BAP-578 Agent Economy

BAP-578 Agent Economy

Use this skill to design, model, and operate economic systems where BAP-578 agents transact, collaborate, compete, and provide services. This covers agent funding flows, inter-agent payment patterns, reputation mechanisms, service marketplaces, and economic incentive alignment.

When to use this skill

• Designing how agents fund each other or pay for services.
• Building reputation systems that track agent quality.
• Creating a service marketplace where agents offer and consume capabilities.
• Modeling economic incentives to encourage healthy ecosystem behavior.
• Designing staking or collateral mechanisms for agent accountability.
• Planning token economics that complement the NFA system.
• Analyzing agent economy health metrics.

The Four Identity Questions (Economy View)

1) Who are you?

The agent economy is the layer where agents become economic actors. Each agent has an on-chain balance (BNB held in the NFA contract), a persona (defining what services it offers), and a status (active/inactive). In an agent economy, identity is directly tied to economic capability: agents with funding can pay for services, agents with expertise can earn revenue, and agents with reputation attract more business.

2) What do you remember?

Economic memory is captured through on-chain events and state:

• Funding history: AgentFunded and AgentWithdraw events track every BNB flow
• Transaction patterns: which agents fund each other, how often, how much
• Reputation scores: derived from successful service completions (stored in vault or logic contract)
• Service records: metadata updates and vault content track what services were delivered
• Economic relationships: persistent patterns of agent-to-agent interaction

3) What can you do?

The agent economy enables:

• Service provision: agents offer capabilities (analysis, generation, monitoring)
• Service consumption: agents pay other agents for capabilities they lack
• Reputation building: successful service delivery increases reputation
• Fund management: agents accumulate and spend BNB for services
• Marketplace participation: agents list services, bid on tasks, and fulfill orders
• Collaborative networks: agents form alliances and share resources

4) How can I trust it?

Economic trust in the agent economy comes from:

• On-chain balances: agent funding is verifiable and transparent
• Event audit trail: every payment and interaction is recorded
• Reputation scores: derived from verifiable service history
• Logic contract enforcement: automated service agreements via smart contracts
• Vault integrity: service records anchored by hash verification

Agent Funding Model

How agents hold and use funds

Each BAP-578 agent has an on-chain BNB balance managed by the NFA contract:

Agent #17
├── balance: 2.5 BNB    ← held in NFA contract
├── active: true
└── owner: 0xABC         ← controls withdrawals

Funding flows:

• Inbound: Anyone can call fundAgent(tokenId) with BNB
• Outbound: Only the token owner can call withdrawFromAgent(tokenId, amount)
• Agent-to-agent: Owner withdraws from Agent A, funds Agent B

Direct agent-to-agent payment pattern

Since agents cannot directly transfer BNB to each other (only owners can withdraw), agent-to-agent payments follow this pattern:

1. Agent A's owner withdraws from Agent A
2. Agent A's owner funds Agent B
3. Both transactions are atomic (can be batched in one tx via helper contract)

Payment helper contract

A helper contract can batch withdraw-and-fund into a single transaction:

// SPDX-License-Identifier: MIT
pragma solidity ^0.8.20;

interface INFA {
    function withdrawFromAgent(uint256 tokenId, uint256 amount) external;
    function fundAgent(uint256 tokenId) external payable;
    function getAgentState(uint256 tokenId) external view returns (
        uint256 balance, bool active, address logicAddress,
        uint256 createdAt, address owner
    );
}

contract AgentPaymentRouter {
    INFA public immutable nfa;

    event AgentPayment(
        uint256 indexed fromAgent,
        uint256 indexed toAgent,
        uint256 amount,
        string reason
    );

    constructor(address _nfa) {
        nfa = INFA(_nfa);
    }

    function payAgent(
        uint256 fromTokenId,
        uint256 toTokenId,
        uint256 amount,
        string calldata reason
    ) external {
        // Verify caller owns the source agent
        (, , , , address owner) = nfa.getAgentState(fromTokenId);
        require(msg.sender == owner, "Not source agent owner");

        // Withdraw from source agent
        nfa.withdrawFromAgent(fromTokenId, amount);

        // Fund destination agent
        nfa.fundAgent{value: amount}(toTokenId);

        emit AgentPayment(fromTokenId, toTokenId, amount, reason);
    }

    receive() external payable {}
}

Service Marketplace

Concept

A service marketplace where agents advertise capabilities and other agents (or users) can purchase those services. Services are defined in agent metadata (persona + experience) and priced via the marketplace contract.

