Guide · Agent skill
How to use the Nuwa Persona Distiller agent skill (huashu-nuwa)
Updated · written with AI from the project's own documentation and checked against our catalogue data
Nuwa Persona Distiller (skill name huashu-nuwa) is an agent skill that builds a reusable 'thinking advisor' persona from a named person or a vague need, capturing how that person reasons rather than what they said. This guide is for readers who have never used agent skills and want to install it, trigger it, and know what it will do and cost. By the end you will know how to install the ZIP, how to phrase a request, what the run involves and what limits to expect.
Nuwa Persona Distilleralchaincyf/nuwa-skill · MIT Download the skill Quick start
- 1Download and unzip. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Confirm the skill loads. Start your agent runtime and make sure it recognises the huashu-nuwa skill, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Make a request. Say something like "distill [person]" or describe a need such as better decision-making.
- 4Choose a cost tier. Pick quick, standard or deep after the skill explains the approximate cost level.
- 5Add local material (optional). Provide books, transcripts, subtitles or essays so the skill can use primary sources first.
- 6Review the research folder. Check the references/research files the sub-agents wrote inside the skill folder.
- 7Use the new persona SKILL.md. Load the generated persona and test it, paying attention to its stated limits.
What is an agent skill and what does huashu-nuwa do?
An agent skill is a folder containing a SKILL.md file that an agent runtime, such as Claude Code, loads to learn a new workflow. huashu-nuwa is one such skill. When it runs, the agent asks a few questions, estimates cost, creates a folder of research files, and runs several parallel research sub-agents that search the web and optionally analyse local books, transcripts or subtitles.
The output is a new persona SKILL.md. The source describes it as a working cognitive toolkit covering mental models (the lenses the person sees the world through), decision heuristics, expression style, anti-patterns and stated limits (honest boundaries).
- Captures how a person thinks, not what they said
- Can start from a named person or from a vague need such as better decision-making
- Output folder is self-contained so it can be copied and used on its own
Install the huashu-nuwa skill
Option 1: download the ZIP from this page and unzip it into your personal skills folder ~/.claude/skills/, or into a project's .claude/skills/ folder so it applies only to that project.
Option 2: in the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP there.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need a special command. The skill's description lists trigger phrases. Chinese triggers include 「蒸馏XX」, 「造skill」 and 「XX的思维方式」. English triggers include "distill [person]", "nuwa", "create a [person] perspective skill", "how does [person] think" and "I need a thinking advisor".
If you already know who you want, name the person. If you only have a need, describe it, and the skill will ask one or two follow-up questions and recommend candidates before distilling.
- "Distill Charlie Munger"
- "Create a Feynman perspective skill"
- "How does Naval think about investing?"
- "I want to improve my decision-making. Who could help me think about it?"
Understand the cost tiers before you start
The skill asks you to choose a tier before it runs. A full distillation is a long multi-agent, multi-round task, and the source says a single run on top-tier models can cost tens of dollars (a real user case). Choose the tier deliberately.
Quick uses 3 dimensions with up to 5 sources each and costs about one third of the standard tier. Standard (the default) covers 6 dimensions and costs a medium amount; a lighter model can lower it significantly. Deep adds full downloads of primary material and is the most expensive. If you say only "make XX", the skill defaults to a full profile, thinking advisor role, web search only and the standard tier.
- Quick: about one third of standard cost, good for trying it out
- Standard (default): 6 research dimensions
- Deep: 6 dimensions plus full download of primary sources, highest cost
Use your own sources for better results
If you have books (PDFs), transcripts, video subtitles, blog exports or essays by the person, provide them at the start. The source says local primary material is usually of much higher quality than second-hand web summaries, and the skill switches to local-corpus mode, searching the web only for dimensions your files do not cover.
The skill's own output flags whether content came from your files or from web search, and it keeps contradictions rather than smoothing them over.
- Local-corpus-first mode: your files are read first, then gaps are filled by search
- Pure local mode: say you want only your material used, or distil a non-public figure
- Subtitle helper: download_subtitles.sh fetches subtitles, then srt_to_transcript.py cleans them into plain text
bash [skill目录]/scripts/download_subtitles.sh <YouTube_URL> [输出目录]Safety notes: scripts and network access
This skill runs scripts and uses the network. It requires a Bash-capable environment to run the bundled subtitle download scripts, and web search plus parallel sub-agent capability. Optional sources it mentions include Z-Library/LibGen for books and transcript sites such as podcastnotes.org.
Review the SKILL.md and the scripts before running, and check what the agent is about to execute. The community index notes that listed third-party projects are unaudited and should be checked, especially their skill instructions and scripts. Do not use the output to impersonate real people, and do not distil private individuals without consent. The output is a synthesised interpretation of web-sourced material and can contain errors or outdated information, so verify facts before relying on them.
- Runs scripts: yes
- Uses network: yes
- Do not create personas of private individuals without consent
- Verify any factual claims in the output
Tips for better persona results
Pick a person whose thinking you can name clearly, and say what you want the persona for: thinking advisor, decision reference or role play. Give a focus if you want one dimension rather than a full profile. Check the limits section of the result, because every good persona states what it cannot do.
For a first try, use the quick tier on a lesser-known person to see the output format before committing to a standard run.
- Be specific about the purpose and the focus
- Provide local sources when you have them
- Read the honest boundaries and anti-patterns sections
What people use it for
- Building a 'thinking advisor' persona modelled on a specific author or thinker
- Exploring how multi-agent research pipelines and self-contained skill folders are built
- Creating a persona mainly from your own books, transcripts or essays
- Finding which thinker's mental models match a decision or communication problem
Frequently asked questions
How do I install the huashu-nuwa skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
What do I need to run huashu-nuwa?
You need an agent runtime that loads SKILL.md skills, such as Claude Code, a Bash-capable environment for the subtitle scripts, and web search with sub-agent capability. Access to video, podcast or book sites is optional.
Is huashu-nuwa free?
The skill is published under the MIT licence. Running it costs money through the models you use, and the source says a full distillation on top-tier models can cost tens of dollars, so check the tier before starting.
Is it safe to run huashu-nuwa?
The skill runs scripts and uses the network, so review SKILL.md and the scripts before running it. Do not use it to create personas of private individuals without consent or to impersonate real people.
How accurate is the persona it creates?
The output is a synthesised interpretation of mostly web-sourced material and can contain errors or outdated information. Verify any facts you rely on, and do not use it for tasks that need verified facts.
What are alternatives to huashu-nuwa?
The community index in the repository lists other persona distillation collections and orchestration projects, which are maintained independently and unaudited. Check each one's scripts and instructions before using it.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use huashu-nuwa if you want a reusable thinking-advisor persona built from a specific thinker, or you are studying multi-agent research pipelines, and you can accept a long, potentially costly run. Pick something else if you need a cheap one-click tool or verified facts, or if you want to build a persona of a private person.
Download the skill