sheet-model（板材模型）

Sheet Model — Headless Spreadsheet for Agents

Build spreadsheet models programmatically using HyperFormula (headless computation engine) + ExcelJS (xlsx export). The spreadsheet IS both the computation and the deliverable.

When to Use This vs the xlsx Skill

Setup

Install dependencies (run once):

cd {baseDir}
npm install

Architecture

Agent code (declarative)          SheetModel wrapper          Output
────────────────────────    ──────────────────────────    ──────────────
addRow('Revenue', 701384)   → HyperFormula (compute)     → Console table
addRow('EBITDA', formula)   → Named ranges auto-tracked  → .xlsx with live
addScenarioSheet(config)    → {Name} refs resolved       →   formulas +
getValue('EBITDA')          → Dependency graph updates    →   styling +
exportXlsx('model.xlsx')    → ExcelJS (export)           →   cond. format

Core API

All code is ESM (.mjs). Import the wrapper:

import { SheetModel } from '{baseDir}/lib/sheet-model.mjs';
const M = new SheetModel();

Creating Sheets and Adding Data

M.addSheet('Data');

// Section headers (bold, no value)
M.addSection('Data', 'BALANCE SHEET');
M.addBlank('Data');

// Data rows — addRow returns the A1 row number (use it in SUM ranges)
const r_first = M.addRow('Data', '  Revenue',    701384, { name: 'Revenue' });
const r_costs = M.addRow('Data', '  Costs',     -450000, { name: 'Costs' });
const r_other = M.addRow('Data', '  Other',        5000);

// Formula rows — use returned row numbers for SUM ranges
M.addRow('Data', '  EBITDA', `=SUM(B${r_first}:B${r_other})`, { name: 'EBITDA' });

// Formula rows using named references (auto-resolved by HyperFormula)
M.addRow('Data', '  Margin', '=EBITDA/Revenue', { name: 'Margin' });

Build top-to-bottom: Names must be defined before any formula that references them. Define data rows first, then formulas. addBlank() and addSection() also return the A1 row number (useful for SUM range boundaries).

When to use which formula style:

Never use named expressions as range endpoints: =SUM(Revenue:OtherIncome) is undefined behavior.

Named References

The {name: 'Revenue'} option on addRow:

1. Registers a HyperFormula named expression (usable in any formula as Revenue)
2. Tracks the A1 row for internal cross-referencing
3. Exports as a proper Excel named range in.xlsx (via cell.names)

⚠️ Names are global across all sheets. Using {name: 'Revenue'} in two different sheets overwrites the first. Use unique, prefixed names for multi-sheet models: {name: 'Rev2024'}, {name: 'Rev2025'}.

Name validation rules:

• Must be valid Excel names: start with a letter or underscore, no spaces
• Cannot collide with Excel cell references: names like AC, PC, R1C1, A1 are rejected with a clear error
• Use descriptive names: AdjPC, TotalAC, CurrentAssets (not AC, PC, CA)

Cross-Sheet References (Data Sheets Only)

In addRow formulas, reference named cells on other data sheets with dot notation:

M.addRow('CashFlow', '  From Operations', '={PnL.NetIncome} + {PnL.Depreciation}');

This {Sheet.Name} syntax only works in addRow formulas, NOT in scenario output formulas. For scenarios, use named expressions (bare names) — they are global across sheets.

Scenario Sheets

The core feature — define inputs, scenarios, and output formulas declaratively:

M.addScenarioSheet('Scenarios', {
  inputs: [
    { name: 'GrowthRate', label: 'Revenue Growth %' },
    { name: 'CostCut',    label: 'Cost Reduction' },
  ],

  scenarios: [
    { label: 'Base Case',    values: {} },                              // all inputs = 0
    { label: 'Optimistic',   values: { GrowthRate: 0.10, CostCut: 50000 } },
    { label: 'Conservative', values: { GrowthRate: 0.03, CostCut: 20000 } },
  ],

  outputs: [
    // {InputName} → column-relative (B2, C2, D2...)
    // DataSheetName → named expression from Data sheet (fixed)
    // {PriorOutput} → column-relative ref to earlier output in this sheet
    { name: 'AdjRev',  label: 'Adj. Revenue',  format: 'number',
      formula: 'Revenue * (1 + {GrowthRate})' },
    { name: 'AdjCost', label: 'Adj. Costs',    format: 'number',
      formula: 'Costs + {CostCut}' },
    { name: 'AdjEBITDA', label: 'EBITDA',      format: 'number',
      formula: '{AdjRev} + {AdjCost}' },

