Wealth Advisory Planner
Overview
This skill produces structured JSON planning outputs for private wealth advisory engagements. It fetches from the advisory API, resolves conflicts across source documents, performs domain calculations, and returns JSON conforming to the supplied answer template.
Workflow
- Read the task
input/prompt.txt, input/payloads/request_memo.md, and input/payloads/answer_template.json.
- Extract
client_id and analysis_type from the memo and template.
- Gather data from the advisory API (see references/api_surface.md).
- Resolve source conflicts (see references/source_resolution.md).
- Execute the analysis (see references/analysis_types.md).
- Fill the answer template conforming to references/answer_templates.md.
- Output only the final JSON object with no surrounding prose.
Step 1: Read Task Inputs
Read all three payload files in input/payloads/ to understand the engagement:
prompt.txt: Task framing and API_BASE reference
request_memo.md: Client ID, engagement type, planning horizon, output instructions
answer_template.json: Required JSON structure with all required_top_level_keys and field enums
The template's analysis_type enum value tells you which analysis to run.
Step 2: Gather API Data
Fetch all data the analysis needs. Always set API_BASE from the prompt environment. Make all GET calls in parallel when possible.
Read references/api_surface.md for the full API surface and field shapes. The minimum set per analysis type:
| Analysis Type |
Endpoints to Fetch |
| roth_conversion_rmd |
clients, source-documents, retirement-accounts, policies/tax, rmd-factors |
| ilit_crummey_implementation |
clients, source-documents, life-insurance, policies/tax |
| trust_comparison |
clients, source-documents, trust-candidates, policies/tax |
| estate_liquidity_action_plan |
clients, source-documents, life-insurance, trust-candidates, policies/tax |
Always filter source-documents, retirement-accounts, life-insurance, and trust-candidates by ?client_id={client_id}.
Step 3: Resolve Source Conflicts
Source documents may disagree. Follow the priority rules in references/source_resolution.md:
- SIGNED_PROFILE overrides ATTORNEY_MEMO, which overrides CRM_NOTE
- CUSTODIAN_EXPORT is authoritative for account balances. Never override with profile data.
- Beneficiary count always from SIGNED_PROFILE
- Life insurance data from the
/api/life-insurance endpoint is authoritative
Step 4: Execute Analysis
Read references/analysis_types.md for the full computation formulas. Key principles:
- Use
curl to the API_BASE for all data; never hardcode values from training examples
- All monetary values in USD rounded to cents (2 decimals)
- Years as integers, dates as ISO YYYY-MM-DD
- Use the rmd-factors endpoint for RMD divisors; use the tax policy endpoint for all rates and thresholds
- Grow accounts year-by-year using
expected_return; apply RMD divisors when age >= rmd_start_age
- For GRAT projections: compound the trust asset at
expected_growth_rate, subtract annuity each year
- For CRAT projections: same compounding, subtract payout each year, cap term at
max_crat_term_years
Step 5: Fill Answer Template
Read references/answer_templates.md for template field reference. Conform exactly to the template's required_top_level_keys. Every value must match its declared type (number, string, boolean, enum, array). Enums must use the exact strings from the template.
Set task_id from the task directory name (e.g. train_001, test_001). Use the same identifier string that appears in the prompt or path.
Step 6: Output
Return only the JSON object. No markdown fences, no explanatory prose, no trailing text. The JSON must be valid and parseable.
Important Constraints
- Never output task-specific values from training examples as defaults. Always compute from API data.
- Do not use a Python script to fetch API data; use
curl shell commands.
- Do not guess API values. Every number must come from an endpoint response.
- The
API_BASE variable is supplied by the task harness; check the environment or prompt for the actual URL.
estate_tax_rate is always 0.4, rmd_start_age is 73 (confirm from retirement account record).
planning_year defaults to 2026 from the client record or tax policy.
