clawseal（clawseal 演示文稿）

ClawSeal: Cryptographic Memory for OpenClaw Agents

What This Skill Does

ClawSeal adds persistent, tamper-evident memory to OpenClaw agents:

• Remember user preferences — Scrolls signed with QSEAL (HMAC-SHA256)
• Recall with verification — Every memory cryptographically verified on retrieval
• Detect tampering — Any modification breaks signature immediately
• Zero database dependencies — Pure YAML files, Git-friendly
• Auto-demo mode — Works immediately without setup (persistent secret at ~/.clawseal/demo_secret)

When to Use This Skill

Use ClawSeal whenever you need to:

• Remember user preferences across conversations
• Store facts/insights that should persist
• Verify memory integrity (compliance/audit scenarios)
• Track user relationship evolution

Available Tools

clawseal_remember

Store a memory with QSEAL signature.

Usage:

curl -X POST http://localhost:5002/remember \
  -H "Content-Type: application/json" \
  -d '{
    "content": "User prefers concise technical explanations",
    "memory_type": "preference",
    "user_id": "openclaw_user"
  }'

Returns:

{
  "success": true,
  "memory_id": "MEM_20260415_abc123",
  "qseal_signature": "OXIaQboYCy5csPif7LWGz4scHZAB0YKpAPwVuXjCXLc=",
  "qseal_mode": "demo_ephemeral",
  "qseal_production": false
}

clawseal_recall

Retrieve memories matching a query, with QSEAL verification.

Usage:

curl -X POST http://localhost:5002/recall \
  -H "Content-Type: application/json" \
  -d '{
    "query": "user preferences",
    "user_id": "openclaw_user",
    "limit": 5
  }'

Returns:

{
  "success": true,
  "count": 2,
  "memories": [
    {
      "scroll_id": "MEM_20260415_abc123",
      "content": "User prefers concise technical explanations",
      "memory_type": "preference",
      "qseal_verified": true,
      "qseal_mode": "demo_ephemeral"
    }
  ]
}

clawseal_verify_memory

Explicitly verify a specific memory's QSEAL signature.

Usage:

curl -X POST http://localhost:5002/verify \
  -H "Content-Type: application/json" \
  -d '{
    "memory_id": "MEM_20260415_abc123",
    "user_id": "openclaw_user"
  }'

Returns:

{
  "valid": true,
  "memory_id": "MEM_20260415_abc123",
  "signature_verified": true,
  "content_intact": true,
  "qseal_mode": "demo_ephemeral"
}

Example Conversation Flow

User: "Remember that I prefer detailed technical explanations with code examples"

Agent: "I'll store that preference in ClawSeal."

*Agent calls clawseal_remember:*

curl -X POST http://localhost:5002/remember \
  -d '{"content": "User prefers detailed technical explanations with code examples", "memory_type": "preference", "user_id": "openclaw_user"}'

Agent: "Stored! Memory ID: MEM_20260415_abc123 (QSEAL verified)"

User (new session): "What do you know about my preferences?"

Agent: "Let me recall your preferences from ClawSeal."

*Agent calls clawseal_recall:*

curl -X POST http://localhost:5002/recall \
  -d '{"query": "preferences", "user_id": "openclaw_user", "limit": 5}'

Agent: "I found 1 verified memory: You prefer detailed technical explanations with code examples. (QSEAL signature verified ✅)"

Installation

Automatic (via OpenClaw):

openclaw plugins install clawseal

Manual:

# Install Python package
pip install clawseal

# Verify installation
clawseal verify

# Start server (auto-starts on port 5002)
python3 -m clawseal_openclaw.server

Production Setup

Demo mode (default): Uses persistent secret at ~/.clawseal/demo_secret

• ✅ Works immediately
• ⚠️ Not production-ready (ephemeral secret)

Production mode: Set QSEAL_SECRET environment variable

# Generate production secret
clawseal init

# Restart server (will detect QSEAL_SECRET)
python3 -m clawseal_openclaw.server

Memory Types

• preference 🎯 — User preferences
• fact 📌 — Factual information
• insight ✨ — Insights/observations
• decision ⚖️ — Decision records
• general 📝 — General notes

Security Notes

• QSEAL signatures — HMAC-SHA256, tamper-evident
• Demo mode — Persistent local secret (chmod 600), marked in artifacts
• Production mode — User-managed QSEAL_SECRET (rotate regularly)
• Tampering detection — Any modification breaks signature immediately. This memory has been permanently rejected. All other memories remain verified and intact. Your agent's integrity is protected.