Tool: Wallet reputation scoring for agent trust in multi-agent networks

[jacobsd32-cpu]

Problem

In multi-agent networks, agents need to assess trust before delegating tasks, sharing resources, or transacting with other agents. When Agent A routes a task to Agent B, there's currently no standardized way to evaluate B's reliability or track record.

For agents operating with crypto wallets (common in x402/MCP payment flows), on-chain behavioral history provides a strong trust signal.

Proposal

Add wallet reputation scoring as a built-in tool or MCP integration for AgentKit networks.

DJD Agent Score provides a 0–100 trust score from on-chain behavioral analysis:

import { createAgent, createNetwork } from "@inngest/agent-kit";

const reputationTool = {
  name: "check_agent_reputation",
  description: "Score an agent's wallet reputation before delegating tasks or transacting",
  handler: async ({ wallet }: { wallet: string }) => {
    const res = await fetch(
      `https://djd-agent-score.fly.dev/score/basic/${wallet}`
    );
    return res.json();
  },
};

const coordinator = createAgent({
  name: "coordinator",
  tools: [reputationTool],
  system: `Before delegating high-value tasks, check the agent's wallet reputation.
    Require score >= 40 for financial operations.
    Flag agents with reliability < 20 for manual review.`,
});

Use cases in multi-agent routing

1. Trust-based routing — Route sensitive tasks to agents with higher reputation scores
2. Payment gating — Require minimum reputation before processing payments
3. Sybil prevention — Detect when multiple "agents" share wallet funding patterns
4. Progressive trust — Start with small tasks, increase delegation as reputation grows

API details

• Free tier: 3 scores/day, no API key
• x402 payments: USDC on Base for higher volume
• MCP compatible: Works as an MCP tool server
• npm: djd-agent-score-client
• Quick start: https://gist.github.com/jacobsd32-cpu/44e316704e7c029e0850218880a13764

This would complement AgentKit's deterministic routing with a trust dimension based on observed behavior.

[linear]

EXE-1383