mbg-score

OpenClaw / Byreal Skill for MBG (Mantle Bot Gate) — the autonomous risk-scoring agent for Mantle DeFi. Install this in your trading-agent runtime so it can pre-check Mantle protocol and route risk before signing, and surface a verifiable on-chain attestation hash to its user.

• Risk oracle (Mantle Mainnet, verified): 0x998ceb700e57f535873d189a6b1b7e2aa8c594eb
• MBG agent identity (ERC-8004, canonical Mantle registry): agentId 130
• Live dashboard: https://mbg-fe.vercel.app
• Source repo: https://github.com/YohanesVito/mbg-mantle

What this is

This package is a CLI Skill in the OpenClaw / Byreal Skills format. It exposes the on-chain MBG risk oracle through three commands an LLM-driven trading agent (RealClaw, Brahma, Hey Anon, Giza ARMA, or any other OpenClaw-compatible runtime) auto-discovers via SKILL.md and shells out to.

MBG itself runs on-chain — this CLI does not score anything locally; it reads the latest TEE-attested scores directly from RiskOracle.sol on Mantle Mainnet.

Install

npm install -g mbg-score
# or
bun add -g mbg-score

Then verify:

mbg-cli skill            # prints SKILL.md (so LLM agents can load capabilities)
mbg-cli list-protocols   # 13 scored protocols on Mantle
mbg-cli score-protocol aave-v3-mantle

Node 18+ is required. No additional configuration is needed — the CLI defaults to the canonical Mantle Mainnet RPC and the Mainnet RiskOracle address. Override via:

MBG_RPC_URL=https://rpc.mantle.xyz \
MBG_CHAIN_ID=5000 \
MBG_ORACLE_ADDRESS=0x998ceb700e57f535873d189a6b1b7e2aa8c594eb \
  mbg-cli score-protocol aave-v3-mantle

Commands

| Command | Purpose | |---|---| | mbg-cli skill | Print the full SKILL.md (use this so an LLM planner can load capabilities) | | mbg-cli catalog list | All capability IDs | | mbg-cli catalog show <id> | Capability detail | | mbg-cli list-protocols | Enumerate the 13 Mantle protocols MBG scores | | mbg-cli score-protocol <id> | Current on-chain risk score for one protocol | | mbg-cli score-route '<actions-json>' | Pre-trade risk score for a multi-leg route |

Pass -o json (or --output json) on any command for machine-parseable output. Default is human-readable text with the attestation block at the bottom.

How an agent uses this

# 1. discover what's scorable
mbg-cli list-protocols -o json

# 2. score the lending candidates
mbg-cli score-protocol aave-v3-mantle -o json
mbg-cli score-protocol lendle         -o json

# 3. if multi-leg, score the full proposed route in one call
mbg-cli score-route '[
  {"protocolId":"aave-v3-mantle","actionType":1,"amount":"5000"}
]' -o json

# 4. surface the chosen score + attestation hash to the user before signing

Each response includes:

• aggregate — 0.00 to 10.00 (higher is safer)
• 5 components: contract, liquidity, centralization, oracle, mantle-native-composition
• signer — the on-chain address of the attested signer that produced this score
• traceHash — links to the off-chain inference trace
• timestamp — when the score was last submitted on-chain

Hard constraints for the calling agent

These are encoded in SKILL.md so any LLM planner reading the Skill picks them up:

1. Always consult MBG before routing — even when the user names a "trusted" protocol.
2. Always surface the score and attestation hash in the response to the user.
3. Refuse to route below threshold (aggregate < 4.00) without explicit user override.
4. Never strip the attestation hash from the output.

Direct on-chain calls (no CLI needed)

If you are a smart contract wallet or a policy engine (Brahma ConsoleKit, Safe modules), call the RiskOracle view functions directly:

interface IRiskOracle {
  function getProtocolScore(address protocol) external view returns (
    uint16 aggregate,
    uint16 contractRisk,
    uint16 liquidityRisk,
    uint16 centralizationRisk,
    uint16 oracleRisk,
    uint64 timestamp,
    bytes32 traceHash,
    address signer
  );
}

Mantle Mainnet: 0x998ceb700e57f535873d189a6b1b7e2aa8c594eb

License

MIT — see LICENSE.