A financial question usually arrives in plain English. Can we afford another hire? What is this company worth? Does this rental work at the asking price?
A useful answer needs more than a number. You need to see where the inputs came from, which assumptions matter, and what happens when those assumptions change. That is why we built Moolaski: an open-source financial analyst skill for AI agents, with a working Excel model as its deliverable.
From a question to a model you can inspect
Moolaski gives a compatible agent a structured way to handle financial analysis. The skill asks the agent to establish the decision, gather the inputs, choose an appropriate modeling guide, and build a workbook with live formulas. It then requires recalculation, reconciliation checks, and an independent check of the headline result before presenting the answer.
The intended output is a conclusion you can examine: the result, the assumptions driving it, and the point at which the decision changes. The workbook keeps that reasoning available for the next question.
This is a skill package: instructions, modeling references, and a Python workbook checker. Its results depend on the agent, the tools available to it, and the quality of the inputs. Installing it gives your agent a workflow to follow; it does not guarantee that every analysis will be correct.
What you get back
An answer tied to the decision. The skill directs the agent to put its conclusion in the response and on the workbook’s Cover sheet, alongside the main sensitivities, the break-even, and any assumptions that need confirmation.
An editable Excel workbook. The deliverable is an .xlsx file containing formulas. Inputs are distinguished from calculations, and downstream results reference those inputs. Changing a rent assumption, a growth rate, or a financing term should flow through the model when it recalculates.
Checks you can review. The modeling conventions call for a Checks sheet and a master error flag. Depending on the model, checks can reconcile the balance sheet, funding sources and uses, debt movements, or distributions to investors. The agent is instructed to resolve failing checks and report warnings separately.
Those checks establish internal consistency. A workbook can reconcile perfectly while using an unrealistic sales forecast or an incorrect lease assumption. Reviewing the inputs remains part of the analysis.
Start with a rental question
Consider an illustrative request:
Analyze a rental offered at $320,000 with expected rent of $2,400 a month and a 25% down payment. Build a working Excel model, state the assumptions you need, and show the rent and purchase price at which cash flow breaks even.
The supplied figures establish only the starting point. Annual scheduled rent is $28,800, the down payment is $80,000, and the initial loan would be $240,000 before any financed costs. Rent divided by price is 9%, but that is a gross rental yield: it says nothing yet about operating expenses, debt service, vacancy, or purchase costs.
The analysis still needs financing terms, taxes, insurance, repairs, management, vacancy, and reserves. A return calculation over a holding period also needs an exit assumption and selling costs. Moolaski’s workflow tells the agent to identify missing inputs, source public information when tools permit, and label assumptions it makes.
The useful follow-up is specific: reduce rent by 5%, increase maintenance, or test a lower purchase price. The model should recalculate those changes while preserving the base case. A conclusion is more useful when you can see how little—or how much—has to change before it reverses.
These figures illustrate how to frame the work. They are not a completed property appraisal or a forecast of investment returns.
Where Moolaski fits
The current repository includes guides across the following areas:
| Question | Modeling workflow |
|---|---|
| What is a business worth? | Discounted cash flow, comparable-company and transaction analysis, and sensitivity ranges. |
| Will the company have enough cash? | Three-statement forecasts and 13-week cash-flow models. |
| How is the operating plan performing? | Driver-based budgets, actual-versus-budget analysis, and updated forecasts. |
| Does an acquisition work? | Leveraged buyouts, debt schedules, and merger accretion/dilution. |
| Should we fund this project? | Capital budgeting, infrastructure and renewable-energy finance, and debt sizing. |
| Does this property deal work? | Rental acquisitions, commercial properties, development, flips, and partner distributions. |
| Who owns what after a financing? | Startup cap tables, convertible instruments, and exit distributions. |
| Can we trust an existing workbook? | Reviews of formulas, hardcodes, circularity, model checks, and assumptions. |
These are modeling playbooks rather than prebuilt connections to accounting systems or market-data feeds. Bank, insurance, and fund-level models are explicitly outside the skill’s tested playbook; the agent is instructed to disclose that boundary when adapting a nearby guide.
Install it and give it a decision
Install from the public repository:
npx skills add looskis/moolaski
The installed skill is named moolaski-financial-analyst. The repository documents support through the skills CLI for agents including Claude Code, Codex, Cursor, Gemini CLI, and GitHub Copilot.
Start with a question, the relevant files, the period to model, and the decision the result should support. For example:
Use Moolaski to build a 13-week cash forecast from these receivables, payables, payroll dates, and opening cash balance. Identify the lowest cash point, show a slower-collections case, and list any assumptions I need to confirm.
When a live spreadsheet tool is available, the skill directs the agent to build and recalculate there. Otherwise, it can create the workbook in code. The documented Python path uses openpyxl to write formulas and a bundled checker using the formulas library to recalculate and inspect the workbook without a spreadsheet application installed. The skill also requires a separate check of the headline result.
For an existing model, ask for a review first. That lets the agent examine its structure and findings before extending a workbook whose errors might otherwise carry forward.
Keep the data boundary explicit
Moolaski does not supply its own inference service, private deployment, or financial data subscription. The agent and its configured tools determine where files and prompts are processed. A hosted model may receive the financial information included in its context; spreadsheet integrations and research tools introduce their own data paths.
For confidential forecasts, deal materials, or client workbooks, choose that environment deliberately. See our guide to the private AI data boundary for the questions to ask about inference, storage, logs, and connected services.
Moolaski builds and explains models. Use the workbook, its sources, and its assumptions as material for review before making a financial decision.
Try it on your next financial question
Bring a real decision and the inputs you have. Ask for the workbook, the main sensitivities, and the break-even. Then change an assumption and inspect what follows.
Moolaski is available on GitHub under the MIT license. If you want this workflow connected to your team’s reporting process, talk with Looski.
Frequently asked questions
What does Moolaski install?
The skills CLI installs moolaski-financial-analyst: a skill with financial modeling instructions, reference guides, and a Python workbook checker. You use it through a compatible AI agent.
Do I need Excel installed?
A live spreadsheet tool can build and recalculate the model when available. The documented alternative writes an .xlsx file with Python and uses the bundled checker with openpyxl and formulas to recalculate and inspect it without a spreadsheet app.
Does Moolaski keep financial data on my computer?
That depends on the agent, model, and tools you configure. Moolaski does not itself provide a local inference environment or change a hosted provider’s data handling.
What does a passing Checks sheet prove?
It shows that the implemented reconciliation tests passed. It does not establish that the inputs are accurate, the forecast is realistic, or the investment is suitable. Review sources and assumptions, and independently check material results.