Lago CEO Anh Tho Chuong open-sourced a Claude Skill that captures their writing voice. The project pulled over 4,000 points on HackerNews. It works by reverse-engineering years of hand-written content, zero AI assistance in the loop, to capture the gap between what an AI drafts and what Chuong actually sends. That gap is the voice. The template lives at github.com/getlago/inside-lago-voice-skill with seven sections including Voice, Core Rules, Anti-Filler Checklist, Audience Adaptation, Channel-Specific Notes, Drafted vs Sent comparisons, and Company Context.
Chuong isn't a native English speaker and spent years treating their direct, shorter-sentence style as a limitation. Building the skill flipped that perspective. The style became a feature, not a bug. The skill handles structure, formatting, repetition. It saves hours weekly. But Chuong is clear: the emotional core can't live in a Claude Skill. That part stays human.
The 'Drafted vs Sent' section matters most. It captures what you keep, what you always delete, how you rewrite every AI draft before hitting send. To build your own, Chuong recommends analyzing 10 to 20 pieces of hand-written content. swyx, who co-organized the Write and Learn workshop where Chuong presented this work, noted they'd never seen someone hit HN's front page so often while also being vulnerable in their writing.
Commercial tools like Jasper.ai and Copy.ai offer brand voice features. They focus on marketing content generation rather than capturing how an individual actually writes. Anthropic's Claude Skills and OpenAI's Custom Instructions handle preference persistence. They can't replicate your full style end to end. Chuong's approach is different because it explicitly encodes the delta between AI output and human-edited final versions.
The template is the easy part. The fuel comes from having something real to say and reading widely enough to process ideas deeply. That's what invites people into your world.