原始内容
Modelling Team — Claude Code Skill
A Claude Code skill that builds polished Excel financial models via a three-agent team: Architect (Opus) designs the blueprint, Coder (Sonnet) writes Python/openpyxl, Challenger (Opus) stress-tests formulas. Supports light (single-tab) and heavy (multi-tab with Dashboard, scenario selector, INDEX-driven scenarios) modes.
How It Works
User request
|
v
Step 0: Clarify scope (heavy or light?)
|
v
Step 1: MODEL ARCHITECT (Opus)
Designs full model blueprint: tabs, rows, formulas, scenarios
|
v
Step 2: MODEL CODER (Sonnet)
Translates blueprint into Python/openpyxl script, runs it
|
v
Step 3: MODEL CHALLENGER (Opus)
Reviews .xlsx: formula audit, scenario validation, design check
|
v
PASS -> deliver | FAIL -> fix and re-check (up to 3 iterations)
Features
- Two modes: Light (single tab, assumptions at top) and Heavy (Dashboard + Assumptions + calculation tabs)
- Scenario selector (heavy mode): Dashboard cell drives Low/Base/High via
INDEX()— no hardcoded scenario columns - Professional design language: Navy/blue palette, Arial font, color-coded cells (blue = inputs, black = formulas, green = cross-sheet links)
- Manual extensibility: All formulas use proper
$-notation (mixed/absolute references) so you can drag formulas to new columns - IRR/NPV/PBP/DPBP: Correctly constructed cash flow streams with negative initial investment
- Zero formula errors: Challenger validates every formula before delivery
Installation
From .skill file
claude install-skill modelling-team.skill
From this repo
Copy the SKILL.md and references/ directory into your Claude Code skills folder:
~/.claude/skills/modelling-team/
├── SKILL.md
└── references/
├── architect-prompt.md
├── coder-prompt.md
└── challenger-prompt.md
Usage
Just ask Claude Code to build a model:
"Build me a DCF model for a coffee shop: $200K revenue, 5% growth, 30% EBITDA margin, 10% discount rate, 4x terminal multiple, 5-year horizon."
"Build me a heavy model for a rental portfolio with 3 properties, scenario analysis for rent growth and vacancy, and a dashboard with IRR and cash-on-cash return."
Claude will ask whether you want a light or heavy model, then run the three-agent workflow automatically.
File Structure
modelling-team/
├── SKILL.md # Main skill definition
├── references/
│ ├── architect-prompt.md # Blueprint design agent prompt
│ ├── coder-prompt.md # Python/openpyxl coding agent prompt
│ └── challenger-prompt.md # Validation/review agent prompt
└── modelling-team.skill # Packaged distributable
Design Language
The skill enforces a consistent visual style across all models:
| Element | Color | Usage |
|---|---|---|
Blue font (0000FF) |
Inputs | Editable assumptions |
Black font (000000) |
Formulas | All calculations |
Green font (008000) |
Cross-sheet | Links between tabs |
Navy fill (1F3864) |
Title rows | Model/section titles |
Medium blue fill (2F5496) |
Section headers | Sub-sections |
Light blue fill (EBF3FB) |
Input cells | Background for editable cells |
Light green fill (E2EFDA) |
Totals/KPIs | Bold summary rows |
Requirements
- Claude Code with access to Opus and Sonnet models
- Python 3 with
openpyxlinstalled - LibreOffice (for formula recalculation via the xlsx skill)
License
MIT