Private Wealth Advisory Skill
Use this skill when the task involves private wealth advisory planning: Roth conversion and RMD tax summaries, ILIT Crummey funding cycles, GRAT vs CRAT comparisons, or estate liquidity action plans.
Environment
The task environment provides a private wealth advisory API. Locate the base URL from the harness-supplied API_BASE environment variable; if unset, default to http://task-env:9008/.
Step 1 — Fetch all advisory data
Pull every relevant endpoint in parallel. All data is JSON.
GET {API_BASE}/api/health — verify service is available
GET {API_BASE}/api/clients — all client records
GET {API_BASE}/api/clients/{client_id} — single client detail (if needed)
GET {API_BASE}/api/source-documents — signed profiles, attorney memos, CRM notes
GET {API_BASE}/api/retirement-accounts — custodian IRA/401(k) exports
GET {API_BASE}/api/life-insurance — ILIT policy records
GET {API_BASE}/api/trust-candidates — GRAT/CRAT trust candidates
GET {API_BASE}/api/policies/tax — tax constants (exclusions, brackets, rates)
GET {API_BASE}/api/rmd-factors — RMD divisors keyed by integer age
Step 2 — Determine the analysis type and target client
Read the prompt and request memo to find:
- The
client_id(e.g. CLT-1001, CLT-2001) - The
analysis_type: one ofroth_conversion_rmd,ilit_crummey_implementation,trust_comparison, orestate_liquidity_action_plan - The
planning_year(always 2026 in staged environments) - Any horizon year (e.g. 2046) from the memo
- The
task_idas supplied (e.g. train_001, test_001)
The analysis type dictates which sections of the output template to populate. The train examples cover four distinct templates; match the template shape to the analysis type exactly.
Step 3 — Resolve conflicting source records
Some client facts appear in multiple source documents with different values. Always resolve conflicts using the priority rules below, derived from the pattern in five training answers.
Source priority for profile facts
Higher-trust sources override lower-trust sources. Choose the most recent record from the highest available tier:
SIGNED_PROFILE(most authoritative)ATTORNEY_MEMOCUSTODIAN_EXPORTCRM_NOTESTALE_MARKETING_INTAKE(least authoritative)
Profile facts resolved this way include: annual_non_ira_income, marginal_tax_rate, beneficiary_count, philanthropic_intent, family_transfer_priority, filing_status, age, liquid_assets, estate_value.
Source priority for account balances
CUSTODIAN_EXPORT(authoritative for balances)SIGNED_PROFILECRM_NOTE
Use the /api/retirement-accounts data as CUSTODIAN_EXPORT authority for traditional_balance, roth_balance, expected_return, rmd_start_age, recommended_conversion_years.
Source priority for life-insurance / ILIT facts
SIGNED_PROFILEATTORNEY_MEMOCUSTODIAN_EXPORTCRM_NOTE
Use /api/life-insurance data for death_benefit, annual_premium, planned_contribution_date, is_existing_policy_transfer.
Source priority for trust assets
ATTORNEY_MEMOSIGNED_PROFILECRM_NOTE
Use /api/trust-candidates data for trust parameters.
Source priority for beneficiary count
SIGNED_PROFILEATTORNEY_MEMOCUSTODIAN_EXPORTCRM_NOTESTALE_MARKETING_INTAKE
Source priority for goals (philanthropic intent, family transfer priority)
SIGNED_PROFILEATTORNEY_MEMOCUSTODIAN_EXPORTCRM_NOTESTALE_MARKETING_INTAKE
In the source_resolution section
For every output template that has a source_resolution block, report the source type that actually controlled the key facts:
controlling_profile_source: the source type (one of the five enums above) whose profile facts were usedcontrolling_account_source: the source type for account balancescontrolling_beneficiary_source: for ILIT taskscontrolling_policy_source: for ILIT/estate taskscontrolling_goal_source: for trust comparison and estate planning taskscontrolling_asset_source: for trust comparison tasks
Step 4 — Compute the analysis-specific values
Use the step-by-step formulas below for each analysis type. Run the calculations with the deterministic Python scripts from calc.py. Do not hand-derive values; invoke the script functions with the resolved parameters.
