What I Built
The Viral Loop Engine — a 3-skill OpenClaw pipeline that analyzes your social media analytics, extracts the patterns behind your best-performing content, and automatically generates new posts following those winning patterns.
Content creators often post blindly. They see their engagement numbers go up or down but can't systematically extract why certain posts go viral and others flop. Repetitive content creation without a strategy wastes time and misses the opportunity to replicate past successes.
This project solves that by turning your analytics into an automated, data-driven content assembly line. Stop guessing and start replicating what actually works.
How I Used OpenClaw
I built 3 custom, highly-specialized OpenClaw skills that work as a multi-stage orchestration pipeline. All of this runs completely locally via Ollama (llama3.1:8b and qwen2.5:3b) so no data is sent to the cloud!
Skill 1: Data Analyst
Parses CSV analytics data (likes, shares, comments, impressions) and calculates a custom Weighted Engagement Rate ((likes*1) + (comments*2) + (shares*3) + (saves*2.5)) / impressions * 100. It then outputs the top 5% of posts as "Winning Case Studies."
Skill 2: Pattern Architect
Analyzes the winning posts for common denominators by evaluating 5 dimensions (Hooks, Structure, Emotional Tone, Content Specificity, Platform Signals). It outputs a concrete "Winning Playbook" with 3-5 data-backed rules.
Skill 3: Strategic Creator
Takes the Winning Playbook + any new source content (e.g., an article link or raw text) and "skins" the new content using the proven patterns. It generates platform-optimized variants (LinkedIn, Twitter/X, Instagram) with rule citations.
Architecture: By splitting the logic into 3 distinct markdown skills, each part of the process can be refined independently. The AGENTS.md and SOUL.md enforce a strict output structure, ensuring OpenClaw doesn't hallucinate data or skip analysis steps.
Demo
Here is the Viral Loop Engine in action running completely locally via the terminal.
Step 1: Data Analysis (openclaw agent --message "Analyze analytics")
The agent parsed the CSV and found the top 3 posts based on the Weighted Engagement Rate (WER):
🏆 Winning Case Studies
Winner #1: "Just launched my new course..."
- Platform: Instagram
- WER: 10.60%
- Primary Metric: Saves (189)
Winner #2: "My morning routine changed..."
- Platform: Instagram
- WER: 10.46%
- Primary Metric: Saves (145)
Winner #3: "I turned down $1M..."
- Platform: LinkedIn
- WER: 10.38%
- Primary Metric: Shares (234)
Step 2: Pattern Architecture (openclaw agent --message "Create playbook")
The agent analyzed the winners and output the Winning Playbook:
📋 Winning Playbook
Rule 1: High Stakes Hooks
Pattern: Winners lead with high-impact life changes or large dollar amounts ($1M).
Application: Start the post with a specific, high-stakes claim.
Rule 2: The "Hidden Secret" Curiosity Gap
Pattern: Using phrases like "Here's why" or "changed everything".
Application: End the hook by teasing the explanation.
Rule 3: High-Value Saveable Content
Pattern: The top 2 posts won because of Saves.
Application: Provide a bulleted list of tools or steps people will want to refer back to.
Step 3: Strategic Creation (openclaw agent --message "Create post about AI coding tools")
The agent applied the playbook to generate a new optimized LinkedIn post:
🔗 LinkedIn Version
Rules Applied: Rule 1 (High Stakes Hook), Rule 2 (Curiosity Gap), Rule 3 (Saveable List)
---
I saved 400 hours of development time this year. Here's how.
Most developers are using AI wrong. They treat it like a search engine instead of a junior developer.
Here is the exact stack that changed everything for me:
1.OpenClaw for autonomous agent orchestration
2. Ollama for private, local LLM execution
3. VS Code for a seamless development experience
Stop doing repetitive work manually. Set these up today.
What is your go-to AI coding tool? Let me know below.
---
📺 Watch the Video Demo Here (https://drive.google.com/file/d/1B2erGMX1v33su42nhvp8WE3TJXtEF47i/view?usp=drive_link)
What I Learned
Building with OpenClaw is fundamentally different from building standard API wrappers. By defining the agent's identity via SOUL.md and USER.md, we didn't have to keep re-prompting the model with our preferences it works autonomously, that is the interesting part of it. It felt less like writing code and more like writing an employee handbook (AGENTS.md).
The composability of SKILL.md files meant we could test our analytics parsing independently from our content generation. Overall, OpenClaw provides a brilliant orchestration layer that bridges the gap between raw LLM intelligence and actionable, multi-step workflows.
Team Members
This project was built in collaboration with:
- @ElielBright
- @jaayy123
Repo & Documentation
You can find the full source code, sample analytics CSV, and setup instructions in the repository below.
Top comments (1)
This is a brilliant use of the OpenClaw orchestration layer. 🛠️ Most people are just prompt engineering, but using a 3-skill pipeline to calculate a Weighted Engagement Rate and then 'skin' new content is next-level automation. The 'Strategic Creator' agent citing its own rules is the transparency we need in AI content. Definitely worth a read if you're looking into local LLM workflows (llama3.1)