Session Memory Check Skill

Version: 1.0.0
Type: Context Enhancement Skill for AI Agents
Language: Bash (Shell Scripts)

---

🎯 What It Does

Intelligently loads memory context based on message type, balancing "memory coverage" and "context token efficiency".

Key Features:

• ✅ Session startup auto-load (prevents AI amnesia)
• ✅ Task-related context injection
• ✅ Amnesia signal detection
• ✅ Normal chat skip (saves tokens)
• ✅ Graded loading (L1/L2/L3)

---

📋 How It Works

Trigger Mechanism (3-Layer Filter)

Message Received
    ↓
Layer 1: Is it the first session message?
├─ Yes → Trigger L1 Light Load
└─ No → Layer 2

Layer 2: Message Type Detection
├─ Task Assignment → Trigger L2 Standard Load
├─ Question → Trigger L2 Standard Load
├─ Correction → Trigger L3 Full Load
└─ Normal Chat → Layer 3

Layer 3: Context Check
├─ AI Amnesia Signal → Trigger L3 Full Load
├─ User Correction → Trigger L3 Full Load
└─ Normal → No Trigger (Save Tokens)

---

📦 Load Levels

Level	Trigger	Content	Token Usage
L1 Light	Session 1st message	Identity + User + Today's Tasks	~500 chars
L2 Standard	Task/Question	L1 + Current Projects + Rules	~1000 chars
L3 Full	Correction/Amnesia	L2 + Lessons + Requirements	~2000 chars
None	Normal chat	Nothing	0 chars

---

🚀 Quick Start

1. Install

# Clone to your skills directory
git clone https://github.com/YOUR_USERNAME/session-memory-check.git
cd session-memory-check

2. Test

# Test session startup
./scripts/session-memory-trigger.sh "Hello" true

# Test task assignment
./scripts/session-memory-trigger.sh "Do XXX task" false

# Test amnesia detection
./scripts/session-memory-trigger.sh "You forgot again" false

# Test normal chat (should not trigger)
./scripts/session-memory-trigger.sh "Nice weather" false

3. Integrate

Add to your agent's system prompt or config:

# On message received
./scripts/session-memory-trigger.sh "$MESSAGE" "$IS_FIRST_MESSAGE"

# Inject output to system prompt prefix

---

📁 File Structure

session-memory-check/
├── SKILL.md                      # Full documentation
├── README.md                     # This file (quick start)
├── scripts/
│   ├── session-memory-loader.sh   # Memory loader
│   ├── session-memory-trigger.sh  # Trigger detector
│   └── run-tests.sh               # Test runner
├── tests/
│   └── test-report.md            # Test results
└── examples/
    └── integration.md            # Integration examples

---

🔧 Scripts

session-memory-trigger.sh

Purpose: Detect message type and decide trigger level

Usage:

./scripts/session-memory-trigger.sh <message> <is_first_message>

Arguments:

• message: The user message content
• is_first_message: true or false

Output:

• Message type (task/question/correction/amnesia/chat)
• Trigger level (L1/L2/L3/none)
• Memory context (if triggered)

---

session-memory-loader.sh

Purpose: Load memory context based on trigger level

Usage:

./scripts/session-memory-loader.sh <level>

Arguments:

• level: L1, L2, or L3

Output:

• Formatted memory context for injection

---

📊 Token Efficiency

Scenario	Without Skill	With Skill	Savings
Session startup	0 chars	~500 chars	-500 (necessary)
Task assignment	0 chars	~1000 chars	-1000 (necessary)
Normal chat (10 msgs)	0 chars	0 chars	✅ 0
Average	0 chars	~900 chars	✅ Efficient

Note: The skill only loads context when necessary, saving tokens on normal chats.

---

✅ Test Results

All tests passed:

Test Case	Input	Expected	Result
Session startup	"Hello" (first)	L1 trigger	✅ Pass
Task assignment	"Do XXX task"	L2 trigger	✅ Pass
Amnesia detection	"You forgot"	L3 trigger	✅ Pass
Normal chat	"Nice weather"	No trigger	✅ Pass

Test Report: tests/test-report.md

---

🎯 Success Metrics

• ✅ Session startup 100% triggers L1
• ✅ Task assignment 100% triggers L2
• ✅ Amnesia signal 100% triggers L3
• ✅ Normal chat 0% triggers
• ✅ Average context < 1000 characters
• ✅ 10 consecutive sessions without amnesia

---

🤝 Integration Examples

Example 1: OpenClaw Agent

## Session Startup Check

In AGENTS.md:
```bash
./skills/session-memory-check/scripts/session-memory-trigger.sh "$MESSAGE" true

### Example 2: Custom Agent

```python
# Python wrapper
import subprocess

def inject_memory_context(message, is_first):
    result = subprocess.run([
        './scripts/session-memory-trigger.sh',
        message,
        str(is_first).lower()
    ], capture_output=True, text=True)
    return result.stdout

Example 3: Node.js Agent

// Node.js wrapper
const { execSync } = require('child_process');

function injectMemoryContext(message, isFirst) {
    const output = execSync(
        `./scripts/session-memory-trigger.sh "${message}" ${isFirst}`
    ).toString();
    return output;
}

---

📝 Configuration

Customize Keywords

Edit scripts/session-memory-trigger.sh:

# Task assignment keywords
case "$msg" in
    *your_keyword*|*another_keyword*)
        echo "task"
        ;;
esac

Customize Load Levels

Edit scripts/session-memory-loader.sh:

# L1: Change line count
read_lines "$WORKSPACE/SOUL.md" 20  # Change 20 to your preference

# L2: Add more sections
read_memory_sections "Your Section"

---

🐛 Troubleshooting

Issue: Script not found

Solution:

chmod +x scripts/*.sh

Issue: Chinese characters not matching

Solution: Ensure your shell supports UTF-8:

export LANG=en_US.UTF-8
export LC_ALL=en_US.UTF-8

Issue: Context too long

Solution: Reduce line count in session-memory-loader.sh:

read_lines "$WORKSPACE/SOUL.md" 10  # Reduce from 20 to 10

---

📄 License

MIT License - See LICENSE file for details

---

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Push to the branch
5. Create a Pull Request

---

📬 Contact

• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

Created: 2026-04-03
Version: 1.0.0
Language: Bash (Shell)
Compatible: Linux, macOS, WSL