/live-workshop

Turn any lecture into a structured, searchable knowledge base -- in real time.

A Claude Code custom skill that sets up a JupyterBook-based workshop journal with live audio transcription, keyword extraction, and structured knowledge base.

Quick Install

# 1. Clone into Claude Code skills directory
git clone https://github.com/nasqret/live-workshop-skill.git ~/.claude/skills/live-workshop

# 2. Install dependencies
brew install sox ffmpeg          # macOS (Linux: apt install sox ffmpeg)
pip3 install openai-whisper jupyter-book

# 3. Use in any project
cd ~/my-project
# Then in Claude Code:
/live-workshop "Workshop Title" --lang en --modules "Topic1,Topic2,Topic3"

What It Does

Four-Phase Pipeline

Phase	Action
INIT	Creates JupyterBook project with modules, journal, exercises, and git repo
RECORD	Live audio capture (sox) + real-time Whisper transcription + keyword prompter
ORGANIZE	Auto-classifies your notes into theory/tools/exercises, updates journal, commits
ANALYZE	Iterative transcript analysis, decodes Whisper errors, extracts speaker insights

Project Structure Generated

your-project/
├── _config.yml              # JupyterBook config
├── _toc.yml                 # Auto-updated table of contents
├── intro.md                 # Landing page
├── journal/                 # Daily meeting notes
├── modules/
│   ├── 01_topic/
│   │   ├── theory.md        # Concepts and frameworks
│   │   ├── tools.md         # Software and commands
│   │   └── exercises.md     # Hands-on tasks
│   └── ...
├── exercises/index.md
└── live_audio/
    ├── start.sh             # Launch recording + transcription
    ├── record_chunks.sh     # sox audio recorder
    └── transcribe_live.py   # Whisper + keyword extraction

Usage

Start a workshop session

/live-workshop "Machine Learning Fundamentals" --lang en --modules "Regression,Neural Networks,Transformers"

With a workshop URL (auto-fetches agenda)

/live-workshop "DevOps Workshop" --url https://workshop.example.com/agenda

Start live recording (in a separate terminal)

./live_audio/start.sh              # defaults: small model, 30s chunks, English
./live_audio/start.sh turbo 45     # better quality, 45s chunks
./live_audio/start.sh tiny 15 de   # fast, 15s chunks, German

During the workshop

Just type your notes in Claude Code. The skill will:

• Classify them into the right module (theory/tools/exercises)
• Update today's journal entry
• Rebuild the JupyterBook
• Commit to git

Analyze transcripts

Say "iterate notes" or "analyze transcripts" to process new audio fragments.

Arguments

Argument	Default	Description
<title>	required	Workshop/meeting title
--lang	en	Whisper transcription language
--modules	auto-detect	Comma-separated module names
--url	-	Workshop URL to fetch agenda from
--chunk-duration	30	Audio chunk length in seconds
--whisper-model	small	Whisper model (tiny/base/small/medium/turbo)

Requirements

• Claude Code CLI
• sox - audio recording
• OpenAI Whisper - transcription
• ffmpeg - audio processing
• JupyterBook - output format
• Python 3.10+
• Git

Tech Stack

sox | Whisper | JupyterBook | Git | Python | zsh

License

MIT