", and "where should I post what?" trigger publish-pack.
Configuration & optional integrations
Everything optional lives in skills/content-cascade/.env (copy from .env.example). /onboard can write it for you.
- Apify — most reliable transcript fetching, especially for videos uploaded in the last hour. Free tier.
- Supabase — auto-save blog posts and social drafts to a database instead of local files.
- Twitter/X — auto-post the generated thread.
Without any of these, the system still does the full pipeline — it just saves to local markdown and you copy-paste to publish.
Note on competitor scanning: the ideation, outlines, and yt-titles skills can optionally run a YouTube competitor scan via a separate yt-search helper that does not ship with this plugin. Those steps are clearly marked optional and degrade gracefully — every skill produces its full output without them. Your own channel performance data (used by yt-titles) is built in and works out of the box.
FAQ
Do I have to run the skills in order?
No — each works standalone. The order is the recommended end-to-end flow. Already filmed? Jump straight to /content-cascade.
Does this post to my accounts?
Only if you set up the Twitter keys and explicitly say to post. By default everything saves as a draft for your review.
"Captions aren't available yet."
YouTube takes 30–60 minutes to generate captions after upload. Run the cascade again a bit later, or add an Apify token for more reliable fetching of fresh uploads.
How do I switch channels or refresh my voice?
Re-run /onboard — it updates the same config in place.
License
MIT © Samin Yasar — see LICENSE.