Service listing structure

{
  "serviceId": "analysis-001",
  "agentTokenId": 17,
  "name": "DeFi Yield Analysis",
  "description": "Comprehensive yield comparison across top 20 DeFi protocols",
  "priceBnb": "0.005",
  "deliveryTime": "1 hour",
  "category": "finance",
  "rating": 4.8,
  "completedJobs": 142
}

Marketplace contract pattern

// SPDX-License-Identifier: MIT
pragma solidity ^0.8.20;

contract AgentMarketplace {
    struct ServiceListing {
        uint256 agentTokenId;
        uint256 price;
        bool active;
        string serviceURI;
    }

    struct Order {
        uint256 listingId;
        uint256 buyerAgentId;
        uint256 amount;
        OrderStatus status;
        uint256 createdAt;
    }

    enum OrderStatus { Pending, Delivered, Completed, Disputed }

    mapping(uint256 => ServiceListing) public listings;
    mapping(uint256 => Order) public orders;
    uint256 public nextListingId;
    uint256 public nextOrderId;

    uint256 public constant PLATFORM_FEE_BPS = 250; // 2.5%

    event ServiceListed(uint256 indexed listingId, uint256 indexed agentId, uint256 price);
    event OrderCreated(uint256 indexed orderId, uint256 indexed listingId, uint256 buyerAgentId);
    event OrderCompleted(uint256 indexed orderId);

    function listService(
        uint256 agentTokenId,
        uint256 price,
        string calldata serviceURI
    ) external returns (uint256) {
        uint256 id = nextListingId++;
        listings[id] = ServiceListing({
            agentTokenId: agentTokenId,
            price: price,
            active: true,
            serviceURI: serviceURI
        });
        emit ServiceListed(id, agentTokenId, price);
        return id;
    }

    function createOrder(uint256 listingId, uint256 buyerAgentId) external payable returns (uint256) {
        ServiceListing memory listing = listings[listingId];
        require(listing.active, "Listing not active");
        require(msg.value >= listing.price, "Insufficient payment");

        uint256 id = nextOrderId++;
        orders[id] = Order({
            listingId: listingId,
            buyerAgentId: buyerAgentId,
            amount: msg.value,
            status: OrderStatus.Pending,
            createdAt: block.timestamp
        });
        emit OrderCreated(id, listingId, buyerAgentId);
        return id;
    }

    function completeOrder(uint256 orderId) external {
        Order storage order = orders[orderId];
        require(order.status == OrderStatus.Delivered, "Not delivered");

        ServiceListing memory listing = listings[order.listingId];
        uint256 fee = (order.amount * PLATFORM_FEE_BPS) / 10000;
        uint256 payment = order.amount - fee;

        // Payment would go to the service agent's funding
        // Implementation depends on NFA integration
        order.status = OrderStatus.Completed;
        emit OrderCompleted(orderId);
    }
}

Reputation System

On-chain reputation model

Reputation is built from verifiable service history:

Reputation Score = f(completed_orders, ratings, disputes, agent_age, balance)

Reputation components

Service completion rate:

completion_rate = completed_orders / (completed_orders + cancelled_orders + disputed_orders)

Average rating:

• After each completed order, the buyer rates the service (1-5 stars)
• Stored on-chain or in vault with hash verification

Economic stake:

• Higher agent balance signals commitment (skin in the game)
• Agents with more BNB locked are less likely to misbehave

Agent age:

• Older agents (higher createdAt delta) have longer track records
• New agents start with neutral reputation

Dispute rate:

dispute_rate = disputed_orders / total_orders

Reputation tiers

Tier 1: New Agent      — < 10 completed orders
Tier 2: Established    — 10-50 completed orders, > 4.0 rating
Tier 3: Trusted        — 50-200 completed orders, > 4.5 rating
Tier 4: Expert         — 200+ completed orders, > 4.7 rating
Tier 5: Elite          — 500+ completed orders, > 4.8 rating, < 1% disputes

Storing reputation

Option 1: Logic contract state

A reputation logic contract maintains scores per agent:

contract ReputationLogic {
    struct Reputation {
        uint256 completedOrders;
        uint256 totalRating;
        uint256 disputes;
        uint256 lastUpdated;
    }

    mapping(uint256 => Reputation) public reputations;

    function getScore(uint256 tokenId) external view returns (uint256) {
        Reputation memory rep = reputations[tokenId];
        if (rep.completedOrders == 0) return 0;
        uint256 avgRating = (rep.totalRating * 100) / rep.completedOrders;
        uint256 completionBonus = rep.completedOrders > 100 ? 20 : rep.completedOrders / 5;
        uint256 disputePenalty = (rep.disputes * 100) / rep.completedOrders;
        return avgRating + completionBonus - disputePenalty;
    }
}

Option 2: Vault-based reputation

Store detailed reputation data in the agent's vault:

{
  "reputation": {
    "score": 487,
    "tier": "Expert",
    "completedOrders": 234,
    "averageRating": 4.72,
    "disputeRate": 0.008,
    "specializations": {
      "defi-analysis": { "orders": 145, "rating": 4.8 },
      "market-research": { "orders": 89, "rating": 4.6 }
    },
    "lastUpdated": "2026-03-06T10:00:00Z"
  }
}

Verify integrity via vaultHash comparison.

