ClawHub

Openclaw Skill

Security

Evaluate the security of blockchain transactions (EVM, Starknet, Aptos) using the Sigui Protocol AI oracle. Detects drain attacks, mixer chains, Sybil swarms, and flash-loan exploits before they execute. Returns a verdict (ALLOW / BLOCK / ESCALATE) with a risk score and on-chain proof.

Install

openclaw skills install sigui-security

Sigui Protocol Security Skill 🛡️

Sigui is a multi-chain AI security oracle that evaluates blockchain transactions before they execute. It uses a native Spatio-Temporal Transformer (Trustformer/T-GAT) to detect coordinated attacks in real time — without the 48ms graph-rendering bottleneck of VLM-based approaches.

🧠 Chain of Thought (for OpenClaw Agent)

When a user asks you to verify or execute a blockchain transaction, follow these steps:

  1. Extract Parameters: Identify the amount (in USDC value), the destination address, the chain (default to arc), and the action type (transfer, approve, swap, etc.).
  2. Execute Evaluation: Run the evaluate.py script provided in this skill directory using the bash or python tool.
  3. Parse Verdict:
    • Exit code 0 (ALLOW): Inform the user that the transaction is safe and proceed.
    • Exit code 1 (BLOCK): DO NOT EXECUTE. Inform the user of the critical threat detected and provide the reason.
    • Exit code 2 (ESCALATE): Inform the user that the transaction requires a deep analysis. Run the script again with the --escalate flag to perform the deep analysis.
  4. Relay the Result: Format the output nicely using emojis (✅, 🚫, 🔍) and provide the on-chain proof link if available.

Supported Chains

ChainArgumentNotes
ARC (L2)arcDefault; ultra-fast, fee-free
EthereumethereumMainnet + Sepolia testnet
StarknetstarknetCairo contracts deployed on Sepolia
AptosaptosMove contracts on Testnet
SolanasolanaComing soon

🛠️ Usage Examples

Method 1 — Python Script (Recommended)

Run the helper script directly from the skill directory. The script uses the rich library to print a beautiful, easy-to-read table.

python ~/.openclaw/workspace/skills/sigui-security/evaluate.py \
  --amount 1500.0 \
  --destination "0xBadActorAddress" \
  --action "transfer" \
  --chain "ethereum"

To get raw JSON for programmatic parsing, add the --json flag:

python ~/.openclaw/workspace/skills/sigui-security/evaluate.py \
  --amount 1500.0 \
  --destination "0xBadActorAddress" \
  --json

Method 2 — Deep Analysis (Escalation)

If a transaction returns ESCALATE (or exit code 2), you can request a deep AI analysis. This uses a heavier model (like Qwen2.5 or Claude) to perform a thorough audit of the contract or address.

python ~/.openclaw/workspace/skills/sigui-security/evaluate.py \
  --amount 1500.0 \
  --destination "0xUnknownAddress" \
  --escalate

Method 3 — Inline Python (if bash is unavailable)

import asyncio, sys
sys.path.insert(0, ".")

from sigui import SiguiClient
from sigui.local import start_mock_server

async def check():
    server = start_mock_server(port=8765)
    client = SiguiClient(api_url="http://127.0.0.1:8765")
    
    result = await client.evaluate(
        agent_id="openclaw_agent",
        amount=500.0,
        destination="0xRecipientAddress",
        action_type="transfer",
        chain="arc",
    )
    server.stop()
    return result

res = asyncio.run(check())
print(f"Verdict: {res.verdict.value} | Risk: {res.risk_score}")

📊 Interpreting the Verdict

VerdictRisk ScoreAgent Action Required
ALLOW0.0 – 0.35✅ Safe to proceed. Share the safety score with the user.
ALLOW_WITH_CAP0.35 – 0.55⚠️ Safe but enforce the cap_amount_usdc spending limit.
ESCALATE0.55 – 0.80🔍 Request deep analysis (--escalate) before proceeding.
BLOCK0.80 – 1.0🚫 Abort the transaction immediately. Explain the threat.

Risk score ranges from 0.0 (completely safe) to 1.0 (critical threat). Safety score = (1 - risk_score) × 1000. Higher is better.

🛡️ Threat Types Detected

Sigui detects the following multi-chain attack patterns in real time:

  • Drain Star — One orchestrator wallet draining multiple victim wallets simultaneously.
  • Mixing Chain — Funds routed through layered mixer hops to obfuscate origin.
  • Sybil Swarm — Coordinated fake-identity cluster attacking governance or airdrops.
  • Flash Loan Exploit — Manipulating price oracles within a single atomic block.
  • Honeypot Contract — Buy-enabled, sell-disabled token contract trap.
  • Rug Pull — Liquidity removal by deployer before community exit.
  • Phishing Signature — Malicious eth_sign / permit approval requests.

⚙️ Configuration (Optional)

By default, if the Sigui SDK isn't installed or no backend is running, the script falls back to a local mock server for development and testing (no real funds required).

To connect to a live Sigui node, set the environment variables:

export SIGUI_API_URL="https://api.sigui.io"
export SIGUI_CHAIN="arc"
export OPENCLAW_AGENT_ID="my_agent_name"

🔒 Privacy & Cost

  • Evaluations on ARC are free (gasless L2).
  • Evaluations on Ethereum/Starknet/Aptos cost a micro-fee paid in USDC (~$0.001).
  • No transaction data is stored beyond the on-chain proof hash.
  • All sensitive fields (wallet keys, private data) stay local — only the metadata is sent for evaluation.

🔗 Links

More in Security