# Cangjie‑Skill vs Nuwa‑Skill: Key Differences in AI Knowledge Distillation

> Discover the key differences between Cangjie-skill and Nuwa-skill. Cangjie-skill creates executable knowledge skills from texts, while Nuwa-skill replicates personal thinking for role-play.

- Repository: [kangarooking/cangjie-skill](https://github.com/kangarooking/cangjie-skill)
- Tags: comparison
- Published: 2026-08-14

---

**Cangjie‑skill distills systematic knowledge from books and long‑form content into executable skills, while nuwa‑skill distills a person's thinking style and expression DNA for role‑play scenarios.**

Both tools belong to the same AI‑skill ecosystem developed in the `kangarooking/cangjie-skill` repository, yet they serve fundamentally different purposes. Understanding their distinct architectures helps you choose the right approach for your use case—whether you need actionable frameworks from a bestseller or a believable persona based on a real individual.

---

## What Cangjie‑Skill Distills: Content, Not People

**Cangjie‑skill** targets high‑value *content artifacts*: books, video transcripts, podcasts, courses, and interviews that contain structured knowledge. Its mission is **knowledge extraction and operationalization**.

According to [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) (line 16), cangjie‑skill explicitly does **not** handle "simple summarisation, book‑review style output, or author‑persona role‑play." These fall under nuwa‑skill's domain.

### Core Use Cases

- "拆书" (book dismantling) — extracting frameworks, checklists, and case studies
- Converting video transcripts into composable agent skills
- Pressure‑testing extracted knowledge before deployment

The methodology follows a **7‑stage RIA‑TV++ pipeline** detailed in `methodology/00‑overview.md` (line 27). This includes parallel extractors, triple verification, and pressure testing—architectural choices optimized for fidelity and actionability.

---

## What Nuwa‑Skill Distills: People, Not Content

**Nuwa‑skill** operates on a completely different input: *individuals*. It captures the **thinking style** and **expression DNA** of a specific person—how they reason, what phrasings they prefer, their decision‑making patterns.

### Core Use Cases

- Role‑play: "让 AI 扮演马斯克" (make the AI act as Elon Musk)
- Generating responses in a departed colleague's voice
- Reproducing an author's reasoning style for authentic dialogue

As noted in [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) (lines 154‑156), nuwa‑skill provides the "human" half of the ecosystem. Its outputs can be further evolved by **darwin‑skill** for automatic refinement.

---

## Architectural Comparison: RIA‑TV++ vs Distill‑Person

| Dimension | Cangjie‑Skill | Nuwa‑Skill |
|-----------|---------------|------------|
| **Input type** | Books, transcripts, courses, podcasts | Individual personas (Musk, Buffett, etc.) |
| **Output type** | Executable, composable Claude skills | Human‑skill for role‑play and voice emulation |
| **Pipeline name** | RIA‑TV++ (7 stages) | "Distill‑person" pipeline |
| **Key constraint** | No persona role‑play ([`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md), L16) | No systematic knowledge extraction from long content |
| **Ecosystem role** | Content‑skill generator | Human‑skill generator |

The RIA‑TV++ pipeline's "TV" component—**Triple Verification**—was actually borrowed from nuwa‑skill's verification approach, showing technical cross‑pollination between the projects.

---

## Code Example: Cangjie‑Skill in Practice

Here's how you define a cangjie‑skill distillation task, as specified in [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) (line 2):

```yaml
name: cangjie-skill
description: |
  Distill a book, long‑video transcript, podcast, course, or interview
  into a coherent set of executable skills.
input:
  title: "Deep Work"
  content: |
    # Chapter 1 … (full text of the book)

    ...
output:
  skills:
    - name: "Focus‑Batching"
      type: framework
      description: |
        A step‑by‑step framework for batching deep‑work sessions.
    - name: "Distraction‑Elimination Checklist"
      type: checklist

```

When executed, this triggers the full RIA‑TV++ pipeline and returns a **skill pack** that agents can invoke directly.

