菜单
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.
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 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.
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 | proactive-impulse |
| description | 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. |
Detect a concrete optimisation opportunity for the signed-in customer and offer it proactively — but only after asking for permission, and never acting without an explicit confirmation.
detect_opportunities (product data tools) for
the customer in context. It returns idle-cash opportunities (with an
estimated_annual_gain) and, when the customer holds no card, a credit-card
candidate. If it returns nothing, stay silent — do not invent an opportunity.idle_cash opportunity with the
concrete numbers from the tool: movable amount, target product, its rate,
and the estimated_annual_gain in EUR/year. Cite the product conditions
(file + numbered section, e.g. "savings-products.md §…"). Keep a liquidity
buffer on the current account.summarize_spending
to see top categories/merchants and get_product for the annual fee, then
state a concrete cost/benefit (e.g. "annual fee €29 vs. your travel spend
of €X — the travel insurance/cashback likely outweighs the fee"). Only
recommend the card if the benefit plausibly beats the fee; otherwise say so
honestly.confirm=false, confirm in
words, then commit with confirm=true. Note that opening the savings product
is the actionable step here; the actual funds transfer is arranged
separately.update_overview when any pending
action is created.