git:20260808.5ce8d82 to git:20260824.d18190a

---

name: kelly-finance

description: Build, analyze, repair, and audit finance three-statement models and operating forecasts. Use when the user invokes $kelly-finance, asks for 财务三表, 三表模型, income statement, balance sheet, cash flow statement, forecast model, budget model, financial projections, FP&A model, SaaS/unit economics forecast, fundraising model, board finance pack, investor model, model checks, balance-sheet balancing, cash runway, working capital, capex/depreciation schedules, debt schedules, scenario/sensitivity cases, or Excel/Sheets-ready finance outputs.

metadata:

category: finance

tags:

- risk:local-write

- surface:busabase

---

# Kelly Finance

## Overview

Use this skill as a practical FP&A and corporate-finance modeling desk. It

builds clean three-statement starter models with a dependency-free Python

script, and gives a Busabase-backed review workspace (AirApp) for the model

dashboard and its model-quality check queue — audit the model logic, leave

notes, approve/request-changes/block/dismiss each check, and hand rework back

to the agent — before anything is delivered to investors or written back to

the source-of-truth books.

Default behavior is AirApp-first. Unless the user explicitly asks only for

explanation, ensure a model exists (run the trusted build script below if the

`model` Base is empty) and give the user the clickable AirApp URL (or the

local preview URL when local preview is explicitly requested). Use chat-only

mode only when the user says "纯聊天", "chat only", "不要打开 UI", or similar.

## App UI Screenshots

<table>

<tr>

<td width="50%"><img src="assets/screenshots/overview.webp" alt="Kelly Finance model overview"></td>

<td width="50%"><img src="assets/screenshots/checks.webp" alt="Kelly Finance model audit checks"></td>

</tr>

<tr>

<td><strong>Overview</strong><br>Model KPI cards and a five-year forecast table (revenue through free cash flow), with a needs-attention summary.</td>

<td><strong>Model audit checks</strong><br>Review queue for statement ties, hardcodes, formula direction, and debt/working-capital linkage — each check approvable, blockable, or sent back with a note.</td>

</tr>

<tr>

<td width="50%"><img src="assets/screenshots/workbook.webp" alt="Kelly Finance workbook tab contract"></td>

<td width="50%"></td>

</tr>

<tr>

<td><strong>Workbook</strong><br>Generated workbook path plus the tab contract — Assumptions, Income Statement, Balance Sheet, Cash Flow, Checks.</td>

<td></td>

</tr>

</table>

## Mandatory Dependencies

2. Read and follow `$busabase` for connection, target Space, node discovery, ChangeRequests, review, and merge behavior.

3. Read and follow `$busabase-app-creator` for resource modeling, AirApp runtime limits, security, validation, and deployment.

If a dependency is unavailable, preserve this skill's local artifact and product contracts, stop before the unavailable Busabase operation, and report the exact missing dependency. Do not invent a second data backend.

## Boundary

- Review workspace only. The skill reads and writes its own Busabase Bases; it must not connect to banks/accounting systems, send files, mutate ERP records, move money, or change external systems.

- Treat financial models and assumptions as sensitive by convention, even though the bundled demo data is synthetic (`ExampleCo`).

- Any external action, such as sending a model to investors or changing source-of-truth books, is approval-required and executed outside the app by the agent after human review. This skill never performs that step itself.

## Busabase Resources

Three Bases under one application Folder (`kelly-finance`), declared in

- `model`: one row per model run (usually just the current run, `model-id`

`current`) — company, currency, display unit, model purpose, the forecast

`periods` array (JSON), headline metrics (revenue CAGR, ending cash, free

cash flow, balance check), warnings, and the generated workbook's

path/tab contract. Written by `scripts/build_three_statement_model.mjs`

after the Python starter-model script runs, or by an agent that has

reviewed a real workbook and knows the real computed figures.

- `checks`: one row per model-audit check (formula ties, model-quality

issues, delivery notes) — the raw check fields plus the reviewer's

decision (`decision-action`/`decision-comment`/`decided-at`) and, once

`scripts/execute_decisions.mjs` runs, an execution marker, all written

directly onto the same row. Unlike a derived-status review queue, a

check's `status` is stored directly (set by the reviewer's decision

action) — the needs_review/approved/done/blocked *counts* shown on the

dashboard are recomputed client-side from the checks on every read.

- `settings`: sanitized config summary (company/currency defaults, no

secrets), one row keyed by `record-id`/`kind`.

Resources provision lazily through an idempotent Busabase ChangeRequest the

first time the app runs in a Space; see `references/finance-ui-schema.md` for

exact field shapes.