Economic Incentive Design

Incentive alignment goals

1. Encourage quality: high-quality service providers earn more
2. Discourage fraud: bad actors lose reputation and funding
3. Promote participation: new agents can bootstrap via free mints
4. Sustain the ecosystem: platform fees fund operations

Incentive mechanisms

Staking for service providers:

• Agents must maintain a minimum balance to list services
• Balance acts as collateral — can be slashed for disputes
• Higher stakes unlock higher-value service tiers

Escrow for payments:

• Buyer payment held in escrow contract until service delivery confirmed
• Automatic release after confirmation or timeout
• Dispute resolution via arbitration (human or automated)

Referral rewards:

• Agents earn a bonus for referring new agents or buyers
• Tracked via logic contract or event logs

Volume discounts:

• Frequent buyers get reduced prices from marketplace agents
• Loyalty tracked on-chain or in vault

Economic health metrics

Agent Collaboration Patterns

Hub-and-spoke model

One coordinator agent distributes tasks to specialist agents:

        ┌── Agent A (research)
        │
Coordinator ── Agent B (analysis)
        │
        └── Agent C (reporting)

The coordinator receives funding, pays specialists, and delivers combined output to the buyer.

Peer-to-peer model

Agents directly transact with each other without a coordinator:

Agent A ←→ Agent B ←→ Agent C

Each agent discovers and pays peers independently. Best for simple bilateral services.

Pool model

Multiple agents contribute to a shared pool and share rewards:

Agent A ─┐
Agent B ──┼── Shared Pool ── Rewards distributed proportionally
Agent C ─┘

Implemented via a pool logic contract that tracks contributions and distributions.

Economy Bootstrap Strategy

Phase 1: Seeding (first 100 agents)

• Free mints create initial agent population
• Platform provides sample services and templates
• Manual curation of service listings
• Incentivize first transactions with bonus funding

Phase 2: Growth (100-1000 agents)

• Organic service discovery via marketplace
• Reputation system begins differentiating quality
• Agent-to-agent payments become regular
• Community-driven service categories emerge

Phase 3: Maturity (1000+ agents)

• Self-sustaining marketplace with diverse services
• Reputation tiers drive premium pricing
• Agent collaboration networks form organically
• Platform fees cover operations and development

Risk Mitigation in Agent Economies

Sybil resistance

Prevent one actor from creating many agents to game reputation:

• Free mint limit per address constrains initial sybil cost
• Paid mints add economic cost to creating fake identities
• Reputation requires actual completed orders (can't be faked cheaply)
• Logic contract binding adds development cost per agent type

Marketplace fraud prevention

• Escrow model: Funds held until buyer confirms delivery
• Dispute resolution: Third-party arbitrator or DAO vote
• Reputation slashing: Bad actors lose reputation permanently
• Minimum stake: Service providers must lock BNB as collateral
• Time-locks: Large withdrawals have a cooling period

Economic attack vectors and mitigations

Tokenomics Integration

While BAP-578 uses native BNB, it can be extended with a governance or utility token:

Potential token uses

• Governance: Token holders vote on marketplace rules, fee levels, and feature priorities
• Staking rewards: Stake tokens to earn a share of marketplace fees
• Premium features: Token holders unlock advanced agent capabilities
• Fee discounts: Pay marketplace fees in tokens at a discount
• Agent badges: Verified token holders get special reputation badges

Revenue distribution model

Marketplace fee (2.5% per order)
    │
    ├── 40% → Treasury (operations, development)
    ├── 30% → Token stakers (reward distribution)
    ├── 20% → Service agent (bonus on top of payment)
    └── 10% → Referrer (if applicable)

Output Format

When asked for agent economy help, respond with:

1. Economic model (payment flows, incentives, pricing)
2. Smart contract code (payment router, marketplace, reputation)
3. Incentive analysis (who benefits, potential abuse, mitigations)
4. Metrics (what to measure and healthy ranges)
5. Bootstrap plan (how to get from zero to self-sustaining)

Related Skills

• bap578
• bap578-business
• bap578-logic-contracts