4a. Roth conversion / RMD analysis (roth_conversion_rmd)
Parameters from data:
income=annual_non_ira_incomefrom the controlling profilebracket_targetfrom/api/policies/tax→conversion_bracket_targets[filing_status]marginal_ratefrom the controlling profiletraditional_balance,roth_balance,expected_returnfrom the controlling account sourcerecommended_conversion_yearsfrom the controlling account sourcermd_start_agefrom the controlling account sourcehorizon_yearfrom the memo
Conversion plan formulas:
annual_conversion_amount = bracket_target - income
total_converted = annual_conversion_amount × recommended_conversion_years
total_conversion_tax = total_converted × marginal_rate
first_conversion_year = planning_year
conversion_years = recommended_conversion_years
conversion_years_positive = recommended_conversion_years
RMD projection formulas:
Two scenarios: baseline (no conversion) and conversion.
RMD at age = traditional_balance / rmd_divisor[age]; divisor from /api/rmd-factors.
For baseline: grow the traditional balance from the starting balance at expected_return each year. At each year where age >= rmd_start_age, compute RMD, tax = RMD × marginal_rate, subtract RMD from balance, then grow. Sum taxes through horizon.
For conversion: same, but also subtract annual_conversion_amount from the traditional balance (and add to Roth) during each conversion year (starting in planning_year, for conversion_years years). The Roth grows at the same expected_return. Sum RMD taxes through horizon.
baseline_rmd_tax_through_horizon = sum of RMD taxes in baseline scenario
conversion_rmd_tax_through_horizon = sum of RMD taxes in conversion scenario
rmd_tax_savings_through_horizon = baseline - conversion
Legacy projection formulas:
projected_roth_balance_horizon = Roth balance after horizon-year growth
projected_traditional_balance_horizon = Traditional balance after horizon-year growth
heir_tax_profile: MOSTLY_TAX_FREE if Roth > 1.5 × Trad at horizon; MOSTLY_TAXABLE if Trad > 1.5 × Roth; MIXED_TAXABLE_AND_TAX_FREE otherwise.
Recommendation rules:
primary_action:STAGED_ROTH_CONVERSIONif positive RMD tax savings and bracket headroom exists;DEFERif near-term RMD (< 2 years out) but bracket headroom;NO_CONVERSIONif no savingssuitability:SUITABLEif savings > 0 and conversion feasible;BORDERLINEif near-RMD;DEFERotherwiserisk_flag:TAX_BRACKET_MANAGEMENTif tax bracket is the primary concern;LIQUIDITY_CONSTRAINTif conversion tax exceeds liquid assets;RMD_NEAR_TERMif RMDs begin within 2 years
4b. ILIT Crummey implementation (ilit_crummey_implementation)
Parameters:
annual_exclusion_per_beneficiaryfrom/api/policies/tax→annual_gift_exclusion[planning_year]beneficiary_countfrom controlling beneficiary sourceannual_premium,death_benefit,planned_contribution_date,is_existing_policy_transferfrom controlling policy source
Gift plan formulas:
annual_exclusion_capacity = annual_exclusion_per_beneficiary × beneficiary_count
premium_gap = max(0, annual_premium - annual_exclusion_capacity)
Administration timeline (Crummey mechanics):
contribution_date=planned_contribution_datenotice_due_date= contribution_date + 7 dayswithdrawal_window_end= notice_due_date + 30 daysearliest_premium_payment_date= withdrawal_window_end + 1 daynotices_required=beneficiary_countdedicated_bank_account_required= true when premium_gap == 0
Estate result:
death_benefitas providedestate_inclusion_risk:LOW_IF_FORMALITIES_METif new policy and formalities followed;EXCLUSION_SHORTFALLif premium > exclusion capacity;THREE_YEAR_LOOKBACKif existing policy transfer;THREE_YEAR_LOOKBACK_AND_EXCLUSION_SHORTFALLif bothprojected_outside_estate_if_implemented= death_benefit if LOW_IF_FORMALITIES_MET else 0tax_liquidity_support= death_benefit × 0.4
Recommendation rules:
