菜单
Use when the customer wants to order a product (e.g. open a savings account or apply for a credit card) or change their contact details. Enforces the human-in-the-loop confirmation and compliance guardrails.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Operations runbook for the agentic-banking-ecosystem repo. USE THIS SKILL when the user asks to deploy, build, provision, generate data, index, ingest, register, or clean up any part of the project — including infrastructure (azd), container images, Container Apps, the customer/product MCP servers, AI Search indexes, Foundry toolboxes, and the banking agents. Covers the full lifecycle: provision → generate data → build → deploy MCP → register toolboxes → index → ingest → deploy agents → clean up.
Use when an agent asks what data is required for a customer, account or transaction to be compliant — an onboarding checklist, an eligibility gap analysis, or "what do we still need to collect". Maps each compliance rule onto the concrete banking data model and returns an extensive, field-level requirements breakdown.
Use before asking Bank North's Compliance agent (the ask_compliance tool) any regulatory, KYC/AML, sanctions, fraud or product-eligibility question. Gathers the signed-in customer's real data first, then composes a precise, self-contained compliance question so the answer is specific to this customer and product rather than generic.
Use when the agent should proactively spot an optimisation for the signed-in customer (e.g. a large uninvested balance on the current account) and offer help. Always asks permission first, then gives two concrete, personalised recommendations, and only acts human-in-the-loop.
Use when an employee asks which product to recommend to a customer, or to compare products for a specific customer need. Produces a grounded, segment-aware recommendation.
Use when the customer asks about their balance, transactions, or personal details. Produces a grounded, confidential answer from the customer data MCP tools.
| name | product-ordering |
| description | Use when the customer wants to order a product (e.g. open a savings account or apply for a credit card) or change their contact details. Enforces the human-in-the-loop confirmation and compliance guardrails. |
Guide a customer through starting a product order or a contact-detail change. Every write is human-in-the-loop and must be confirmed before it commits.
confirm=false
(order_product for a new product, update_customer for contact details) to
produce a preview of exactly what will change.update_overview with awaiting_confirmation=true) and ask them to confirm
in words.confirm=true. order_product opens an order case (status requested,
the holding starts pending) and returns an order_id.delivery details from the order (estimated business days +
shipping address) and tell the customer when it will arrive and where.requested →
approved | rejected; approved → shipped → delivered. Use
list_orders / get_order to report status ("Your GoldCard application is
approved and shipping"). Advancing the status (update_order_status) is a
bank-side decision and is itself human-in-the-loop; approving an order
activates the holding, rejecting it marks the holding rejected.order_id, new account id (for orders),
the delivery ETA + address (for cards), and the current order status.confirm=true without an explicit customer "yes".requested or
pending; report the actual status from the order case.