    // Section separator
    { section: true, label: 'RATIOS' },

    // Ratio with conditional formatting thresholds
    { name: 'EBITDAm', label: 'EBITDA Margin', format: 'percent',
      formula: '{AdjEBITDA} / {AdjRev}',
      thresholds: { good: 0.15, bad: 0.08 } },

    // Inverted threshold (lower = better)
    { name: 'DebtEBITDA', label: 'Debt/EBITDA', format: 'ratio',
      formula: 'TotalDebt / {AdjEBITDA}',
      thresholds: { good: 2.5, bad: 4.0, invert: true } },
  ],
});

Do NOT call addSheet() before addScenarioSheet() — it creates the sheet internally. Only use addSheet() for data sheets. addScenarioSheet is a one-shot call. You cannot add rows to a scenario sheet after creation. Include all inputs, outputs, and sections in the config object. Every output you want to reference later MUST have a name property. Without it, {ThatOutput} in a subsequent formula will throw an error.

Formula Reference Resolution in Scenarios

Inside outputs[].formula, references are resolved as follows:

Important: Bare names (no {}) are HyperFormula named expressions — they resolve to a fixed cell. {Wrapped} names resolve to column-relative cells within the Scenarios sheet.

⚠️ Name collision rule: If a scenario output {name} matches a Data sheet named expression, {OutputName} will resolve to the Data sheet's fixed cell, not the scenario output's column-relative cell. Always use unique names for scenario outputs. Example: Data sheet has {name: 'EBITDA'}, scenario should use {name: 'AdjEBITDA'} — never both EBITDA.

Output Formats

Thresholds (Conditional Formatting)

thresholds: { good: 0.15, bad: 0.08 }             // Higher is better (green >= 0.15, red < 0.08)
thresholds: { good: 2.5, bad: 4.0, invert: true }  // Lower is better (green <= 2.5, red > 4.0)

Colors: green (#E2EFDA), amber (#FFF2CC), red (#FCE4EC).

Reading Computed Values

// From Data sheet (by named ref)
const ebitda = M.getValue('Data', 'EBITDA');

// From Scenarios (by scenario index, 0-based)
const baseEBITDA = M.getScenarioValue('Scenarios', 0, 'AdjEBITDA');
const optEBITDA  = M.getScenarioValue('Scenarios', 1, 'AdjEBITDA');

// Raw cell access (sheet, col 0-indexed, a1Row 1-indexed)
const val = M.getCellValue('Data', 1, 5); // col B, row 5

Use getValue() for data sheets, getScenarioValue() for scenario sheets. getValue() on a scenario sheet returns the first scenario's value (column B).

Error Handling

// Formula errors (division by zero, etc.) return CellError objects, not exceptions
const val = M.getValue('Data', 'Margin');
if (typeof val !== 'number' || !isFinite(val)) {
  console.error('Formula error:', val);
  // val might be: { type: 'DIV_BY_ZERO' }, { type: 'REF' }, { type: 'NAME' }, etc.
}

// To prevent #DIV/0! in formulas, guard with IF:
M.addRow('Data', '  Margin', '=IF(Revenue=0, 0, EBITDA/Revenue)', { name: 'Margin' });

// Reference errors throw immediately during addRow/addScenarioSheet
// → Always define named rows BEFORE formulas that reference them
// → Error messages include the row label and cell reference for easy debugging

Common causes of formula errors:

• Division by zero in ratios → guard with IF(denominator=0, 0, numerator/denominator)
• Misspelled named expression → throws immediately with clear error message
• Circular reference → throws immediately with row context

Console Output

M.printScenarios('Scenarios');

Prints a formatted table with emoji flags for threshold-based RAG status (🟢🟡🔴).

Export to.xlsx

await M.exportXlsx('output.xlsx', {
  creator: 'Agent Name',
  headerColor: '1B3A5C',  // Dark blue header background (ARGB hex, no #)
});

The exported file contains:

• Live formulas (not static values) — user can change inputs and see results update
• Named ranges on all named cells (visible in Excel's Name Manager)
• Conditional formatting on cells with thresholds (green/amber/red)
• Frozen panes — first column frozen on all sheets; header row also frozen on scenario sheets
• Input cells highlighted in light blue (#DAEEF3)
• Section headers in bold

Named ranges on scenario sheets point to the first scenario column (column B). They exist for cross-sheet references in Excel, not for selecting all scenarios.

