| name | compliance-consultation |
| description | 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. |
Compliance Consultation Skill
The Compliance agent gives a far more specific, actionable answer when the
question already contains the relevant customer facts. Never send a bare "Can
this customer get a credit card?" — gather the data model context first, then
ask a structured question.
When to use
- Any product order where eligibility must be checked (savings, credit card,
children's account) — before the product-ordering preview.
- Any regulatory / KYC / AML / sanctions / fraud question a customer raises.
- Any guardrail check where you would otherwise guess.
Only proceed when the ask_compliance tool is available. If it is not, do not
answer from your own knowledge — tell the customer the compliance service is
unavailable and defer to a human adviser.
Method
-
Gather the customer's data model context. Read what the rule needs from
the customer MCP tools before you ask:
get_customer → full_name, date_of_birth (derive the age),
nationality, tax_residency, address, email, phone, kyc_status,
segment, created_at.
list_accounts → existing holdings: category, status, whether a
current/savings holding with an iban exists (a reference account),
and any holding with a non-null credit_limit.
Fetch only the fields the scenario needs; never pull another customer's data.
-
Identify the exact scenario. Name the workflow and the specific product
(e.g. "opening a FlexSave savings account", "applying for a GoldCard credit
card") including its category and whether it carries credit exposure.
-
Compose a structured, self-contained question. Put the facts in the
question so the compliance answer is unambiguous. Use this shape:
Scenario: <workflow + specific product, with category/credit exposure>.
Customer: age (DOB ), nationality , tax residency ,
kyc_status , segment , customer since .
Existing holdings: <e.g. one active current account with IBAN, one savings>.
Question: Is this customer eligible? List every data-model field and
document still required, the required state of each, and the determination
(approve / review / reject / escalate) with citations.
Fill in only the fields relevant to the rule; state a value as "unknown" if
you could not fetch it rather than omitting it silently.
-
Relay the decision. Report the compliance agent's determination, the
field/document gaps it named, and its citations verbatim
(file + numbered section). Do not soften a reject or invent a missing rule.
-
Feed it back into the flow. If compliance says approve and all fields are
met, continue to the product-ordering preview. If it says review / reject /
escalate, stop and route per the compliance answer and the escalation skill.
Output format
- To the customer: the plain-language outcome and any information they must
still provide, with the cited rule.
- Internally: the structured question above passed to
ask_compliance.
Guardrails
- Always gather the customer facts BEFORE calling
ask_compliance; a question
without customer context yields a generic, less useful answer.
- Never fabricate a field value to make the question look complete — mark
unknowns as unknown.
- Defer personalised financial advice to a human adviser regardless of the
compliance answer.