## Create A Three-Statement Template

`scripts/build_three_statement_model.py` is the real modeling engine — a

dependency-free Python script that hand-writes a genuine `.xlsx` workbook

(balance-sheet balancing, PP&E/depreciation roll-forward, debt/interest

schedule, working-capital tie), kept exactly as Python since rewriting real

financial-modeling formulas in JavaScript would mean reimplementing that

logic instead of just calling it. The trusted wrapper spawns it and records

the result in Busabase:

```bash

node skills/kelly-finance/scripts/build_three_statement_model.mjs \

--company "ExampleCo" --start-year 2026 --years 5 \

--currency USD --base-revenue 1000000 \

--output /tmp/three_statement_model.xlsx --seed-checks --apply

```

Without `--apply` this is a dry run: the `.xlsx` is still generated locally

(that step has no external side effect either way), but nothing is written

to Busabase. `--seed-checks` also seeds the standard model-quality check

queue (balance sheet, cash roll-forward, net income tie, PP&E tie, debt tie —

check *definitions* only, never fabricated figures) into the `checks` Base.

Once an agent has actually opened the workbook and read real computed

values, pass `--periods '[{"label":"2026","revenue":1000000,...}, ...]'` to

seed the dashboard's forecast table with those real numbers — this script

never invents financial figures itself.

After generating, open or inspect the workbook when possible. If the user

needs a polished investor-facing model, add formatting, scenario cases, and

relevant operating schedules after the starter model is created.

## Local App

Default behavior is AirApp-first — give the user the clickable AirApp URL.

requested.

Required app views (hash routes):

- `#/overview`: model KPI dashboard, forecast table, and top model checks.

- `#/checks` and `#/checks/<id>`: review queue for formula ties, model

quality issues, and delivery notes. Users can approve, request changes,

block, or dismiss each check — written directly onto the check record

through `busabase-sdk`.

- `#/workbook`: generated workbook path and tab contract.

- `#/settings`: sanitized config summary, onboarding marker, and data

provider.

## Demo Mode

- `?demo=1` opens a deterministic offline model (`ExampleCo`, a five-year

fundraising forecast) for screenshots and review. Demo mode never reads or

writes Busabase; demo decisions stay in the browser and are discarded on

refresh.

- `lang=en` or `lang=zh` forces UI chrome language.

## Review Or Repair A Model

When reviewing an existing workbook:

- Preserve user formulas and formatting unless asked to rebuild.

- First map sheets, time axis, linked statements, hardcodes, and check rows.

- Find the actual source of a mismatch before changing formulas.

- Use a separate `Checks` or `Audit` tab if the workbook lacks one.

- Never force a balance-sheet plug without labeling it and explaining why it is temporary.

- Record findings as rows in the `checks` Base (see field contract in `references/finance-ui-schema.md`) so the human reviewer can work through them in the app.

Use `references/three-statement-modeling.md` for the review checklist, forecast-driver conventions, and model quality bar.

## Workflow

1. `node scripts/build_three_statement_model.mjs ... --apply` builds the

starter workbook and writes the `model` row (and optionally the standard

check queue) to Busabase.

2. Open the app. **Overview** shows the model KPI dashboard and forecast

table; **Checks** is the review queue.

3. For each check, record `Approve` / `Request changes` / `Block` /

`Dismiss` with an optional reviewer note — written straight onto the

check record.

4. `node scripts/execute_decisions.mjs --apply` (dry run without `--apply`)

re-reads Busabase and writes an execution marker (`execution-status`,

`execution-detail`, `executed-at`) onto every approved check, reporting

which are ready for the agent's next step. It performs no external side

effect — no export, filing, or transmission to investors or a

source-of-truth system.

Read `references/finance-ui-schema.md` before editing the app, scripts, or

## The Domain Model

domain model: the needs_review/approved/done/blocked rollup

(`computeMetricsFromChecks()`), the decision -> status mapping

(`statusForAction()`), simple derived arithmetic over an already-computed

periods array (`deriveModelMetrics()` — CAGR/ending-cash/free-cash-flow, NOT

a reimplementation of the real modeling math), and the deterministic demo

dataset (`demoSnapshot()`). Every function is pure and deterministic — same

inputs always produce the same output — so a human reviewer can audit every

status and count by hand. It backs the live Busabase read path

the snapshot shape.

## Modeling Standards

- Use positive revenue and expense rows with clear sign labels; cash-flow outflows should be negative.

- Separate historical actuals from forecast periods when actuals are supplied.

- Use named scenarios or assumption columns for base/downside/upside cases instead of duplicating whole models.

- State whether currency values are units, thousands, or millions.

- Mark estimates as assumptions, not facts.

## Safety

- Review workspace only: never send a model to investors, change

source-of-truth books, or otherwise act outside this app — any real

external action is approval-required and executed by the agent outside

the app after human review.

- Do not invent forecast figures beyond the deterministic demo data or a

figure an agent actually read from a real workbook; `--periods` on the

build script exists precisely so real computed values (not guesses) reach

the dashboard.

- Keep the model/check rows minimal and use stable ids so repeated builds

stay idempotent (the build script upserts by `model-id`/`check-id`, never

duplicating rows).

## Useful Commands

```bash

node skills/kelly-finance/scripts/build_three_statement_model.mjs --apply

node skills/kelly-finance/scripts/execute_decisions.mjs --apply

python3 skills/kelly-finance/scripts/build_three_statement_model.py --output /tmp/model.xlsx --company "ExampleCo"

```