A framework for teaching AI to write like you. Not like a better version of you. Like you.
I built a Claude Skill that writes in my voice. It captures my anti-patterns, my audience splits, the gap between what AI drafts and what I actually send. It saves me hours every week.
People asked me to share it. The skill itself wouldn't help you — it's calibrated to me. But the structure works for anyone.
This template has the same categories my skill has, empty for you to fill in.
SKILL-TEMPLATE.md — the template. Seven sections:
- Voice — how you actually write (not how you wish you wrote)
- Core Rules — your non-negotiables, derived from your own past writing
- Anti-Filler Checklist — the patterns you always delete from AI drafts
- Audience Adaptation — your different registers (internal vs external, technical vs business)
- Channel-Specific Notes — what changes per platform (LinkedIn vs HN vs email)
- Drafted vs Sent — real examples of AI drafts vs what you actually sent, with lessons
- Company Context — how you talk about your product, competitors, positioning
- Fork this repo
- Read 10-20 things you wrote by hand, without AI. Look for what's consistent.
- Fill in each section. Be specific. "Write clearly" is useless. "Cut any sentence that announces the smart thing before saying it" is actionable.
- The most important section is Drafted vs Sent. Every time you rewrite an AI draft before sending, save both versions. The gap is the voice.
- Keep calibrating. This is a living document.
Save the filled-in template as a Claude Skill or drop it into your system prompt. Claude will use it as a style guide for everything it writes on your behalf.
The template is the easy part. It gives you structure. What it can't give you is something worth saying. Read widely, absorb context, sit with ideas until they're yours. The voice skill encodes the craft. The soul is on you.
This template comes from a talk at @swyx and @MilksandMatcha's Write and Learn workshop. The original skill was built by studying years of hand-written content — no AI in the loop — and reverse-engineering the patterns.
Built by Anh-Tho Chuong, CEO & Co-founder at Lago — open-source billing infrastructure.