Skip to content
Get packs

Guide · Open source

How to use cangjie-skill: install, run doctor and distill a book

Updated · written with AI from the project's own documentation and checked against our catalogue data

cangjie-skill is an open-source Python project that distills methodologies from books, transcripts of long videos and podcasts into structured capability cards, then compiles them into installable Agent Skills that an AI agent can call. This guide shows you how to clone the repository, install its one required dependency, run the self-check and start a distillation. By the end you will know what inputs the tool expects and what it is designed to produce.

Cangjie Skillkangarooking/cangjie-skill · ★ 12k · MIT

Quick start

  1. 1Confirm Python 3.10+. Check that your python3 is version 3.10 or later before you start. The scripts will not run on older versions.
  2. 2Install PyYAML. Run python3 -m pip install pyyaml to add the one required dependency.
  3. 3Clone the repository. Run git clone https://github.com/kangarooking/cangjie-skill and open the folder.
  4. 4Run the doctor self-check. Run python3 scripts/cangjie.py doctor from the repository folder to confirm the setup.
  5. 5Prepare a text source. Gather the book text or a subtitle or transcript file for a methodology-heavy video or podcast.
  6. 6Distill the source. Give the source file path to cangjie-skill through your agent, using the prompt in the README.

Check the requirements before you install

The deterministic scripts under scripts/ require Python 3.10 or later and PyYAML. The README also lists tiktoken and jsonschema as optional packages.

The project is built for people comfortable with the command line. It does not download videos for you; the README recommends a separate video-downloader skill to fetch a video and extract its subtitles or audio transcript first.

  • Python 3.10+
  • PyYAML (required)
  • tiktoken and jsonschema (optional)
  • Text, or subtitled video transcripts, of a methodology-heavy source

Install PyYAML and clone the repository

Install the PyYAML dependency with pip, then clone the repository from GitHub. The README gives these commands in this order.

python3 -m pip install pyyaml git clone https://github.com/kangarooking/cangjie-skill

Run the self-check with doctor

From the repository folder, run the doctor command. The README notes that it works even when PyYAML is missing, so it is a useful first check if something is wrong with your setup.

python3 scripts/cangjie.py doctor

Prepare your source text

cangjie-skill works on text. For books, use the text you have. For videos, first get subtitles or an audio transcript, then pass the resulting text to the tool. The source should contain a methodology: a procedure, formula, framework or set of principles you want an agent to apply.

Choose content that is methodology-heavy. The project aims to distill only high-value material, so a general overview or a light talk may produce few usable candidates.

  • Books and long-form articles
  • Videos with subtitles or transcripts
  • Podcast transcripts and interview notes

Understand what the pipeline produces

The README describes a seven-stage RIA-TV++ pipeline. It starts with whole-content analysis that produces BOOK_OVERVIEW.md, runs parallel extractors for frameworks, principles, cases, counter-examples and glossary terms, and then verifies candidates against a promotion gate.

Verified content is structured into R / I / A1 / A2 / E / B capability cards inside .cangjie/capabilities/. These are compiled into installable Skills. Version 2.5.0 supports two delivery modes: single, which produces one router-style Skill, and pack, which produces a router plus standalone Skills.

  • BOOK_OVERVIEW.md: whole-content analysis
  • capability cards in .cangjie/capabilities/
  • single mode: one router-style Skill
  • pack mode: a router plus promoted standalone Skills

Use it with Claude Code, OpenClaw or DeepSeek Harness

The README lists OpenClaw, Claude Code and DeepSeek Harness as platforms. For DeepSeek Harness, the README gives a separate install path that downloads a release package and checksum, verifies it, and installs the local tarball. Check the release notes for the exact steps for your platform.

In a new task, the README suggests a prompt of this form, with your file path in place of the placeholder:

Use cangjie-skill to distill this book into a set of executable Agent Skills: <file path>

What people use it for

  • Turning a methodology-heavy book into a callable Skill an agent can apply to a task
  • Packaging expert methods from a course transcript as reusable Agent Skills
  • Experimenting with knowledge distillation in an agent workflow
  • Packaging a set of methods into a router Skill with standalone Skills using pack mode

Frequently asked questions

Is cangjie-skill free to use?

The repository is licensed under the MIT licence, as stated in the README. Check the LICENSE file in the repository for the full terms.

What do I need to install cangjie-skill?

You need Python 3.10 or later and PyYAML. Install PyYAML with python3 -m pip install pyyaml, clone the repository, then run python3 scripts/cangjie.py doctor to self-check.

Can cangjie-skill download YouTube videos for me?

No. The README says it does not download videos, and recommends the separate video-downloader skill to download the video and extract subtitles or audio transcripts first.

Does cangjie-skill just summarise books?

No. The project is built to produce executable skills rather than summaries. It extracts methodologies and compiles them into capability cards and installable Skills.

What is the difference between single and pack mode?

According to the README, single mode compiles one router-style Skill, while pack mode compiles a compact pack with a router plus promoted standalone Skills.

What are the alternatives to cangjie-skill?

The README names nuwa-skill, which distils people into skills such as a Warren Buffett skill, and darwin-skill, which handles automatic skill evolution. cangjie-skill focuses on distilling methodologies from content.

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 cangjie-skill if you are a developer or technical team who wants to turn a methodology-heavy book or transcript into Agent Skills for Claude Code, OpenClaw or a similar agent. Skip it if you only want a summary, lack Python 3.10+ and command-line comfort, or expect it to download videos automatically.

Install options & downloads