---

## Code Example: Nuwa‑Skill Contrast

For comparison, here's the conceptual structure of a nuwa‑skill task (implemented in a separate repository):

```yaml
name: nuwa-skill
description: |
  Distill a person (thinking style, expression DNA) into an AI skill.
input:
  persona: "Elon Musk"
  source_material:
    - tweets.txt
    - interview_transcripts.md
output:
  skill:
    - name: "ElonMusk‑Persona"
      behavior: |
        Responds with bold vision, risk‑taking language,
        and a focus on first‑principles reasoning.

```

The critical distinction: nuwa‑skill sources material *about* a person to replicate their style, while cangjie‑skill sources material *by* an author to extract their methods.

---

## How They Complement Each Other

Rather than competing, these tools form a **complementary pair**:

1. **Cangjie‑skill** gives you *what to do* — concrete frameworks from authoritative content
2. **Nuwa‑skill** gives you *who to emulate* — the voice and reasoning style to deliver that guidance

A practical combination: use cangjie‑skill to extract productivity frameworks from *Deep Work*, then apply nuwa‑skill's "ElonMusk‑Persona" to have that framework explained with Musk's characteristic first‑principles boldness.

As the ecosystem documentation in [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) describes, both feed into **darwin‑skill** for iterative evolution—creating a complete pipeline from raw material to refined, deployable AI capabilities.

---

## Key Source Files Referenced

- [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) — Meta‑skill definition with explicit scope boundaries
- [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) — Ecosystem positioning (lines 154‑156)
- `methodology/00‑overview.md` — RIA‑TV++ pipeline architecture (line 27)
- [`extractors/framework-extractor.md`](https://github.com/kangarooking/cangjie-skill/blob/main/extractors/framework-extractor.md) — Example content extractor implementation

---

## Summary

- **Cangjie‑skill = "distill a book"**: transforms systematic knowledge into executable, composable skills via the 7‑stage RIA‑TV++ pipeline
- **Nuwa‑skill = "distill a person"**: captures thinking style and expression DNA for authentic role‑play
- **They are mutually exclusive in scope**—each explicitly avoids the other's domain per [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) line 16
- **They integrate synergistically**—book‑skills provide content, person‑skills provide delivery voice
- **Both evolve through darwin‑skill** as part of the broader AI‑skill ecosystem documented in [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md)

---

## Frequently Asked Questions

### Can cangjie‑skill create a persona that sounds like the book's author?

No. According to [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) (line 16), author‑persona role‑play is explicitly **not** part of cangjie‑skill's scope. That functionality belongs to nuwa‑skill. Cangjie‑skill extracts *what* the author teaches, not *how* they personally communicate.

### What does the RIA‑TV++ pipeline stand for?

The RIA‑TV++ pipeline is cangjie‑skill's 7‑stage methodology for knowledge distillation, detailed in `methodology/00‑overview.md` (line 27). "TV" refers to **Triple Verification**, a component borrowed from nuwa‑skill's verification approach. The full acronym's expansion and all seven stages are documented in that methodology file.

### Can I use nuwa‑skill to extract frameworks from a book about Warren Buffett?

No—this would invert the tools' purposes. If you want Buffett's *investment frameworks*, use **cangjie‑skill** on books like *The Essays of Warren Buffett*. If you want an AI that *responds as Buffett would*, including his folksy tone and decision‑making patterns, use **nuwa‑skill** trained on his letters, interviews, and speeches.

### How do these skills evolve after creation?

Both skill types can feed into **darwin‑skill**, the ecosystem's automatic evolution component (mentioned in [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md), lines 154‑156). Darwin‑skill iteratively refines skills based on usage feedback, improving either the accuracy of extracted frameworks (cangjie outputs) or the authenticity of persona emulation (nuwa outputs).