Nuwa: distill a public figure's way of thinking into a reusable skill

Who it's for

Use Nuwa when you want a public figure's way of thinking as a skill you can call again, so later decisions are analyzed with that mental model and those heuristics. What it does is “distill”: pull a framework out of public statements, not copy inspiration or private thoughts.

Accept three hard limits first: intuition cannot be distilled (a framework can be extracted, inspiration cannot); sudden shifts cannot be captured (it is only a snapshot as of research time); public statements are not real thoughts, and you can work only from public writings. If the task is to recover unpublished judgments, or to treat the distillate as the person's still-complete thinking today, do not take this path.

The runtime precondition is an agent compatible with the Agent Skills protocol. The notes name Claude Code, Codex, Cursor, OpenClaw, Hermes Agent, CodeBuddy, Workbuddy, Gemini CLI, OpenCode, and more, and write 50+ compatible runtimes; that is an example of the compatibility range. Do not write that only listed environments work, and do not invent install steps for unnamed platforms.

First try in 3 minutes

Official page:

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

Protocol notes:

https://agentskills.io/

Recommended install is only this one. First confirm the repository is alchaincyf/nuwa-skill, then run in a local terminal:

npx skills add alchaincyf/nuwa-skill

The notes do not say which directory the files land in. After install, do not invent a path; only check in a currently compatible agent whether the skill can be invoked. If the command fails, stop; do not switch to an installer that never appeared.

The first time, do not generate a new skill file, and do not use the distillate to lock a real decision. Pick a public figure (the notes' examples are Paul Graham, Zhang Xiaolong, Duan Yongping) and only ask: which frameworks can be pulled from public statements, and what each of the three limits — intuition, sudden shifts, and public statements versus real thoughts — will miss.

Success signal: the answer stays on “distilling a way of thinking” and names those three limits on its own. If the agent starts writing a new Skill file, stop it first, confirm the side effects, then continue.

Use cases

Case 1: Distill a named person's perspective skill. After install, you can say “distill a Paul Graham”, “make a Zhang Xiaolong perspective Skill”, or “help me make a Duan Yongping Skill”. This is a file-writing operation: first have the agent list the frameworks and source boundaries it will extract; after you agree, generate the local skill, then open the new file and check it matches those items. The notes only demonstrate this “make a so-and-so Skill”; they give no batch distill or non-public-figure method.

Case 2: Use an already distilled way of thinking to analyze a problem. The official sentence is to have Jobs, Musk, Munger, and Feynman work for you — that is, look at one concrete problem through that framework. First pick a reversible analysis (compare options, break down risk); do not start by changing production config. The notes write no slash command and no fixed filename; the only path that exists is “analyze with that perspective”. After analysis you still judge yourself, because public statements are not real thoughts.

Boundaries

Intuition cannot be distilled, sudden shifts cannot be captured, and public statements cannot be treated as real thoughts. These three are boundaries the skill wrote itself, not usage tips.

The install command is written only as npx skills add alchaincyf/nuwa-skill. There is no official landing directory; do not add an unnamed path or a second installer.

Named clients come from compatible-runtime examples and write 50+; do not infer “only these work”, and do not apply this npx command to marketplaces or platforms the notes never named.

The notes give no language coverage, pricing, or operations besides distill / invoke a perspective. Keep undocumented items unknown.