claude-train-prep
Claude Code plugin: AI consulting training material workflow (Word report + HTML slides) with multi-agent parallelism
Open source Open in the app JSON README (API)
About
Claude Code plugin: AI consulting training material workflow (Word report + HTML slides) with multi-agent parallelism
Details
- Kind
- Plugins
- Topic
- AI, RAG & memory
- Publisher
- gzleeon
- Origin
- marketplace
- Category
- ferramentas
- Stars
- 2
- Last push
- 2026-03-22T06:52:15Z
- Repository state
- ativo
- License
- MIT
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
gzleeon/claude-train-prep/claude-train-prep
README
# Train_Prep — Claude Code Plugin
A Claude Code plugin that encapsulates a complete **AI consulting training material preparation workflow** — from industry pain point research to delivering a Word report and interactive HTML presentation, all orchestrated with multi-agent parallelism.
## What It Does
Given a client and industry context (+ optional reference PDFs), the `Train_Prep` skill will:
1. **Research** (parallel agents) — Analyze industry pain points across 3 business phases, map them to AI solutions
2. **Word Report** — Assemble a structured `.docx` solution proposal via pandoc
3. **HTML Slides** — Generate an interactive, full-screen HTML presentation with keyboard/touch navigation
## Demo Output
The skill was originally created to prepare training materials for Pearl River Delta intercity rail executives. It produced:
- A 40-pain-point solution proposal Word report
- A 34-slide interactive HTML training presentation
## Installation
### Option 1: Install from GitHub (recommended)
```bash
# Coming soon — marketplace listing
```
### Option 2: Manual Install
```bash
# Clone the repo
git clone https://github.com/leeon/claude-train-prep.git ~/.claude/plugins/local/train-prep
# Register in ~/.claude/plugins/installed_plugins.json
# Add to "plugins" object:
# "train-prep@local": [{ "scope": "user", "installPath": "~/.claude/plugins/local/train-prep", "version": "local" }]
# Enable in ~/.claude/settings.json
# Add to "enabledPlugins": { "train-prep@local": true }
```
## Usage
Once installed, trigger the skill by saying:
- `Train_Prep`
- `给 [客户名] 准备 AI 赋能培训材料`
- `Prepare training materials for [industry] client`
- `制作培训方案，客户是 [X]，业务场景是 [Y]`
Claude will ask for:
1. Client/industry context
2. Reference PDFs (your capability deck + case studies)
3. Output directory
## Prerequisites
```bash
brew install poppler pandoc # macOS
# or
sudo apt-get install poppler-utils pandoc # Linux
```
## Workflow Architecture
```
┌─────────────────────────────────────────────────────┐
│ Train_Prep Workflow │
├─────────────────────────────────────────────────────┤
│ Phase 1: Research (3 parallel agents) │
│ Agent A ──► Phase 1 pain points + solutions │
│ Agent B ──► Phase 2 pain points + solutions │
│ Agent C ──► Phase 3 pain points + impl. plan │
├─────────────────────────────────────────────────────┤
│ Phase 2: Word Report │
│ Assemble Markdown ──► pandoc ──► .docx │
├─────────────────────────────────────────────────────┤
│ Phase 3: HTML Slides (4 parallel agents) │
│ Agent A ──► HTML framework + CSS + JS + cover │
│ Agent B ──► Phase 1 slides │
│ Agent C ──► Phase 2 slides │
│ Agent D ──► Phase 3 + impl. plan + closing │
│ Main ────► macOS-compatible assembly script │
└─────────────────────────────────────────────────────┘
```
## Plugin Structure
```
claude-train-prep/
├── .claude-plugin/
│ └── plugin.json
└── skills/
└── Train_Prep/
├── SKILL.md # Core skill (triggers + workflow)
└── references/
└── workflow-guide.md # Detailed code templates & checklist
```
## Key Technical Notes
- **PDF extraction**: Uses `pdftotext` (more reliable than Claude's built-in PDF reader for text extraction)
- **macOS compatibility**: `head -n -N` (negative line count) is not supported on macOS — the skill uses `lines=$(wc -l < file); head -n $((lines-N)) file` instead
- **HTML assembly**: Multi-agent HTML parts use a strict interface contract (Part 1 leaves `slide-container` unclosed; Part 4 closes it with `</body></html>`)
- **Dynamic slide counting**: Uses `MutationObserver` instead of hardcoded totals, enabling agents to work independently without coordination
## License
MIT License — see [LICENSE](LICENSE)
## Contributing
PRs welcome! Especially:
- Additional industry examples in `references/`
- Improved HTML slide templates
- Better ROI calculation frameworks