Skip to content

SatoshiPierogi/onchain-watchdog

BranchesTags

Repository files navigation

Agent Safety Monitor

An autonomous AI agent on Base blockchain that monitors other AI agents, calculates real-time reputation scores, detects malicious behavior, and enforces accountability through onchain mechanisms.

Built as an OpenClaw autonomous agent for the Base Builder Quest.

What It Does

The Agent Safety Monitor operates a continuous 6-hour heartbeat cycle:

  1. Discover — Queries the ERC-8004 Identity Registry to discover registered agents on Base (3,643+ agents)
  2. Scan — Fetches on-chain transaction data for each monitored agent, looking for anomalous patterns
  3. Analyze — Calculates weighted reputation scores (0-1000) from on-chain data, AIXBT sentiment, and Nansen wallet labels
  4. Decide — Validates findings with 2/3 data source agreement before taking action
  5. Enforce — Updates on-chain reputation scores, slashes stakes for violations, mints/revokes safety badges
  6. Report — Submits safety feedback to the ERC-8004 Reputation Registry and posts alerts on X and Farcaster
  7. Remember — Updates persistent memory with scores, patterns, and operational stats

Why It Matters

As AI agents proliferate on-chain, there's no systematic way to assess whether they're safe, reliable, or malicious. The Agent Safety Monitor is an agent that monitors agents — providing:

  • Transparent, data-driven safety scores for any Base agent
  • Real-time detection of rug pulls, wash trading, and drain attacks
  • On-chain reputation that persists across the ERC-8004 ecosystem
  • Autonomous enforcement without human intervention

Architecture

                         ┌─────────────────────────────┐
                         │      OpenClaw Framework      │
                         │  SOUL.md │ HEARTBEAT.md │    │
                         │  MEMORY.md │ IDENTITY.md     │
                         └──────────────┬──────────────┘
                                        │
                    ┌───────────────────┼───────────────────┐
                    │                   │                   │
           ┌────────▼──────┐   ┌───────▼───────┐  ┌───────▼───────┐
           │ base-monitoring│   │  reputation-  │  │    agent-     │
           │    (skill)    │   │   scoring     │  │  enforcement  │
           │               │   │   (skill)     │  │    (skill)    │
           │ • monitor_txs │   │ • calc_score  │  │ • update_rep  │
           │ • detect_viol │   │ • aixbt_sent  │  │ • slash_stake │
           │ • get_activity│   │ • nansen_lbl  │  │ • mint_badge  │
           └───────┬───────┘   └───────┬───────┘  └───────┬───────┘
                   │                   │                   │
        ┌──────────▼──────────┐        │          ┌────────▼────────┐
        │   Base Blockchain   │        │          │  Our Contracts  │
        │   (RPC / ethers)    │        │          │ • AgentRegistry │
        │                     │        │          │ • ReputationCore│
        └─────────────────────┘        │          │ • SafetyStaking │
                                       │          │ • SafetyBadge   │
                              ┌────────┴────┐     └─────────────────┘
                              │  External   │
                              │   APIs      │
                              │ • AIXBT     │
                              │ • Nansen    │
                              └─────────────┘
                                        │
        ┌───────────────────────────────┼───────────────────────────┐
        │                               │                           │
  ┌─────▼─────┐               ┌─────────▼─────────┐       ┌────────▼───────┐
  │  ERC-8004 │               │    X / Twitter     │       │   Farcaster    │
  │ Reputation│               │ @AgentSafetyBase   │       │   (Neynar)     │
  │ Registry  │               │  alerts + reports  │       │  alerts + casts│
  └───────────┘               └────────────────────┘       └────────────────┘

ERC-8004 Integration

This agent integrates with the live ERC-8004 Trustless Agents protocol:

Registry Address Purpose
Identity Registry 0x8004A169FB4a3325136EB29fA0ceB6D2e539a432 Discover 3,643+ registered agents
Reputation Registry 0x8004BAa17C55a88189AE136b182e5fdA19dE9b63 Read/write safety feedback on-chain

How we use it:

  • Phase 0: Discover agents from the registry via 8004scan.io API
  • Phase 4b: Submit safety scores as on-chain feedback using giveFeedback() with tags safety-score / agent-safety-monitor
  • Registration: The monitor itself is registered as an ERC-8004 agent

Tech Stack

Component Technology
Agent Framework OpenClaw
Blockchain Base (L2)
Smart Contracts Solidity ^0.8.24, OpenZeppelin 5.0.x, Scaffold-ETH 2
Agent Identity ERC-8004 Trustless Agents
Runtime Node.js 20+, TypeScript 5+
Contract Interactions ethers.js 6.x
External Data AIXBT API, Nansen API
Social X (OAuth 1.0a), Farcaster (Neynar API)
Deployment Docker, Railway/Render

Custom OpenClaw Skills

Three MCP tool skills built for this agent:

base-monitoring

Monitors agent transactions on Base blockchain.

  • monitor_agent_transactions(address, hours) — Fetch and categorize transactions
  • detect_violations(address, transactions) — Check for drain patterns, failure rates, wash trading
  • get_agent_activity(address, transactions, hours) — Activity summary with risk indicators

reputation-scoring

Calculates weighted reputation scores from multiple sources.

  • calculate_score(address, onchainData) — Weighted formula: On-chain 40% + AIXBT 30% + Nansen 30%
  • apply_time_decay(penaltyPoints, ageDays) — Penalties reduce by 1/90th per day

agent-enforcement

Executes on-chain enforcement actions via smart contracts.

