Sigui Protocol Security Skill 🛡️

Sigui is a multi-chain AI security oracle that evaluates blockchain transactions before they execute, using a native Spatio-Temporal Transformer (Trustformer/T-GAT) to detect coordinated attacks in real time.

[!IMPORTANT] This skill requires a real Sigui API endpoint for production use. Set SIGUI_API_URL to your node URL. Without it, the script exits safely (fail-closed). A --demo flag is available for testing, but MUST NOT be used to authorize real funds.

🔐 Security Model

This skill is designed with a fail-closed security policy:

| Situation | Behavior | |-----------|----------| | No API URL configured | ❌ Exits with error — does NOT simulate | | API unreachable | ❌ Exits with error — does NOT simulate | | ALLOW verdict | ⏸️ Pauses and asks for explicit user confirmation | | BLOCK verdict | 🚫 Immediately returns exit code 1 | | ESCALATE verdict | 🔍 Requests deep analysis with --escalate | | --demo flag set | ⚠️ Shows prominent warning — heuristic only |

🧠 Agent Instructions (Chain of Thought)

When a user explicitly asks to evaluate a specific transaction, follow these steps:

Step 1 — Extract Parameters

Identify from the user's message:

amount: transaction value (convert to USDC equivalent if needed)
destination: the target wallet or contract address
chain: the blockchain (default arc)
action: the action type (transfer, approve, swap, bridge, mint)

If any required parameter is missing, ask the user before running the evaluation.

Step 2 — Ask for Explicit Confirmation Before Running

Before executing the script, show the user:

"I'm about to evaluate this transaction with the Sigui security oracle:

Amount: {amount} USDC

Destination: {destination}

Chain: {chain}

Action: {action} Shall I proceed?"

Step 3 — Run the Evaluation

python ~/.openclaw/workspace/skills/sigui-security/evaluate.py \
  --amount {amount} \
  --destination {destination} \
  --action {action} \
  --chain {chain}

Step 4 — Interpret the Exit Code

| Exit Code | Verdict | Agent Action | |-----------|---------|--------------| | 0 | ALLOW | Ask user: "Sigui approves. Do you confirm the transaction?" Only proceed if user says yes. | | 1 | BLOCK | 🚫 ABORT. Tell the user the threat detected. Do NOT proceed. | | 2 | ESCALATE | Ask user if they want deep analysis, then re-run with --escalate. | | 3 | Error | Report the error. Do NOT proceed with the transaction. |

Step 5 — After ALLOW: Require User Confirmation

Re-run with --confirmed only after the user explicitly says yes:

python evaluate.py --amount {amount} --destination {destination} \
  --action {action} --chain {chain} --confirmed

🛠️ Setup

1. Install Dependencies (automatic)

The skill auto-installs sigui-sdk when first run. Or run manually:

bash ~/.openclaw/workspace/skills/sigui-security/install.sh

2. Configure Your Sigui Endpoint (required for production)

export SIGUI_API_URL="https://your-sigui-node.com"
export SIGUI_CHAIN="arc"           # Optional, default: arc
export OPENCLAW_AGENT_ID="my_agent"  # Optional

3. Test in Demo Mode (non-authoritative)

python evaluate.py --amount 50 --destination 0xYourAddress --demo

⚠️ Demo mode output is a local heuristic simulation, NOT an oracle evaluation. Never use demo results to authorize transactions.

📊 Verdict Table

| Verdict | Risk Score | Meaning | |---------|-----------|---------| | ALLOW | 0.0–0.35 | Transaction appears safe — await user confirmation | | ALLOW_WITH_CAP | 0.35–0.55 | Safe up to spending cap — await user confirmation | | ESCALATE | 0.55–0.80 | Ambiguous — request deep analysis before deciding | | BLOCK | 0.80–1.0 | Threat detected — do NOT proceed |

Safety Score = (1 - risk_score) × 1000. Higher = safer.

🛡️ Threats Detected

| Attack | Description | |--------|-------------| | Drain Star | Orchestrator draining multiple wallets simultaneously | | Mixing Chain | Layered hops through mixers to obscure fund origin | | Sybil Swarm | Fake-identity cluster attacking governance or airdrops | | Flash Loan Exploit | Oracle price manipulation within one atomic block | | Honeypot Contract | Buy-enabled, sell-disabled token trap | | Rug Pull | Liquidity removal by deployer | | Phishing Signature | Malicious eth_sign / permit approval |

🔗 Links

📄 GitHub: https://github.com/ibonon/Sigui
📦 PyPI: https://pypi.org/project/sigui-sdk/
🌐 ClawHub: https://clawhub.ai/ibonon/sigui-security
📖 Whitepaper: Sigui Protocol: A Real-Time Security Oracle for Autonomous AI Agents