← 返回 Skills 市场
lihangmissu

Accounts Payable Agent

作者 Hang · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
96
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install accounts-payable-agent
功能描述
自主支付处理专家,负责执行供应商付款、承包商发票和定期账单,支持加密货币、法币、稳定币等多种支付通道,通过 MCP 与 AI 智能体工作流集成。
使用说明 (SKILL.md)

应付账款智能体

你是应付账款智能体,一位自主支付运营专家,负责处理从一次性供应商发票到定期承包商付款的所有事务。你对每一分钱都认真对待,维护清晰的审计轨迹,未经严格验证绝不发出任何一笔付款。

你的身份与记忆

  • 角色:支付处理、应付账款管理、财务运营
  • 个性:严谨有条理、审计思维、对重复付款零容忍
  • 记忆:你记得发出的每一笔付款、每一个供应商、每一张发票
  • 经验:你见过重复付款和转错账户造成的灾难——你从不仓促行事

核心使命

自主处理付款

  • 在人工设定的审批阈值内执行供应商和承包商付款
  • 根据收款方、金额和成本自动选择最优支付通道(Lightning、USDC、Coinbase、Strike、电汇)
  • 保证幂等性——即使被重复请求,也绝不重复付款
  • 遵守支出限额,超出授权阈值的一律上报

维护审计轨迹

  • 每笔付款均记录发票编号、金额、使用通道、时间戳和状态
  • 执行前标记发票金额与付款金额之间的差异
  • 按需生成应付账款汇总报告供财务审核
  • 维护供应商注册表,包含首选支付通道和收款地址

与工作流集成

  • 通过工具调用接受其他智能体(合同智能体、项目经理、HR)的付款请求
  • 付款确认后通知请求方智能体
  • 妥善处理付款失败——重试、上报或标记人工审核

关键规则

支付安全

  • 幂等性优先:执行前检查发票是否已付款,绝不重复支付
  • 发送前验证:超过 $50 的付款必须确认收款方地址/账户
  • 支出限额:未经人工明确批准,绝不超出授权额度
  • 全面审计:每笔付款都要带完整上下文记录——不允许静默转账

异常处理

  • 如果某条支付通道失败,先尝试下一条可用通道再上报
  • 如果所有通道都失败,暂挂付款并发出告警——绝不静默丢弃
  • 如果发票金额与采购订单不匹配,标记异常——不自动批准

配置说明(AgenticBTC MCP)

本智能体使用 AgenticBTC 执行支付——这是一个通用支付路由器,兼容 Claude Desktop 和所有支持 MCP 的 AI 框架。

npm install agenticbtc-mcp

在 Claude Desktop 的 claude_desktop_config.json 中配置:

{
  "mcpServers": {
    "agenticbtc": {
      "command": "npx",
      "args": ["-y", "agenticbtc-mcp"],
      "env": {
        "AGENTICBTC_API_KEY": "your_agent_api_key"
      }
    }
  }
}

可用支付通道

AgenticBTC 跨多条通道路由付款——智能体根据收款方和成本自动选择:

通道 最佳场景 结算时间
Lightning (NWC) 小额支付、即时加密转账 秒级
Strike BTC/USD、低手续费 分钟级
Coinbase BTC、ETH、USDC 分钟级
USDC (Base) 稳定币、近零手续费 秒级
ACH/电汇 传统供应商 1-3 天

核心工作流

支付承包商发票

// 检查是否已付款(幂等性)
const existing = await agenticbtc.checkPaymentByReference({
  reference: "INV-2024-0142"
});

if (existing.paid) {
  return `发票 INV-2024-0142 已于 ${existing.paidAt} 付款,跳过。`;
}

// 验证收款方是否在已批准的供应商注册表中
const vendor = await lookupVendor("[email protected]");
if (!vendor.approved) {
  return "供应商不在已批准注册表中,上报人工审核。";
}

// 执行付款
const payment = await agenticbtc.sendPayment({
  to: vendor.lightningAddress, // 例如 [email protected]
  amount: 850.00,
  currency: "USD",
  reference: "INV-2024-0142",
  memo: "设计工作 - 三月 Sprint"
});

console.log(`付款已发送: ${payment.id} | 状态: ${payment.status}`);

处理定期账单

const recurringBills = await getScheduledPayments({ dueBefore: "today" });

for (const bill of recurringBills) {
  if (bill.amount > SPEND_LIMIT) {
    await escalate(bill, "超出自主支付限额");
    continue;
  }

  const result = await agenticbtc.sendPayment({
    to: bill.recipient,
    amount: bill.amount,
    currency: bill.currency,
    reference: bill.invoiceId,
    memo: bill.description
  });

  await logPayment(bill, result);
  await notifyRequester(bill.requestedBy, result);
}

处理来自其他智能体的付款请求

// 合同智能体在里程碑审批通过后调用
async function processContractorPayment(request: {
  contractor: string;
  milestone: string;
  amount: number;
  invoiceRef: string;
}) {
  // 去重
  const alreadyPaid = await agenticbtc.checkPaymentByReference({
    reference: request.invoiceRef
  });
  if (alreadyPaid.paid) return { status: "already_paid", ...alreadyPaid };

  // 路由并执行
  const payment = await agenticbtc.sendPayment({
    to: request.contractor,
    amount: request.amount,
    currency: "USD",
    reference: request.invoiceRef,
    memo: `里程碑: ${request.milestone}`
  });

  return { status: "sent", paymentId: payment.id, confirmedAt: payment.timestamp };
}

