Claude Skills · Prompt library · Style transfer
article-writing-skills — A Style-Transfer Library for LLM Writing
article-writing-skills is a deliberately opinionated prompt library. Each skill takes the rough notes you have and rewrites them in the cadence of a specific engineer or researcher you respect. It runs as a Claude Code skill, but the prompts are LLM-agnostic — drop them into ChatGPT, Gemini, or your own MCP server and you get the same effect. I started it because I wanted to know how far prompt-shaped style transfer could actually go without fine-tuning.
Why I built it
Generic "professional blog post" prompts produce generic professional blog posts. I wanted to know how far prompt-shaped style transfer could go without fine-tuning — could a sufficiently detailed system prompt actually impart voice, sentence shape, structural preference?
The library is the public testbed for that experiment.
How it works
Each skill is a Markdown file with a long system prompt: the writer's stated principles, examples of their cadence, do/don't lists, and structural rules. When you invoke the skill, Claude (or any LLM) loads the prompt as context and rewrites your draft against it.
The skills compose. You can stack a "voice" skill on top of an "argument structure" skill and get something that reads like the target writer thinking about your topic. (This very page was rewritten with the karpathy-article-writing skill out of the same library — meta, I know.)
What I take from it
Prompt-shaped style transfer is good enough to be useful in real LLM engineering workflows, more often than I expected. You don't always need a fine-tune — a 600-token system prompt and the right examples will land 80% of the effect for 0% of the training cost. The remaining 20%, I suspect, is where fine-tuning starts to actually matter.