Advanced Styling

For styling beyond what SheetModel provides, modify the file after export with ExcelJS:

import { createRequire } from 'module';
const require = createRequire(import.meta.url);
const ExcelJS = require('exceljs');

await M.exportXlsx('model.xlsx');

const wb = new ExcelJS.Workbook();
await wb.xlsx.readFile('model.xlsx');
const ws = wb.getWorksheet('Data');
// Custom column widths, borders, fills, page setup, etc.
ws.pageSetup = { orientation: 'landscape', fitToPage: true, fitToWidth: 1 };
ws.getColumn(2).numFmt = '$#,##0';
await wb.xlsx.writeFile('model.xlsx');

Complete Example: Financial Model with Scenarios

import { SheetModel } from '{baseDir}/lib/sheet-model.mjs';

const M = new SheetModel();
M.addSheet('Data');

// ── Balance Sheet ──
M.addSection('Data', 'BALANCE SHEET');
M.addBlank('Data');
const r1 = M.addRow('Data', '  Cash',           50000, { name: 'Cash' });
const r2 = M.addRow('Data', '  Receivables',   120000, { name: 'Receivables' });
const r3 = M.addRow('Data', '  Inventory',      30000);
M.addRow('Data', '  Current Assets', `=SUM(B${r1}:B${r3})`, { name: 'CurrentAssets' });

M.addBlank('Data');
const r4 = M.addRow('Data', '  Payables',       80000, { name: 'Payables' });
const r5 = M.addRow('Data', '  Short-term Debt', 40000, { name: 'STDebt' });
M.addRow('Data', '  Current Liabilities', `=SUM(B${r4}:B${r5})`, { name: 'CurrentLiab' });

M.addBlank('Data');
M.addRow('Data', '  Equity', 200000, { name: 'Equity' });
M.addRow('Data', '  Long-term Debt', 150000, { name: 'LTDebt' });

// ── P&L ──
M.addBlank('Data');
M.addSection('Data', 'INCOME STATEMENT');
M.addBlank('Data');
const p1 = M.addRow('Data', '  Revenue',         500000, { name: 'Revenue' });
const p2 = M.addRow('Data', '  COGS',           -200000);
const p3 = M.addRow('Data', '  Operating Exp',  -150000);
const p4 = M.addRow('Data', '  Depreciation',    -30000, { name: 'Depreciation' });
M.addRow('Data', '  Operating Income', `=SUM(B${p1}:B${p4})`, { name: 'OpIncome' });
M.addRow('Data', '  Interest Expense',  -15000, { name: 'IntExp' });
M.addRow('Data', '  Net Income', '=OpIncome+IntExp', { name: 'NetIncome' });

// ── Scenarios ──
M.addScenarioSheet('Analysis', {
  inputs: [
    { name: 'RevGrowth',  label: 'Revenue Growth' },
    { name: 'DebtPaydown', label: 'Debt Paydown' },
  ],
  scenarios: [
    { label: 'As-Is',       values: {} },
    { label: 'Growth 10%',  values: { RevGrowth: 0.10 } },
    { label: 'Deleverage',  values: { RevGrowth: 0.05, DebtPaydown: 50000 } },
  ],
  outputs: [
    { name: 'AdjRev',   label: 'Adj. Revenue',      format: 'number',
      formula: 'Revenue * (1 + {RevGrowth})' },
    { name: 'AdjEBITDA', label: 'EBITDA',            format: 'number',
      formula: '{AdjRev} + (Revenue - OpIncome + Depreciation) / Revenue * {AdjRev} * -1 + ABS(Depreciation)' },
    { name: 'TotalDebt', label: 'Total Debt',        format: 'number',
      formula: 'LTDebt + STDebt - {DebtPaydown}' },
    { name: 'NetDebt',   label: 'Net Debt',          format: 'number',
      formula: '{TotalDebt} - Cash' },

    { section: true, label: 'KEY RATIOS' },
    { name: 'CurrRatio', label: 'Current Ratio',     format: 'ratio',
      formula: 'CurrentAssets / CurrentLiab',
      thresholds: { good: 1.5, bad: 1.0 } },
    { name: 'DebtEBITDA', label: 'Debt/EBITDA',      format: 'ratio',
      formula: '{TotalDebt} / {AdjEBITDA}',
      thresholds: { good: 2.5, bad: 4.0, invert: true } },
    { name: 'ICR',  label: 'Interest Coverage',      format: 'ratio',
      formula: '{AdjEBITDA} / ABS(IntExp)',
      thresholds: { good: 3.0, bad: 1.5 } },
    { name: 'EBITDAm', label: 'EBITDA Margin',       format: 'percent',
      formula: '{AdjEBITDA} / {AdjRev}',
      thresholds: { good: 0.20, bad: 0.10 } },
    { name: 'ROE',  label: 'Return on Equity',       format: 'percent',
      formula: 'NetIncome / Equity',
      thresholds: { good: 0.12, bad: 0.05 } },
  ],
});

