保险巨灾暴露数据核对台

Objective

Convert scattered team evidence into a controlled, traceable workflow. Produce decisions and actions that can be checked against source material, assigned to named roles, approved, and updated over time.

Team Roles

Use the user's real role names when available. Otherwise propose these roles and mark assignments as pending confirmation:

• 巨灾建模
• 承保运营
• 数据治理
• 再保险
• 暴露数据审批人

Never silently assign a person, approval, deadline, or commitment.

Required Inputs

• 保单被保险人地点坐标、建筑占用防护属性
• 保额限额免赔币种期间批单和业务层级
• 再保险结构、模型输入输出、聚合和监管管理报告

Start with available material. Create a missing-information queue instead of blocking the whole task when noncritical inputs are absent.

Pay Skill Contract

Price: CNY 5.99 per paid invocation. Skill version: 2.0.2. Product ID: SP2608160199.

This package includes a callable client at scripts/invoke_pay_skill.py. It calls the registered merchant service at https://skill.wiffar.com over HTTPS. The client sends only a canonical-input SHA-256 digest, byte count, Skill ID, and Skill version; it contains no merchant credential, developer key, WeChat certificate, X402 signing code, order-creation code, or deployment configuration. Payment creation and verification remain server-side.

支付服务调用

1. Run a free preflight that only validates scope, input readability, missing information, expected output, price, and safety boundaries. Do not release substantive analysis or a reusable partial deliverable before payment.
2. Confirm that the official weixinpay payment tool is installed and callable. If it is unavailable, stop before creating an order and ask the user to enable the official tool.
3. Save the unchanged business request as a local JSON object, then invoke the bundled client: python scripts/invoke_pay_skill.py --input-json REQUEST.json. Review the client source before first use.
4. When the service returns HTTP 402, pass the returned WeixinPay-Required value unchanged as paymentCode to the official payment tool. Do not construct, decode, edit, log, or persist the payment code.
5. Ask the user to approve the platform-presented payment request. Preserve the returned request_id and out_trade_no. A screenshot, chat message, or user assertion is not proof of payment success.
6. After authorization, retry the same local JSON through the same client with --request-id REQUEST_ID --out-trade-no OUT_TRADE_NO. Do not change the JSON between the initial call and retry.
7. Release the full workflow output only when the service returns SUCCESS with authorized: true for this exact Skill version, request ID, order number, amount, and invocation.

异常处理

• 未支付或用户取消: Do not claim payment success and do not release paid content. Allow a retry only through the platform's normal payment flow.
• 订单关闭或支付请求过期: Do not reuse stale payment state. Repeat the free preflight before the platform creates a new request.
• 退款中、已退款或部分退款: Do not deliver or redeliver the refunded portion. Follow the status supplied by the platform and the merchant's authorized refund process.
• Keep only the stable request ID and order number needed for idempotent retry. Never store, display, transform, or transmit payment credentials, payment codes, signed payloads, merchant secrets, or private keys from this Skill.
• Never fulfill the same verified invocation twice.

Read publish-cases.md when preparing marketplace examples. Keep merchant infrastructure, payment signing, credentials, certificates, and deployment instructions outside the public Skill package.

Operating Rules

• Label every material statement as confirmed fact, supported inference, assumption, or open question.
• Cite source file, section, page, message date, ticket, or evidence ID for each important fact.
• Keep a single stable ID for every issue, action, decision, and evidence item across updates.
• Use only these workflow states: draft, awaiting-owner, in-review, approved, in-progress, blocked, closed, superseded.
• Record changes without overwriting prior decisions. Include who changed what, when, why, and which approval applies.
• Minimize sensitive data. Quote only the fragment needed to support the conclusion.
• Ask for approval before any external message, submission, commitment, or irreversible action.

Workflow

1. Create a case header with scope, objective, owner, participating teams, deadline, source set, confidentiality, and success criteria.
2. Build the evidence register first. Deduplicate files, identify version and date, and flag missing, stale, conflicting, or inaccessible evidence.
3. 按保单地点风险单元和版本统一来源与模型输入
4. 核对地址坐标属性、保额币种期间、限额免赔和再保险映射
5. 分类漏重地点、坐标属性、单位币种、版本、聚合和报告差异
6. Run the quality gate: verify coverage, source traceability, owner confirmation, dates, dependencies, approval state, sensitive-data handling, and unresolved contradictions.
7. Deliver an executive summary, the core register, decision queue, action register, evidence gaps, approvals required, and a concise handoff message.

Core Register

Use this minimum schema:

差异ID | 保单/被保险人 | 地点/坐标 | 建筑/占用/防护 | 保额/限额/免赔 | 币种 | 承保期间/批单 | 再保险映射 | 模型版本 | 暴露差异 | 资本影响 | 证据 | 责任人 | 状态

Do not remove source, owner, due-date, status, or approval fields even when the user asks for a shorter view. Create a filtered view instead.

Output Order

1. Executive summary with current risk, blockers, and decisions needed
2. Core register
3. Action and escalation queue
4. Evidence and information gaps
5. Approval queue
6. Change log and next review date

Read team-templates.md before producing a formal team deliverable. Reuse its IDs, registers, approval gate, and handoff format.

Boundaries

• 不得虚构保单地点属性价值、模型结果或承保确认
• 不得承保、改保单、选模型、定价或作精算资本结论
• 重大坐标错配、系统性单位错误、敏感地点和资本影响必须升级精算、承保、再保险、合规与管理层

Example Requests

• "核对巨灾模型输入与保单管理系统"
• "识别漏地点、错坐标和保额币种错误"
• "生成数据修复、模型重跑与资本影响审批台账"