Self-Distillation: Repurposing Long-form Articles to Viral Social Posts with Open-Source Agent Skill
Author: 泥瓦匠 @bysocket
Format: Course Article
Est. Read Time: 3 min
Author: Niwajian @bysocket•Date: 2026-07-27 10:00
Special thanks to ChenChang @changchen_cc for sharing insights on KOL Marketing and viral content distribution logic during the Workshop. Deeply inspired by his work!
1. The Results
The GitHub open-source repository for the Agent Skill is available here:
https://github.com/JeffLi1993/content-repurposing-skills
Let's walk through a live case study:
Source Article:https://bysocket.com/seo-link-building-conversion-data/
As shown above, it outputs the following two structured results:
Result 1:
Result 2:
Select Post 2, and it immediately suggests visual image options:
Pick an image option, and automatically publish the generated visual + copy!
2. Why Build This?
Content creators know that finishing a 3,000-word post is just the beginning. The real time sink is distribution.
Splitting a long article into short posts for LinkedIn, X, or RED requires tweaking tone, changing structures, drafting hooks, and creating images for every single platform.
Having spent half a year handling manual distribution, I can summarize it in one word: repetitive. Every publish follows the same steps:
- Extracting 2-3 core insights from the long post
- Rewriting openings per channel (LinkedIn prefers professional narrative, X demands punchy brevity)
- Drafting hooks that stop scrollers in their tracks
- Adding engagement questions and soft CTAs
- Pairing with an image that conveys the core message
80% of the work is strategic repetition, and only 20% requires human judgment.
When building seo-audit-skill, I used a Script + LLM dual architecture where scripts handled deterministic checks and LLM parsed semantics.
But content repurposing is different: there are no APIs to call and no HTML to parse. The core is strategic knowledge. What hook captures attention? How to extract different angles? How to make 3 posts distinct yet complementary?
So the idea here is: Distill strategic knowledge into Prompts, allowing the Agent to execute my methodology.
3. What Can It Do?
Give it any long-form content (blogs, whitepapers, case studies, guides), and it outputs 3 differentiated social posts for LinkedIn and X.
Each post includes:
- ICP Deduction: Automatically infers target audience persona, pain points, and awareness stage.
- Hook Selection: Evaluates 5 candidates and picks the highest-scoring opening.
- Value Content: Delivers source-backed value instead of vague summaries.
- Engagement Prompt: Questions designed to trigger reader comments.
- Soft CTA: Natural transition to action without hard selling.
- First Comment CTA: Conversion links optimized for comment section pin.
The 3 posts are not rewordings of the same post, but a coordinated campaign package. After post selection, it suggests 3 visual graphic options for instant generation.
4. How to Use
Two installation methods:
Method 1: Install via CLI
npx skills add JeffLi1993/content-repurposing-skills --skill long-blog-to-viral-social-posts
Method 2: Install as Plugin
/plugin marketplace add JeffLi1993/content-repurposing-skills/plugin install long-blog-to-viral-social-posts
Once installed, simply prompt: Repurpose this post into 3 social posts: https://your-blog-url.com/post. It automatically scrapes, cleans, extracts, and generates the assets.
5. Core Design Principles
The biggest difference between this Skill and seo-audit: Zero script code, 100% strategic knowledge distillation.
The logic of seo-audit was "don't let LLM guess what code can verify" - an HTTP request easily verifies if robots.txt exists. But content repurposing has no deterministic checks; it is entirely strategic decision-making.
So what I did was encode my experience from doing hundreds of manual content repurposing runs into an executable workflow:
1️⃣ Content Atom Framework: Long posts shouldn't be summarized, but atomized. I defined 9 Content Atoms (Claims, Data, Insights, Frameworks, Stories, Proof, Quotes, Objections, Resources) for the Agent to extract before drafting.
2️⃣ Combination Strategy: The 3 posts cannot share the same Atom, Hook family, or engagement mechanism, preventing repetitive packaging.
3️⃣ 70/20/10 Scoring Model: Evaluates posts out of 100 points: Attention accounts for 70 (Hook 25 + Claim 15 + Insight 15 + Pattern Interrupt 15), Engagement 20, Conversion 10. Anything below 80 points is automatically revised.
4️⃣ Hook Generation Flow: Generates 5 candidates across different families and scores them based on tension, curiosity gap, conciseness, and ICP relevance.
5️⃣ De-AI Writing Rules: Restructures phrasing using native social media rhythm, avoiding robotic translationese and generic AI buzzwords.
The entire Skill consists of a SKILL.md + 2 reference docs (Hook templates and Platform rules), without script dependencies.
6. Project Directory Structure
content-repurposing-skills/
└── long-blog-to-viral-social-posts/
├── SKILL.md # Skill definition + 12-step workflow
└── references/
├── hook-templates.md # Hook template library
└── platform-rules.md # Platform rules for LinkedIn / XGet in Touch
This tool was born out of my own content distribution needs. Hope it helps fellow content creators!
If you:
- Often write long posts but lack distribution time - Use it directly
- Find any issues with generated posts - Submit an Issue
- Have better Hook templates or platform rules - Open a PR
Open source & free. Looking forward to your feedback!
Final Remarks
Self-distillation means turning implicit experience into explicit rules, and encoding explicit rules into an executable workflow.
AI cannot replace you in this process. It can execute your strategy, but you must pave the strategy yourself first.
So my advice remains: Do it yourself first until you can teach others, then let AI scale it for you. Hope this tool saves you distribution time so you can focus on writing better content 🚀