Wealth Advisory Planning
This skill produces structured JSON for private wealth advisory engagements covering Roth conversions, ILIT Crummey funding, trust comparisons, and estate liquidity planning. All data comes from a task-group advisory API; no local computation can replace API lookups.
High-level workflow
- Parse the request memo for
client_id, engagement type, and planning horizon. - Read the
answer_template.jsonin the task payloads to know the exact output schema. - Query the advisory API for all relevant data endpoints.
- Resolve conflicting source records using the priority rules below.
- Compute numeric outputs using the formulas in references/formulas.md.
- Select enum values using the decision logic in references/enums.md.
- Return a single JSON object matching the template; no prose outside the JSON.
API reference
The harness exposes the advisory API at API_BASE or via the configured base URL from environment_access.md. All endpoints are read-only GETs. See references/api-reference.md for the full endpoint catalog with field descriptions.
For any client engagement, query these endpoints in parallel, filtering by client_id where applicable:
GET /api/clients/{client_id}— demographic, estate value, liquid assets, filing statusGET /api/source-documents— filter to the client; used for profile facts, priority, and conflict resolutionGET /api/retirement-accounts— filter to the client; traditional/roth balances, expected return, recommended conversion yearsGET /api/life-insurance— filter to the client; death benefit, premium, contribution date, existing-policy-transfer flagGET /api/trust-candidates— filter to the client; asset value, growth rate, GRAT/CRAT parametersGET /api/policies/tax— static tax constants (annual gift exclusion, estate tax exemption, estate tax rate, bracket targets, charitable deduction rate)GET /api/rmd-factors— static RMD divisor table by age
The /api/clients list endpoint is optional and only needed for client-id lookup or context.
Source resolution
Client records may conflict because systems imported data at different times. Resolve every conflict using the priority hierarchy in references/source-resolution.md. The controlling source for each fact category feeds both the source_resolution output and the numeric calculations.
The most recent SIGNED_PROFILE (by effective_date) is the default authority for profile facts (age, income, beneficiaries, filing status, philanthropic intent, family transfer priority). CUSTODIAN_EXPORT controls account balances, RMD start age, and expected returns. ATTORNEY_MEMO controls estate value and can override asset source selections for trust engagements.
A CRM_NOTE is stale and used only when no other source for that fact exists. A STALE_MARKETING_INTAKE source type, if present, ranks below CRM notes.
Analysis dispatch
Read the request memo and the answer_template.json to determine the analysis type. The template’s analysis_type field (or the required_top_level_keys) tells you which output schema applies:
| Template signature | Analysis type | Reference section |
|---|---|---|
conversion_plan + rmd_projection + legacy_projection |
roth_conversion_rmd |
references/formulas.md: Roth Conversion |
gift_plan + administration + estate_result |
ilit_crummey_implementation |
references/formulas.md: ILIT Crummey |
grat + crat + estate_context |
trust_comparison |
references/formulas.md: Trust Comparison |
ilit + trust_transfer + action_set |
estate_liquidity_action_plan |
references/formulas.md: Estate Liquidity |
Output rules
Every output must be a single JSON object with no surrounding prose or markdown. Apply these rules uniformly:
task_idis the stable task identifier from the prompt (e.g.,"train_001"or the test environment equivalent).client_idis the client identifier from the memo.- All USD amounts are JSON numbers rounded to two decimal places (cents).
- All dates use ISO 8601
YYYY-MM-DDformat. - Enum values must exactly match the allowed values in the answer template; no approximations or near-matches.
action_setarrays in estate liquidity outputs must be sorted alphabetically.- Numbers must be JSON number types, never strings.
Use the bundled Python script scripts/compute.py for deterministic numeric calculations. Call it with the relevant arguments for each analysis section; it prints JSON to stdout for integration into the final output.