References
- API Surface - Full endpoint listing with field shapes
- Source Conflict Resolution - Priority rules for conflicting documents
- Analysis Types - Computation formulas for each analysis type
- Answer Templates - Template field reference and enum values
1---2name: wealth-advisory-planner-43description: "Structured wealth advisory planning for private wealth teams. Produces JSON outputs for Roth conversion and RMD analysis, ILIT Crummey funding implementation, GRAT vs CRAT trust comparisons, and combined estate liquidity action plans. Use when the task involves a client from the advisory environment, an answer_template.json with wealth advisory fields, API calls to the advisory API surface (clients, source documents, retirement accounts, life insurance, trust candidates, tax policies, RMD factors), or request memos referencing Roth conversions, ILIT, Crummey notices, GRAT, CRAT, estate liquidity, or source document conflict resolution."4license: MIT5compatibility: designed for deepagents-code6metadata:7 source: https://skillmd.com/skills/prism-shadow/wealth-advisory-planner-48---910# Wealth Advisory Planner1112## Overview1314This skill produces structured JSON planning outputs for private wealth advisory engagements. It fetches from the advisory API, resolves conflicts across source documents, performs domain calculations, and returns JSON conforming to the supplied answer template.1516## Workflow17181. Read the task `input/prompt.txt`, `input/payloads/request_memo.md`, and `input/payloads/answer_template.json`.192. Extract `client_id` and `analysis_type` from the memo and template.203. Gather data from the advisory API (see [references/api_surface.md](references/api_surface.md)).214. Resolve source conflicts (see [references/source_resolution.md](references/source_resolution.md)).225. Execute the analysis (see [references/analysis_types.md](references/analysis_types.md)).236. Fill the answer template conforming to [references/answer_templates.md](references/answer_templates.md).247. Output only the final JSON object with no surrounding prose.2526### Step 1: Read Task Inputs2728Read all three payload files in `input/payloads/` to understand the engagement:2930- `prompt.txt`: Task framing and `API_BASE` reference31- `request_memo.md`: Client ID, engagement type, planning horizon, output instructions32- `answer_template.json`: Required JSON structure with all `required_top_level_keys` and field enums3334The template's `analysis_type` enum value tells you which analysis to run.3536### Step 2: Gather API Data3738Fetch all data the analysis needs. Always set `API_BASE` from the prompt environment. Make all GET calls in parallel when possible.3940Read [references/api_surface.md](references/api_surface.md) for the full API surface and field shapes. The minimum set per analysis type:4142| Analysis Type | Endpoints to Fetch |43|---|---|44| roth_conversion_rmd | clients, source-documents, retirement-accounts, policies/tax, rmd-factors |45| ilit_crummey_implementation | clients, source-documents, life-insurance, policies/tax |46| trust_comparison | clients, source-documents, trust-candidates, policies/tax |47| estate_liquidity_action_plan | clients, source-documents, life-insurance, trust-candidates, policies/tax |4849Always filter source-documents, retirement-accounts, life-insurance, and trust-candidates by `?client_id={client_id}`.5051### Step 3: Resolve Source Conflicts5253Source documents may disagree. Follow the priority rules in [references/source_resolution.md](references/source_resolution.md):5455- SIGNED_PROFILE overrides ATTORNEY_MEMO, which overrides CRM_NOTE56- CUSTODIAN_EXPORT is authoritative for account balances. Never override with profile data.57- Beneficiary count always from SIGNED_PROFILE58- Life insurance data from the `/api/life-insurance` endpoint is authoritative5960### Step 4: Execute Analysis6162Read [references/analysis_types.md](references/analysis_types.md) for the full computation formulas. Key principles:6364- Use `curl` to the `API_BASE` for all data; never hardcode values from training examples65- All monetary values in USD rounded to cents (2 decimals)66- Years as integers, dates as ISO YYYY-MM-DD67- Use the rmd-factors endpoint for RMD divisors; use the tax policy endpoint for all rates and thresholds68- Grow accounts year-by-year using `expected_return`; apply RMD divisors when age >= rmd_start_age69- For GRAT projections: compound the trust asset at `expected_growth_rate`, subtract annuity each year70- For CRAT projections: same compounding, subtract payout each year, cap term at `max_crat_term_years`7172### Step 5: Fill Answer Template7374Read [references/answer_templates.md](references/answer_templates.md) for template field reference. Conform exactly to the template's required_top_level_keys. Every value must match its declared type (number, string, boolean, enum, array). Enums must use the exact strings from the template.7576Set `task_id` from the task directory name (e.g. `train_001`, `test_001`). Use the same identifier string that appears in the prompt or path.7778### Step 6: Output7980Return only the JSON object. No markdown fences, no explanatory prose, no trailing text. The JSON must be valid and parseable.8182## Important Constraints8384- Never output task-specific values from training examples as defaults. Always compute from API data.85- Do not use a Python script to fetch API data; use `curl` shell commands.86- Do not guess API values. Every number must come from an endpoint response.87- The `API_BASE` variable is supplied by the task harness; check the environment or prompt for the actual URL.88- `estate_tax_rate` is always 0.4, `rmd_start_age` is 73 (confirm from retirement account record).89- `planning_year` defaults to 2026 from the client record or tax policy.9091## References9293- [API Surface](references/api_surface.md) - Full endpoint listing with field shapes94- [Source Conflict Resolution](references/source_resolution.md) - Priority rules for conflicting documents95- [Analysis Types](references/analysis_types.md) - Computation formulas for each analysis type96- [Answer Templates](references/answer_templates.md) - Template field reference and enum values