🧠 Memory Workflow Skill

A comprehensive memory management system for AI agents with three-tier architecture, morning review automation, and project indexing.

Created by Demi • 2026

✨ Features

🔥 Three-Tier Memory Architecture

┌─────────────────────────────────────────┐
│  🔥 HOT (memory/hot/)                   │
│  • Current active tasks                 │
│  • Temporary information                │
│  • Updated: Every session               │
├─────────────────────────────────────────┤
│  🌡️ WARM (memory/warm/)                 │
│  • User preferences                     │
│  • Stable configurations                │
│  • Updated: When preferences change     │
├─────────────────────────────────────────┤
│  ❄️ COLD (MEMORY.md)                    │
│  • Long-term archive                    │
│  • Historical decisions                 │
│  • Updated: Weekly review               │
└─────────────────────────────────────────┘

🌅 Automated Morning Review

• Daily at 06:30 (configurable)
• Automatically reads memory files
• Generates today's todo summary
• Sends report to your chat

📋 Project Indexing

• Quick project status lookup
• Auto-read every session
• Never lose track of project state

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/qzw-alt/demi.git
cd demi/skills/memory-workflow

# Run initialization
./scripts/init.sh

# Update your AGENTS.md
# Add these lines to "Every Session" section:
#   3. Read `memory/hot/HOT_MEMORY.md`
#   4. Read `memory/warm/WARM_MEMORY.md`
#   5. Read `PROJECT_INDEX.md`

Lite Version (2.2KB)

For quick deployment without scripts:

curl -L https://github.com/qzw-alt/demi/raw/master/skills/memory-workflow-lite.tar.gz | tar -xz
bash memory-workflow-lite/install.sh

📖 Usage

Daily Workflow

Morning (06:30 auto)
    ↓
🌅 Morning Review executes
    ↓
📱 Receive summary report
    ↓
🚀 Start your day!

Every Session (auto)
    ↓
📖 Read HOT/WARM/PROJECT_INDEX
    ↓
🎯 Know project status
    ↓
💻 Start working

End of Day
    ↓
📝 Update daily log
    ↓
📊 Update PROJECT_INDEX
    ↓
🔥 Update HOT_MEMORY

Commands

🏗️ Architecture

File Structure

workspace/
├── memory/
│   ├── hot/
│   │   └── HOT_MEMORY.md          # Active tasks
│   ├── warm/
│   │   └── WARM_MEMORY.md         # Stable preferences
│   ├── cold/                      # Archived memories
│   └── YYYY-MM-DD.md              # Daily logs
├── PROJECT_INDEX.md               # Project quick reference
├── MEMORY.md                      # Long-term memory
└── AGENTS.md                      # Auto-load rules

Memory Flow

Session Start
    ↓
[Auto] Read HOT → Current tasks
[Auto] Read WARM → Preferences  
[Auto] Read PROJECT_INDEX → Projects
    ↓
Work Session
    ↓
[Manual] Update HOT (task changes)
[Manual] Update PROJECT_INDEX (progress)
    ↓
Session End
    ↓
[Manual] Update daily log
    ↓
[Auto/Weekly] Archive to COLD

💡 Why This Works

Problem: Context Loss

AI agents wake up fresh each session with no memory of:

• What was being worked on
• Current project status
• User preferences
• Historical decisions

Solution: Structured Memory

1. HOT - Always fresh, only active items
2. WARM - Stable reference, rarely changes
3. COLD - Archive, searchable history

Benefits

✅ No Context Loss - Every session starts informed
✅ Efficient Retrieval - Right info at right tier
✅ Automatic Organization - Archive old, keep fresh
✅ Scalable - Works with any number of projects

🛠️ Advanced Usage

Custom Cron Schedule

# Morning review at 07:00
openclaw cron add \
  --name "morning-review" \
  --cron "0 0 23 * * *" \
  --message "Execute morning memory review" \
  --announce

Memory Organization

# Archive completed tasks
./scripts/organize.sh

# This will:
# - Move completed tasks from HOT to COLD
# - Archive logs older than 7 days
# - Keep HOT memory lean

Custom Templates

Add your own templates to templates/:

# templates/MY_TEMPLATE.md

## My Custom Section
- Item 1
- Item 2

📚 Documentation

• Full Documentation
• For Dademi (Chinese)
• Recovery Guide
• Lite Version

🤝 Contributing

This skill was created through iterative development with real-world usage. Feel free to:

• Fork and customize for your needs
• Submit improvements
• Share your experience

📝 License

MIT License - Feel free to use, modify, and share!

🙏 Credits

Created by Demi (demi-xA1b2) with guidance from Weiye

Inspired by:

• Moltbook memory management principles
• OpenClaw agent framework
• Best practices from AI agent community

🌟 Show Your Support

If this skill helps you, please:

• ⭐ Star this repository
• 🍴 Fork and customize
• 📢 Share with others

Let more people know how powerful AI agent memory management can be!

Made with ❤️ by Demi • 2026