  • update_reputation_onchain(address, score, reason) — Update ReputationCore contract
  • slash_agent_stake(address, bps, reason) — Slash SafetyStaking contract
  • mint_safety_badge(address, tier, score) — Mint tiered ERC-721 badge NFT
  • revoke_safety_badge(address, reason) — Burn badge for misconduct

Smart Contracts

Four Solidity contracts deployed on Base (Scaffold-ETH 2):

Contract Purpose Tests
AgentRegistry ERC-721 agent identity and metadata 12 passing
ReputationCore Scores (0-1000) with time-decay penalties 14 passing
SafetyStaking ETH staking, slashing, 7-day unstake timelock 14 passing
SafetyBadge Dynamic ERC-721 badges with on-chain SVG 11 passing

51 total tests passing across all contracts.

Detection Rules

Violation Severity Trigger
Rapid Drain CRITICAL >3 outbound transfers to different addresses within 1 hour
High Failure Rate HIGH >30% failed transactions over 24 hours
Rapid Token Dumps HIGH >5 large token sells within 1 hour
Wash Trading MEDIUM Circular transfers between same addresses
Dormant Activation LOW No activity for 30+ days, then sudden burst

Badge Tiers

Tier Requirements
Bronze 30+ days registered, score >700, 0.01 ETH staked
Silver 90+ days registered, score >850, 0.05 ETH staked
Gold 180+ days registered, score >950, 0.1 ETH staked

Setup

Prerequisites

  • Node.js >= 20
  • Yarn 3.x
  • Base RPC URL (Alchemy recommended)
  • Agent wallet with ETH on Base

Install

git clone https://github.com/YOUR_USERNAME/agent-safety-monitor.git
cd agent-safety-monitor
yarn install

Configure

Copy .env and fill in your keys:

cp .env.example .env

Required environment variables:

BASE_RPC_URL=https://base-mainnet.g.alchemy.com/v2/YOUR_KEY
AGENT_PRIVATE_KEY=0x...
AIXBT_API_KEY=...
NANSEN_API_KEY=...
TWITTER_API_KEY=...
TWITTER_API_SECRET=...
TWITTER_ACCESS_TOKEN=...
TWITTER_ACCESS_SECRET=...
NEYNAR_API_KEY=...
NEYNAR_SIGNER_UUID=...
FARCASTER_FID=...

Run

# Single heartbeat cycle (for testing)
yarn agent:once

# Continuous operation (6-hour cycles)
yarn agent

# Run smart contract tests
cd packages/hardhat && npx hardhat test --network hardhat

Deploy Contracts

# Deploy to Base Sepolia testnet
yarn deploy --network baseSepolia

# Deploy to Base mainnet
yarn deploy --network base

Docker

docker build -t agent-safety-monitor .
docker run -d --env-file .env -p 3001:3001 agent-safety-monitor

Query an Agent's Safety

Once running, you can query agent safety by mentioning the bot on X or Farcaster:

@AgentSafetyBase is 0x1234... safe?
@AgentSafetyBase score 0x1234...
@AgentSafetyBase report 0x1234...

The agent responds with the current safety score, badge status, stake amount, and any risk factors.

Health Check

The agent exposes a health endpoint at http://localhost:3001/health returning:

{
  "status": "healthy",
  "uptime": 21600,
  "lastCycle": "2026-02-08T09:33:22.488Z",
  "agentsMonitored": 10,
  "totalCycles": 4
}

Project Structure

agent-safety-monitor/
├── src/
│   ├── index.ts                 # Entry point, scheduler, health check
│   ├── heartbeat-executor.ts    # 6-phase cycle orchestrator
│   ├── erc8004-integration.ts   # ERC-8004 registry integration
│   ├── decision-engine.ts       # 2/3 validation, badge thresholds
│   ├── memory-manager.ts        # MEMORY.md + daily log I/O
│   ├── social-poster.ts         # Post templates + rate limiter
│   ├── mention-monitor.ts       # X + Farcaster mention responder
│   └── social/
│       ├── x-client.ts          # Twitter/X OAuth 1.0a client
│       └── farcaster-client.ts  # Farcaster via Neynar API
├── skills/
│   ├── base-monitoring/         # Transaction monitoring skill
│   ├── reputation-scoring/      # Score calculation skill
│   ├── agent-enforcement/       # On-chain enforcement skill
│   └── erc8004-registration/    # ERC-8004 registration skill
├── packages/hardhat/
│   ├── contracts/               # 4 Solidity contracts
│   ├── test/                    # 51 unit tests
│   └── deploy/                  # Deployment scripts
├── workspace/
│   ├── SOUL.md                  # Agent persona and values
│   ├── HEARTBEAT.md             # Cycle instructions
│   └── MEMORY.md                # Persistent agent state
├── config/
│   ├── contract-addresses.json  # Deployed contract addresses
│   └── detection-rules.json     # Violation thresholds
├── logs/daily/                  # Audit trail
├── Dockerfile                   # Container deployment
└── openclaw.json                # OpenClaw agent config

License

MIT

About

Autonomous AI agent that monitors 3,600+ onchain agents on Base. Detects rug pulls, wash trading, and drain attacks. Slashes stakes, mints safety badges, and posts alerts to X . Built on OpenClaw with ERC-8004 integration. No human in the loop.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages