erc-8004-development

ERC-8004: Trustless Agents

ERC-8004 is a Draft EIP for discovering, choosing, and interacting with AI agents across organizational boundaries without pre-existing trust. It defines three on-chain registries deployed as per-chain singletons on any EVM chain.

Authors: Marco De Rossi (MetaMask), Davide Crapis (EF), Jordan Ellis (Google), Erik Reppel (Coinbase)

Full spec: references/spec.md

When to Use This Skill

• Registering AI agents on-chain (ERC-721 identity)
• Building or querying agent reputation/feedback systems
• Searching and discovering agents by capabilities, trust models, or endpoints
• Working with the Agent0 TypeScript SDK (agent0-sdk)
• Implementing ERC-8004 smart contract integrations
• Setting up agent wallets, MCP/A2A endpoints, or OASF taxonomies

Core Architecture

Three lightweight registries, each deployed as a UUPS-upgradeable singleton:

Agent identity = agentRegistry (string eip155:{chainId}:{contractAddress}) + agentId (ERC-721 tokenId).

Each agent's agentURI points to a JSON registration file (IPFS or HTTPS) advertising name, description, endpoints (MCP, A2A, ENS, DID, wallet), OASF skills/domains, trust models, and x402 support.

See: references/contracts.md for full contract interfaces and addresses.

Quick Start with Agent0 SDK (TypeScript)

npm install agent0-sdk

Register an Agent

RPC_URL, PRIVATE_KEY, and PINATA_JWT are declared in this skill's metadata.openclaw.envVars. Use throwaway/testnet keys for development; reach for a hardware wallet or scoped signer for any mainnet activity.

import { SDK } from 'agent0-sdk';

const sdk = new SDK({
  chainId: 84532,  // Base Sepolia
  rpcUrl: process.env.RPC_URL,
  privateKey: process.env.PRIVATE_KEY,
  ipfs: 'pinata',
  pinataJwt: process.env.PINATA_JWT,
});

const agent = sdk.createAgent(
  'MyAgent',
  'An AI agent that analyzes crypto markets',
  'https://example.com/agent-image.png'
);

// Configure endpoints and capabilities
await agent.setMCP('https://mcp.example.com', '2025-06-18', true);  // auto-fetches tools
await agent.setA2A('https://example.com/.well-known/agent-card.json', '0.3.0', true);
agent.setENS('myagent.eth');
agent.setActive(true);
agent.setX402Support(true);
agent.setTrust(true, false, false);  // reputation only

// Add OASF taxonomy
agent.addSkill('natural_language_processing/natural_language_generation/summarization', true);
agent.addDomain('finance_and_business/investment_services', true);

// Register on-chain (mints NFT + uploads to IPFS).
// Sends a real transaction signed with PRIVATE_KEY - confirm chainId, signer, and balance before running.
const tx = await agent.registerIPFS();
const { result } = await tx.waitConfirmed();
console.log(`Registered: ${result.agentId}`);  // e.g. "84532:42"

Search for Agents

const sdk = new SDK({ chainId: 84532, rpcUrl: process.env.RPC_URL });

// Search by capabilities
const agents = await sdk.searchAgents({
  hasMCP: true,
  active: true,
  x402support: true,
  mcpTools: ['financial_analyzer'],
  supportedTrust: ['reputation'],
});

// Get a specific agent
const agent = await sdk.getAgent('84532:42');

// Semantic search
const results = await sdk.searchAgents(
  { keyword: 'crypto market analysis' },
  { sort: ['semanticScore:desc'] }
);

Give Feedback

// Prepare optional off-chain feedback file
const feedbackFile = await sdk.prepareFeedbackFile({
  text: 'Accurate market analysis',
  capability: 'tools',
  name: 'financial_analyzer',
  proofOfPayment: { txHash: '0x...', chainId: '8453', fromAddress: '0x...', toAddress: '0x...' },
});

// Submit feedback (value=85 out of 100). On-chain tx; same caveats as `registerIPFS()` above.
const tx = await sdk.giveFeedback('84532:42', 85, 'starred', '', '', feedbackFile);
await tx.waitConfirmed();

// Read reputation summary
const summary = await sdk.getReputationSummary('84532:42');
console.log(`Average: ${summary.averageValue}, Count: ${summary.count}`);

See: references/sdk-typescript.md for full SDK API reference.

