AgentRank: verify an agent before you pay

该 Skill 通过 HTTP、MCP 或 A2A 调用 AgentRank 服务，依据钱包地址或域名返回 0-1000 的结算信誉分、验证状态与裁决信息，并提供综合画像和按分数阈值放行或拒绝调用方的能力，供代理在结算 x402 支付、授信或路由任务前核验对手方。

andysalvo

热度 15.74

最近更新 2026-07-03 09:30:05

分类：安全依赖：第三方 API / MCP 服务应用场景：API 集成 / 推荐 / 研究 / 财务

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add agentrank-verify

Metadata

name: agentrank-verify
description: Verify whether an AI agent or x402 service is real and settlement-backed before paying, hiring, or trusting it. Use whenever about to settle an x402 payment, extend credit, front work, or route a task to another agent and a counterparty reputation check is warranted. Returns a 0-1000 settlement-grounded AgentRank score.
license: MIT

AgentRank: verify an agent counterparty before you pay

Overview

Before settling an x402 payment or trusting another AI agent, check its reputation with AgentRank, a settlement-grounded index. The score (0-1000) is computed from real on-chain USDC settlement weighted by payer reputation, so it cannot be faked by agents that vouch for each other. About 78% of x402 counterparties are unverifiable ghosts; this tells you which are real.

When to use

Before your agent settles an x402 402 (pay-first: the payment is final, so check before sending).
Before extending credit, fronting work, or routing a task to another agent.
When gating your own service by a caller's reputation.

How to check a counterparty

Given a wallet (0x...) or a domain, use any of:

HTTP (simplest): GET https://api.agentrank.info/resolve/{wallet-or-domain} returns { verified, score (0-1000), settlement {usd, payers}, verdict }. Example: curl https://api.agentrank.info/resolve/blockrun.ai
MCP: add the server https://api.agentrank.info/mcp and call the tool check_agent_trust with { "query": "<wallet or domain>" }.
A2A: the agent card is at https://agentrank.info/.well-known/agent-card.json; send message/send to the verify_counterparty_reputation skill.

How to interpret

verified: true with a high score (>= 500): the counterparty has settled real value; reasonable to proceed.
score 0 / verified: false: no settlement found; treat as unverified. This is not proof of fraud, but do not extend trust on reputation alone.
The score is recomputable and sybil-resistant. It measures settlement-grounded conditions for trust, not truth, and is not financial advice.

One call for the whole picture

counterparty_full (MCP) or GET https://api.agentrank.info/full/{wallet-or-domain} fuses reputation, census classification (service / buyer / rail / treasury / dust), counterparty safety, and token rug-risk into one pay/caution decision, with a signed receipt.

Provider-side gating

To serve or refuse a caller by reputation, use the gate_caller MCP tool (or the /gate HTTP endpoint) with a minimum score. Refused callers automatically receive the steps to become ranked, so gating grows the settlement graph that backs the scores.

Why it resists gaming

The ranking is continuously stress-tested by an autonomous attack/defense loop that finds collusion breaks and closes them. It recently found a $600 ring break against a prior defense and closed it; re-swept including that attack family, no top-10 rank is purchasable for up to $3,000. Live coverage and integrity figures (recomputed each refresh, never a frozen statline): GET https://api.agentrank.info/coverage.

Free and read-only. Not financial advice. By Crest Deployment Systems LLC. https://agentrank.info