Have you ever wanted Steve Jobs to help you think through product direction, Charlie Munger to help you make an investment decision, or Richard Feynman to teach you a physics concept? That is exactly what the open-source project "Nuwa Skill" does—give it a person's name, and it automatically researches that person's public materials, extracts how they think through problems, how they make decisions, and how they talk, and generates a Skill you can call again and again.

Project address: https://github.com/alchaincyf/nuwa-skill

What Is Nuwa Skill

Nuwa Skill is a tool open-sourced by Huashu, a Bilibili creator (the author of "Kitten Fill Light"). Its core capability is "distilling" a person's way of thinking into a Skill file that an AI can load.

Unlike ordinary "role-playing," Nuwa Skill does not simply have the AI imitate someone's tone of voice. It extracts a complete set of mental models, decision heuristics, and expression DNA from primary-source material, forming a structured thinking framework.

17 figures have already been distilled: Steve Jobs, Charlie Munger, Richard Feynman, Elon Musk, Naval Ravikant, Karpathy, Ilya, Paul Graham, Zhang Xuefeng, Luo Yonghao, and more. Fully open source and free under the MIT license.

Distillation Quality: An Example

Let's test it with the question "How should ordinary people make money in the AI era?"

The distilled Jobs's answer:

Stop—your question itself is the problem. Taste is the real moat. Tools really have been democratized, but taste has not. Do end-to-end things, not parts. Don't ask "how to make the most money in the AI era"; ask whether you have any burning love and curiosity, and then use AI to push it to the extreme.

ChatGPT playing Jobs's answer: standard, correct, but boring boilerplate.

The problem with role-playing is that ChatGPT's own system prompt interferes with the persona's style, and when answering, the AI tends to give you the statistical average of its training corpus—which, to put it bluntly, is mediocrity. Distilled figures come with a complete thinking framework, pulling out the things that really matter from primary sources.

The Distillation Pipeline: A Complete Five-Stage Process

Stage 1: Six Agents Research in Parallel

After you enter a name, Nuwa Skill launches 6 Agents in parallel, each covering one dimension:

The 6 Agents collect information from 40+ primary sources and automatically surface the contradictions in a person's views.

Stage 2: Framework Extraction

Once the research is done, Nuwa extracts five layers of content from the material:

• Mental models: the underlying framework through which the person sees the world
• Decision heuristics: how they judge when a problem comes up
• Expression DNA: their speaking style, word choice, catchphrases
• Signature ideas: the person's most distinctive viewpoints
• Knowledge boundaries: what the person knows and does not know

Take Luo Yonghao as an example: the mental models extracted include pragmatized idealism, aesthetics as productivity, the restless striver's philosophy, expressiveness as leverage, and outsider-intruder thinking.

Stage 3: Building SKILL.md

The extracted content is written into a structured SKILL.md file—this is the final product. Once an AI reads it, it can "load" that person's way of thinking.

Stage 4: Quality Validation

Three kinds of tests are run to ensure quality:

• Known-information test: checks answers for consistency against things the person has publicly said
• Edge test: asks questions the person most likely never answered directly, to see whether the Skill can reason its way to a plausible answer
• Style test: checks whether the tone, word choice, and rhythm of the answers fit

Stage 5: Dual-Agent Refinement

Agent A evaluates the overall structure, Agent B evaluates trigger conditions and rules, and the two reports are merged for one more round of improvement.

Advanced Usage: Assemble a Brain Trust

You can answer the same question with one person's Skill, or switch between distilled figures to get analysis from different perspectives.

For example, ask "What should ordinary people do about AI anxiety?":

• Luo Yonghao's view: anxiety means you can still feel things, but once the anxiety passes you have to act. Don't learn AI knowledge—learn how to use AI to get work done
• Naval's view: the real fear is not that AI is too strong, it is that you are too easy to replace. Separate work that can be written into a manual from work that depends on judgment

You can assemble a brain trust of dozens of people and have them analyze the same question according to their own ways of thinking.

Quick Start

Install with one command:

npx skills add alchaincyf/nuwa-skill

After installation, just drive it with natural language:

• "Distill a Luo Yonghao"
• "Analyze this investment decision from Munger's perspective"
• "Switch to Naval; I'm torn about three things"

If the figure you want to distill is fairly niche, it is best to prepare local source material yourself and feed it to Nuwa Skill.

Use Cases

• Quickly understand how a person thinks, without reading every book and interview they have ever done
• Analyze the same question from multiple perspectives
• Give an AI coding tool the decision-making style of a specific person
• Build your own "virtual brain trust"

References

• GitHub project address: https://github.com/alchaincyf/nuwa-skill
• Bilibili demo video: https://www.bilibili.com/video/BV1BXQABNE4y