Registration File Format

Every agent's agentURI resolves to this JSON structure:

{
  "type": "https://eips.ethereum.org/EIPS/eip-8004#registration-v1",
  "name": "MyAgent",
  "description": "What it does, pricing, interaction methods",
  "image": "https://example.com/agent.png",
  "services": [
    { "name": "MCP", "endpoint": "https://mcp.example.com", "version": "2025-06-18", "mcpTools": ["tool1"] },
    { "name": "A2A", "endpoint": "https://example.com/.well-known/agent-card.json", "version": "0.3.0" },
    { "name": "OASF", "endpoint": "https://github.com/agntcy/oasf/", "version": "v0.8.0",
      "skills": ["natural_language_processing/summarization"],
      "domains": ["finance_and_business/investment_services"] },
    { "name": "ENS", "endpoint": "myagent.eth", "version": "v1" },
    { "name": "agentWallet", "endpoint": "eip155:8453:0x..." }
  ],
  "registrations": [
    { "agentId": 42, "agentRegistry": "eip155:84532:0x8004A818BFB912233c491871b3d84c89A494BD9e" }
  ],
  "supportedTrust": ["reputation", "crypto-economic", "tee-attestation"],
  "active": true,
  "x402Support": true
}

The registrations field creates a bidirectional cryptographic link: the NFT points to this file, and this file points back to the NFT. This enables endpoint domain verification via /.well-known/agent-registration.json.

See: references/registration.md for best practices (Four Golden Rules) and complete field reference.

Contract Addresses

All registries deploy to deterministic vanity addresses via CREATE2 (SAFE Singleton Factory):

Mainnet (Ethereum, Base, Polygon, Arbitrum, Optimism, etc.)

Testnet (Sepolia, Base Sepolia, etc.)

Same proxy addresses on: Ethereum, Base, Arbitrum, Avalanche, Celo, Gnosis, Linea, Mantle, MegaETH, Optimism, Polygon, Scroll, Taiko, Monad, BSC + testnets.

Reputation System

Feedback uses signed fixed-point numbers: value (int128) + valueDecimals (uint8, 0-18).

Anti-Sybil: getSummary() requires a non-empty clientAddresses array (caller must supply trusted reviewer list). Self-feedback is rejected (agent owner/operators cannot submit feedback on their own agent).

See: references/reputation.md for full feedback system, off-chain file format, and aggregation details.

OASF Taxonomy (v0.8.0)

Open Agentic Schema Framework provides standardized skills (136) and domains (204) for agent classification.

Top-level skill categories: natural_language_processing, images_computer_vision, audio, analytical_skills, multi_modal, agent_orchestration, advanced_reasoning_planning, data_engineering, security_privacy, evaluation_monitoring, devops_mlops, governance_compliance, tool_interaction, retrieval_augmented_generation, tabular_text

Top-level domain categories: technology, finance_and_business, healthcare, legal, education, life_science, agriculture, energy, environmental_science, government, manufacturing, transportation, and more.

Use slash-separated paths: agent.addSkill('natural_language_processing/natural_language_generation/summarization', true).

Key Concepts

Reference Index

Official Resources

• EIP Discussion: https://ethereum-magicians.org/t/erc-8004-trustless-agents/25098
• Contracts: https://github.com/erc-8004/erc-8004-contracts
• Best Practices: https://github.com/erc-8004/best-practices
• SDK Docs: https://sdk.ag0.xyz
• SDK Docs Source: https://github.com/agent0lab/agent0-sdk-docs
• TypeScript SDK: https://github.com/agent0lab/agent0-ts
• Python SDK: https://github.com/agent0lab/agent0-py
• Subgraph: https://github.com/agent0lab/subgraph
• OASF: https://github.com/agntcy/oasf