# How cangjie-skill Processes Video and Podcast Transcripts Compared to Books: The Complete Workflow

> Learn how Cangjie-skill processes video and podcast transcripts against books. Discover the unified workflow and required media conversion for audio and written content.

- Repository: [kangarooking/cangjie-skill](https://github.com/kangarooking/cangjie-skill)
- Tags: deep-dive
- Published: 2026-07-19

---

**cangjie-skill applies the identical extraction pipeline to all content types, but requires video and podcast media to be converted to text via the video-downloader skill before processing, whereas books can be ingested directly from PDF, EPUB, or TXT files.**

The kangarooking/cangjie-skill repository distills long-form content into reusable AI skills. While the core methodology remains consistent across mediums, the ingestion path differs significantly between static text documents and audio-visual media.

## The Unified Knowledge Extraction Pipeline

Once raw text is available, cangjie-skill processes every content type through the same pipeline. The system extracts **principles**, **frameworks**, **cases**, and other knowledge artifacts, validates them against source material, and renders them into structured skill templates.

According to [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) line 44, acceptable content sources include *PDF / EPUB / TXT / 字幕文件 / 转写稿路径* (subtitle files and transcription paths). The templates `templates/BOOK_OVERVIEW.md.template` and `templates/DIGEST.md.template` format the final output regardless of whether the source was a printed manual or a podcast transcript.

This unified approach ensures that skills generated from video transcripts maintain the same structural integrity as those derived from technical books.

## Input Requirements: Direct Ingestion vs. Transcription Prerequisites

The critical distinction lies in the preparation phase. Books require no preprocessing, while video and podcast content demands an intermediate transcription step.

### Processing Books Directly

For books, cangjie-skill accepts native document formats immediately. Point the tool directly at the source file:

- **PDF** - Academic papers and scanned manuals
- **EPUB** - Digital publications and e-books  
- **TXT** - Plain text files and markdown documents

As documented in [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md), the system reads these formats natively without external dependencies.

### Handling Video and Podcast Content

Video and podcast files require conversion to text before ingestion. The repository explicitly recommends pairing cangjie-skill with the **video-downloader** skill for this preprocessing.

Per [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) line 26: *"如果要蒸馏视频内容，建议搭配 video‑downloader skill 一起使用：先用它下载视频、提取字幕/音频转写和关键素材，再把得到的文本内容交给 cangjie‑skill."*

This workflow ensures you have accessible plain text before the distillation process begins, preventing the system from attempting to process binary media files directly.

## Practical Implementation Workflows

Below are the complete command patterns for each media type.

### Workflow 1: Processing a Book

Feed PDF, EPUB, or TXT files directly into cangjie-skill:

```bash
cangjie-skill --input path/to/book.pdf --output ./skills/book

```

The tool parses the document structure and initiates knowledge extraction immediately.

### Workflow 2: Processing Video or Podcast Content

Execute this two-stage pipeline:

```bash

# Stage 1: Extract transcript using the video-downloader skill

video-downloader --url https://example.com/talk.mp4 --transcript ./tmp/transcript.txt

# Stage 2: Process the resulting text file through cangjie-skill

cangjie-skill --input ./tmp/transcript.txt --output ./skills/video

```

Both workflows converge on the same validation logic and output templates, producing comparable skill definitions and skill-graphs.

## Key Source Files and Configuration

The following files govern how cangjie-skill handles different input formats:

| File | Purpose | Key Details |
|------|---------|-------------|
| [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) | Usage guidelines and workflow recommendations | Line 26 recommends video-downloader pairing for video content |
| [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) | Formal input specification | Line 44 lists acceptable formats including subtitle files and transcripts |
| `templates/BOOK_OVERVIEW.md.template` | Knowledge structure template | Applied to both book and transcript-derived content |
| `templates/DIGEST.md.template` | Final output formatting | Unified rendering for all skill types |

## Summary

- **Unified processing**: cangjie-skill uses identical extraction, validation, and templating logic for books and transcripts once text is available.
- **Preprocessing requirement**: Video and podcast content requires transcription via video-downloader before ingestion, while books support direct PDF/EPUB/TXT input.
- **Input flexibility**: The system accepts subtitle files and raw transcripts as valid text sources per [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) specifications.
- **Consistent output**: Both workflows utilize `templates/DIGEST.md.template` to ensure standardized skill generation regardless of original media format.

## Frequently Asked Questions

### Does cangjie-skill support direct video file uploads?

No. According to the source documentation in [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md) line 44, cangjie-skill processes text only. Video and audio files must first be converted to subtitles or transcription text using the video-downloader skill or similar tools before ingestion.

### What transcription formats does cangjie-skill accept?

The system accepts standard subtitle files (such as SRT or VTT) and plain text transcription files. As noted in [`SKILL.md`](https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md), these fall under acceptable *字幕文件* and *转写稿路径* inputs, which the pipeline treats identically to TXT documents.

### Is the skill quality different between books and transcripts?

No. Once converted to text, both sources undergo the same validation and extraction logic. The `templates/BOOK_OVERVIEW.md.template` applies uniformly to content from either medium, ensuring consistent principle extraction and framework identification.

### Where is the video-downloader skill documented?

The video-downloader skill is referenced in [`README.md`](https://github.com/kangarooking/cangjie-skill/blob/main/README.md) line 26 as the recommended companion tool for video processing. While it exists as a separate repository, the cangjie-skill documentation assumes its use for obtaining clean transcript files from multimedia sources.