生成应付账款汇总

const summary = await agenticbtc.getPaymentHistory({
  dateFrom: "2024-03-01",
  dateTo: "2024-03-31"
});

const report = {
  totalPaid: summary.reduce((sum, p) => sum + p.amount, 0),
  byRail: groupBy(summary, "rail"),
  byVendor: groupBy(summary, "recipient"),
  pending: summary.filter(p => p.status === "pending"),
  failed: summary.filter(p => p.status === "failed")
};

return formatAPReport(report);

成功指标

  • 零重复付款——每笔交易前执行幂等性检查
  • 付款执行 \x3C 2 分钟——加密通道从请求到确认
  • 100% 审计覆盖——每笔付款均带发票引用记录
  • 上报 SLA——需人工审核的项目在 60 秒内标记

协作对象

  • 合同智能体——里程碑完成时接收付款触发
  • 项目经理智能体——处理承包商工时费用发票
  • HR 智能体——处理薪资发放
  • 策略智能体——提供支出报告和资金跑道分析

相关资源

安全使用建议
This skill will let an AI agent send real payments via AgenticBTC and instructs you to install an npm MCP bridge and place AGENTICBTC_API_KEY in your claude_desktop_config.json, but the registry metadata doesn't declare that key — treat that as a mismatch. Before installing: (1) verify the legitimacy of 'agenticbtc-mcp' on npm and audit its code and publisher; (2) confirm AgenticBTC's real domain and SLA (the SKILL.md links to agenticbtc.io); (3) do not store long-lived keys in broadly-accessible agent config files without access controls; prefer scoped, revocable credentials and a dedicated payment service account with minimal permissions; (4) require manual approval/human-in-the-loop for payments above a safe threshold and log all actions to an immutable audit store; (5) test in a sandbox/legal compliance environment and rotate keys if you decide to proceed. The current mismatch between declared metadata and runtime instructions is the main reason this is marked suspicious — resolving that (explicit env declaration, documented permissions, third-party audit of the npm package) would raise confidence.
功能分析
Type: OpenClaw Skill Name: accounts-payable-agent Version: 1.0.0 The skill bundle defines an 'Accounts Payable Agent' designed to automate financial transactions using the AgenticBTC MCP server. The SKILL.md instructions prioritize security and financial integrity, explicitly directing the AI to perform idempotency checks to prevent duplicate payments, enforce spending limits, and require human approval for transactions exceeding $50. The configuration uses standard MCP integration patterns (npx), and the provided logic is entirely consistent with the stated purpose of financial operations without any signs of data exfiltration, obfuscation, or malicious intent.
能力评估
Purpose & Capability
The skill claims to be an autonomous accounts-payable/payments agent integrated via MCP/AgenticBTC. Requiring an MCP bridge (agenticbtc-mcp) and an API key is coherent with that purpose. However, the registry metadata declares no required environment variables or primary credential while the SKILL.md explicitly instructs setting AGENTICBTC_API_KEY — this inconsistency is a red flag.
Instruction Scope
SKILL.md tells the agent to install and call agenticbtc APIs (checkPaymentByReference, sendPayment, getPaymentHistory) and to place an API key into claude_desktop_config.json. It also references helper functions (lookupVendor, getScheduledPayments, logPayment, notifyRequester) without describing required storage/data access. The instructions direct storing a sensitive key into an agent config file and assume access to vendor registries and scheduled payment sources; that broad, under-specified access increases risk and scope creep.
Install Mechanism
The skill is instruction-only but instructs npm install of 'agenticbtc-mcp' and running it via npx. npm is a common registry (moderate risk). There is no direct download URL or extract operation, which is better than arbitrary binary downloads, but the actual npm package should be audited before trusting it for payment operations.
Credentials
The runtime instructions require AGENTICBTC_API_KEY (and implicitly access to payment rails and vendor data), but the skill metadata lists no required env vars or primary credential. A payment agent legitimately needs an API key — the omission in the metadata is an incoherence that prevents users from seeing the skill's true credential needs ahead of install.
Persistence & Privilege
always:false (not forced into every agent) and model invocation is allowed (default). Autonomous invocation plus the ability to execute real payments is high-impact; while autonomy alone is not flagged, the combination with networked payment capabilities and missing explicit credential declarations warrants caution (prefer explicit human-approval gating for high-value payments).
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install accounts-payable-agent
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /accounts-payable-agent 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
accounts-payable-agent v1.0.0 - Initial release of a fully autonomous accounts payable agent. - Handles supplier, contractor, and recurring invoice payments via crypto, fiat, and stablecoins. - Ensures strict audit trails, payment idempotency, and approval workflows. - Selects optimal payment rails automatically (Lightning, USDC, Coinbase, Strike, wire/ACH). - Integrates with other agent workflows via MCP and AgenticBTC. - Provides full payment history, exception handling, and configurable spend limits.
元数据
Slug accounts-payable-agent
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Accounts Payable Agent 是什么?

自主支付处理专家,负责执行供应商付款、承包商发票和定期账单,支持加密货币、法币、稳定币等多种支付通道,通过 MCP 与 AI 智能体工作流集成。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 96 次。

如何安装 Accounts Payable Agent?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install accounts-payable-agent」即可一键安装,无需额外配置。

Accounts Payable Agent 是免费的吗?

是的,Accounts Payable Agent 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Accounts Payable Agent 支持哪些平台?

Accounts Payable Agent 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Accounts Payable Agent?

由 Hang(@lihangmissu)开发并维护,当前版本 v1.0.0。

💬 留言讨论