UltraSkills
Back to Blog
AI Automation

How We Built a Full AI Content Pipeline With Claude Code

A step-by-step breakdown of how we use Claude Code to generate, refine, and schedule daily social media content autonomously.

Mar 18, 20263 min read

Six months ago our team was spending 4 hours every day writing, editing, and manually posting content across five platforms. Today that same output runs automatically — and it costs us $34/month in tools.

Here's the exact pipeline we built using Claude Code, n8n, and a handful of open-source tools.

The Architecture

The pipeline has four stages: Generate → Refine → Schedule → Publish. Each stage is a separate n8n workflow triggered by a webhook. Claude Code handles the intelligence layer.

Stage 1: Content Brief → Script

Every morning a cron job fires a webhook with a topic brief from Supabase. Claude Code reads the brief and writes a full script — hook, body, CTA — in the brand voice I've trained through a system prompt.

// Claude Code skill: generate-script.md
# Generate Video Script
Given a topic brief, write a 60-second video script.
Format: HOOK (8s) → PROBLEM (15s) → SOLUTION (30s) → CTA (7s)
Voice: Direct, confident, no fluff.

Stage 2: Refinement Loop

The raw script goes through a second Claude Code pass that evaluates hook strength on a 1–10 scale. Anything below 7 gets rewritten. This two-pass system eliminated our worst-performing content before it ever got published.

Stage 3: Format Variants

One script becomes five formats automatically: LinkedIn long-form, Twitter thread, Instagram caption, TikTok hook variant, and a newsletter blurb. Each has its own formatting rules baked into a Claude Code skill.

Stage 4: Publish

n8n schedules each format to the right platform at the optimal time. Supabase tracks every post's performance. The feedback loop feeds back into the content brief generator.

The Result

12x content output. 20+ hours reclaimed per week. Zero copy-paste. The entire pipeline runs while we sleep.

The key insight: Claude Code doesn't replace your taste — it executes it at scale. The skills we wrote encode our judgment. The automation just runs it faster than we can type.

Related Posts