South A Claude Code Content-Engine Playbook
Stop Writing
Blog Posts.
How we used Claude to turn our meeting notes, voice memos, and Slack threads into a blog that publishes daily.
The thesis
The goal wasn’t AI that sounds human.
It was a system for AI to create rankable content.
Reads your thinking
Meeting transcripts, voice memos, and Slack threads go through Cowork to extract themes. Those themes seed the planner. Every post starts from real work.
Researches with real data
Scout adds current SERP data, pricing, and named tools to each topic. WebSearch and an optional SERP MCP pull fresh numbers and competitor word counts.
Audits itself against editorial and SEO bars
Writer drafts against the voice skill. Creative places visual components. SEO Auditor runs the full checklist. Em-dashes hard fail. Nothing ships until it passes.
Publishes
Commit, push, promote to main via GitHub Actions, notify IndexNow. Vercel deploys. No human in the loop for auto-publish tracks.
What you get
Three pieces. Two are free.
The primer · Free
Stop Writing Blog Posts
The 10-minute version
Eleven pages. The architecture, the daily pipeline, the voice rules, the cost breakdown. What you read on a Tuesday afternoon before deciding whether to build this.
Download the primerPDF · 129 KB
The full playbook · Free
Stop Writing Blog Posts
The 40-page playbook
Every voice rule. Every thing that bit us. Every prompt excerpt. The full build documentation, written to be read cover-to-cover or dropped into an agent’s context.
Download the playbookPDF · 447 KB · Markdown for agents
The toolkit · Free
The Claude Content Engine Toolkit
Twenty working files. Nine prompts (Scout, Captain, five sub-agents, Bootstrap). Five skills. Four scripts including the GitHub Actions promotion workflow. Drop into a Next.js + MDX repo and go.
Download the toolkitZip · 36 KB
What’s inside the book
Forty pages of working specifics.
The 7-agent architecture
Two scheduled agents (Scout, Captain) plus five sub-agents (Writer, Creative, Affiliate Linker, SEO Auditor, Publisher). Three data files. Five skills. One safety hook. The whole system on one diagram.
The insight-extraction step
Fifty meeting notes, voice memos, and Slack threads in. Twenty-four distinct topic ideas out. Seven recurring theme categories become the track assignments. Why your planner seed is the most important file in the system.
The voice rules
Specific phrases to kill (delve, leverage, landscape, seamlessly). The Kill / Use instead table. Three recurring AI structural patterns and the prose-level fixes. Why "add personality" is the wrong instruction.
What bit us — nine real platform gotchas
Including the headline constraint: Claude Code Routines can only push to claude/* branches, not main. The GitHub Actions promotion workflow that works around it. Eight more specific hours-long failures.
The cost breakdown
Claude Pro at $20/mo is the floor. $20-60/mo fully loaded with optional SERP MCP, domain, and transcription. $100-140/mo on Max. Cheaper than a freelance writer producing one post a month.
How to use the book with your agent
Drop the markdown into your project context. Paste the bootstrap prompt to scaffold a fresh build. Point your agent at southarcdigital.com/llms.txt for URL-based research.
This is for you if
- ●You run an agency or you build solo, you ship client work all day, you never have time to publish content.
- ●You know SEO matters and that AI slop does not work, but you have not found a system that produces posts you’d put your name on.
- ●You are comfortable with git, terminal commands, and editing markdown skill files.
- ●You want a system you own, not another SaaS tool that writes generic content you rename.
This is not for you if
- ○You want a no-code tool with a dashboard. This runs on Claude Code Routines, a Next.js repo, and a few scripts.
- ○You want a “10x your traffic” pitch. We have published ten posts so far. Authority compounds over months.
- ○You want to rotate Reddit accounts, buy backlinks, or run PBNs. That is the other side of the AI content market.
Who wrote this
Vignesh Ramakrishnan
Founder · South Arc Digital
Former lawyer, current AI engineer. I build production AI systems for non-tech businesses in India, mostly bridging legacy ERPs, WhatsApp-first workflows, and voice interfaces with modern LLM infrastructure.
South Arc Digital has shipped to clients across textiles, industrial gas, FMCG distribution, and legal analytics. We also run Nicherly, a SaaS product for AI-automation agency owners that replaces manual prospecting with AI-personalized outreach.
The content engine this book describes is what we built when we realized we needed to publish to build authority, could not afford the time to write by hand, and refused to publish slop. If you build your own version and it works, let us know.
vignesh@southarcdigital.com · ram@southarcdigital.com
Questions
The things everyone asks.
What Claude plan do I actually need?
Pro at $20/month is the floor and includes Claude Code, Cowork, and cloud routines. That has been sufficient for our run rate of one Captain per day plus one Scout per week plus interactive dev. Max at $100/month gives 5x the usage headroom if you publish more aggressively or use Claude Code heavily for client work on the same account.
What if I don’t use Cowork?
The shape of the input matters more than the tool. Any extraction pipeline that turns your meeting transcripts, voice dumps, or written thinking into a list of themes works. Whisper plus a Claude summarization prompt works. Fathom, Granola, or Otter work. What you want to avoid is starting the planner from a keyword tool’s list of "trending queries," because that is where the genericness enters.
How does this compare to n8n, ZimmWriter, or SEO content generators?
Those are generic systems that produce posts from prompts. This one starts from your own thinking (meeting notes, voice memos) and ends at published posts that sound like you wrote them. The architecture, voice rules, and grounding step are all calibrated around that specific difference. You can read the book for free and decide whether that distinction matters to you.
Can I skip the toolkit?
Yes. Everything the toolkit contains is described in the book with enough specificity that a coding agent can rebuild it. The toolkit exists so you do not have to extract the prompts and scripts from the text yourself. If you have the agent do it, you save an hour.
Why is the whole stack free?
The book is positioning content for South Arc Digital and Nicherly. We would rather every serious operator read it than nickel-and-dime the download. If you build the engine, ship great posts, and one of them links back, that is the payoff we wanted. If you want to hire us for custom integration work later, the contact form is in the footer.
Is this still current?
As of April 2026, yes. Claude Code Routines is in research preview and platform behavior (especially the claude/* branch restriction, documented as the headline gotcha) may change. We will keep the llms.txt and the markdown updated. Subscribe to the South Arc Digital journal if you want notifications.
Future playbooks
Want to know when we ship the next one?
We publish one of these every few months. No marketing sequence, no drip. Just the next playbook when it is ready.
One email per playbook. Unsubscribe any time.
Ready
Grab the primer.
It’s ten minutes.
No sign-up. No email required. Just the file.