primary_action:FUND_WITH_CRUMMEY_NOTICESif premium fits within exclusion;USE_LIFETIME_EXEMPTION_FOR_SHORTFALLif premium gap > 0 without lookback;USE_NEW_POLICY_OR_ACCEPT_LOOKBACKif existing transfer without gap;DISCLOSE_LOOKBACK_AND_USE_EXEMPTIONif existing transfer with gapsuitability:SUITABLE_WITH_ADMINISTRATIONif low risk;BORDERLINEif gap exists;NOT_SUITABLEif lookback + gaprisk_flag: matchesestate_inclusion_risk
4c. Trust comparison (trust_comparison)
Parameters from trust candidate:
asset_value,expected_growth_rate,grat_term_years,grat_annuity_rate,crat_term_years,crat_payout_rate
Tax parameters: estate_tax_rate (0.4), charitable_deduction_rate (0.35) from /api/policies/tax
GRAT formulas:
projected_remainder_to_heirs = asset_value × (1 + expected_growth_rate)^grat_term_years - asset_value × grat_annuity_rate × grat_term_years
estimated_estate_tax_reduction = projected_remainder_to_heirs × estate_tax_rate
mortality_inclusion_risk = "TERM_SURVIVAL_REQUIRED"
term_years = grat_term_years
CRAT formulas:
projected_charitable_remainder = asset_value × (1 + expected_growth_rate)^crat_term_years - asset_value × crat_payout_rate × crat_term_years
estimated_income_tax_deduction = projected_charitable_remainder × charitable_deduction_rate
family_transfer_fit = "LOW" (CRATs go to charity)
term_years = crat_term_years
Estate context:
taxable_estate= estate_value - estate_tax_exemption (from/api/policies/tax)estate_tax_exposure= taxable_estate × estate_tax_rateliquid_assets_available= liquid_assetsliquidity_gap_before_planning= max(0, estate_tax_exposure - liquid_assets)
Recommendation rules:
preferred_strategy:GRATif family_transfer_priority is "high" and philanthropic_intent is "low" or "moderate";CRATif philanthropic_intent is "high"rationale_code:CHILDREN_TRANSFER_PRIORITYfor GRAT preference;PHILANTHROPIC_PRIORITYfor CRATalternate_role:SECONDARY_CHARITABLE_TOOLwhen GRAT preferred;SECONDARY_FAMILY_TRANSFER_TOOLwhen CRAT preferred
4d. Estate liquidity action plan (estate_liquidity_action_plan)
This is a compound analysis combining ILIT and trust comparison with an action set.
Compute the ILIT section using 4b formulas and the trust_transfer section using 4c formulas.
Estate context:
taxable_estate= estate_value - estate_tax_exemptionestate_tax_exposure= taxable_estate × 0.4liquidity_gap_before_planning= max(0, estate_tax_exposure - liquid_assets)
Action set: Choose from these alphabetically-sorted actions based on the gap analysis:
ATTORNEY_DRAFT_REVIEW— always includeCRAT_FOR_CHARITABLE_REMAINDER— include when philanthropic_intent is "high"GRAT_FOR_APPRECIATING_SHARES— include when family_transfer_priority is "high" and trust candidate has positive GRAT remainderILIT_CRUMMEY_NOTICE_CYCLE— include when ILIT premium gap is 0LIFETIME_EXEMPTION_ALLOCATION— include when liquidity gap > 0
Recommendation rules:
primary_action:COMBINE_ILIT_AND_GRATwhen both ILIT and GRAT are feasible;CRAT_WITH_LIQUIDITY_REVIEWwhen philanthropic intent dominates;ILIT_WITH_EXEMPTION_REVIEWwhen ILIT is feasible but GRAT is notsequencing:ILIT_FIRST_THEN_GRATwhen both are needed and no lookback issue;TRUST_DECISION_FIRSTwhen trust dominates;ILIT_FIRST_THEN_ATTORNEY_REVIEWwhen lookback existsrisk_flag:LOW_IF_FORMALITIES_METwhen no concerns;EXCLUSION_SHORTFALLwhen premium > exclusion capacity;THREE_YEAR_LOOKBACKwhen existing policy transfer;THREE_YEAR_LOOKBACK_AND_EXCLUSION_SHORTFALLwhen both
Step 5 — Produce the JSON output
Follow the exact answer template shape for the analysis type. Templates are provided in templates/. The template dictates every required key, enum value, and data type.
Rules:
- All dollar amounts are JSON numbers rounded to cents (two decimal places)
- All years are JSON integers
- All dates are ISO 8601 YYYY-MM-DD strings
- All enums use EXACTLY the values from each template's
fieldsspecification task_idis the stable task identifier from the prompt- Output only the JSON object; no prose outside it
See calc.py for the deterministic implementation of all formulas.