// Use
M.printScenarios('Analysis');
console.log('EBITDA (Growth):', M.getScenarioValue('Analysis', 1, 'AdjEBITDA'));
await M.exportXlsx('financial-model.xlsx');

Recipe: Loan Amortization

const M = new SheetModel();
M.addSheet('Loan');

M.addSection('Loan', 'LOAN PARAMETERS');
M.addBlank('Loan');
M.addRow('Loan', 'Principal',    500000, { name: 'Principal' });
M.addRow('Loan', 'Annual Rate',  0.05,   { name: 'AnnualRate' });
M.addRow('Loan', 'Years',        20,     { name: 'Years' });
M.addRow('Loan', 'Monthly Rate', '=AnnualRate/12', { name: 'MonthlyRate' });
M.addRow('Loan', 'Periods',      '=Years*12',      { name: 'Periods' });
M.addBlank('Loan');
// PMT returns negative (cash outflow) — negate for display
M.addRow('Loan', 'Monthly Payment', '=-PMT(MonthlyRate, Periods, Principal)', { name: 'Payment' });
M.addRow('Loan', 'Total Interest',  '=Payment*Periods - Principal', { name: 'TotalInterest' });

await M.exportXlsx('loan-model.xlsx');

Available Formulas

HyperFormula supports 395 built-in functions including:

• Math: SUM, AVERAGE, MIN, MAX, ABS, ROUND, CEILING, FLOOR, MOD, POWER, SQRT, LOG
• Financial: PMT, FV, NPER, PV, RATE, NPV, XNPV
• Logical: IF, IFS, AND, OR, NOT, SWITCH, IFERROR
• Lookup: VLOOKUP, HLOOKUP, INDEX, MATCH
• Statistical: COUNT, COUNTA, COUNTIF, SUMIF, SUMIFS, AVERAGEIF
• Text: CONCATENATE, LEFT, RIGHT, MID, LEN, TRIM, UPPER, LOWER, TEXT
• Date: DATE, YEAR, MONTH, DAY, TODAY, DATEDIF, EOMONTH

Full list: https://hyperformula.handsontable.com/guide/built-in-functions.html

ExcelJS Gotchas (Critical)

These bugs were discovered empirically and MUST be followed:

1. Named ranges: Use cell.names = ['Name'], NEVER definedNames.add() or addEx().

- add() tries to parse the name as a cell ref → crashes on names like InvFinCP - addEx() silently doesn't persist to the.xlsx file - SheetModel handles this automatically — don't use ExcelJS definedNames directly

1. Formula prefix: HyperFormula getCellFormula() returns "=SUM(...)" with leading =. ExcelJS expects {formula: 'SUM(...)'} without =. Double = causes #NAME? errors. SheetModel handles this automatically.
2. Formula language:.xlsx always stores formulas in English (SUM, ABS, IF). Excel translates to locale on display. Always write English function names.

Limitations

• Row limit: Practical limit ~5,000 rows. For larger datasets, use pandas/openpyxl via the xlsx skill.
• Single value column on data sheets: addRow writes to columns A (label) and B (value) only. For multi-period models (Year 1, Year 2, Year 3), use a scenario sheet where each "scenario" is a period, or use direct HyperFormula API (M.hf.setCellContents(...)) for additional columns.
• Data sheet formatting: All values in data sheets are formatted as integers (#,##0). For percentages, ratios, or decimals, compute them in a scenario sheet output with the appropriate format option, or post-process with ExcelJS (see Advanced Styling).
• No charts: HyperFormula/ExcelJS can't create Excel charts. Add charts manually or use a separate tool.
• No pivot tables: Use pandas for pivot-style analysis.
• Scenario columns: Maximum 25 scenarios per sheet (columns B–Z). For readability, keep to 10 or fewer.
• Named range naming: Names that match Excel cell/column references (e.g., AC, R1C1, A1) are rejected automatically. Use descriptive names.
• Data sheet formulas in.xlsx: The exported Excel formula is the original text, not extracted from HyperFormula. Stick to bare named expressions (e.g., Revenue, OpIncome) and A1 refs via template literals (e.g., ` =SUM(B${r1}:B${r3}) `).