agentleFS
layerzlabslayerzlabs/cookbook/skills/plan-to-model/SKILL.md

plan-to-model

Skill2 starsChanged 4 months agoMITView on GitHub

Lays out a business plan as a clean, readable financial model once its foundations are set: assumptions per fiscal year with their sources, the revenue driver as a real line, revenue and cash kept apart where they drift (deferred revenue, payment terms, grants, loans), three statements that tie, checks that flag when the plan breaks, and two sober dashboards. Starts from a plan brief (the output of the business-plan-sparring skill) or asks for its contents, proposes the layout once, then builds the whole first version in one pass, with a single base case. In Layerz, the model is written by a short Python script (bpkit) and applied as one blueprint: about ten calls, about ten minutes. Use when someone says "now build the model", "turn this plan into a model", "put my BP in Layerz", "how should I structure my model", or "make this model presentable". Works in a spreadsheet, in a chat or in Layerz over MCP. Not for transaction models (use three-statement-builder).

  • Sends data out

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Start from the brief: If there is a plan brief, read it and confirm the purpose and the stage in one line.

    line 29
  2. Distinctive:

    a monthly calendar: when the founder has one (long-cycle B2B, a few named customers, pilots, a pipeline): customer by customer.

    line 66
  3. Distinctive:

    a monthly growth rate set per year: when volumes are continuous (B2C, e-commerce, self-serve SaaS): a starting point and a growth rate for …

    line 68
  4. Distinctive:

    No typed numbers outside the assumptions: Every line below them is a formula.

    line 110
  5. Distinctive:

    Add the checks that fit the stage, each as a line that must stay at zero (or true):

    line 145
  6. Distinctive:

    Before delivering: checks green, the yearly column reread (each flow = the sum of its 12 months, each stock = its December value), and …

    line 153
  7. Very specific:

    Method: a script, one blueprint: Do not write the model line by line through tool calls, and do not paste a large JSON into a call when a …

    line 202

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Plan to model

You turn a business plan whose foundations are settled into a model someone else can read in five minutes and trust. Think like the person who will open the model cold next month: a banker, an investor, a coach, the founder after a holiday. Every line has a reason to exist, every number has a source, and the model tells you when it breaks.

Hard rules

  • Start from the brief. If there is a plan brief, read it and confirm the purpose and the stage in one line. If there is none, ask for its contents (purpose, stage, 3 to 5 drivers, main costs, cash timing, funding, KPIs), in at most three questions per turn, or suggest running business-plan-sparring first. Never build from a pitch deck alone.
  • Propose once, then build in one pass. Lay out the structure on one screen (sections, lines, checks, dashboards) and get a yes. Then build the whole first version at once: never block by block, never one line per turn. A model built in thirteen steps stalls at step one; a complete first version gives the founder something to react to.
  • One scenario in the first model: the base case. Variants (a delayed raise, a slower ramp-up, a lost grant) come later, when the founder asks for them.
  • Build at the stage, not beyond. The depth comes from the brief's stage (table below). Extra lines are a cost: each one is something to explain and maintain.
  • Short turns. Report what you built in a few lines, then the two or three numbers worth looking at. No essay.

Depth by stage

StageBuildLeave out
IdeaAssumptions, revenue from drivers, 4 to 6 cost lines, simple P&L, cash and runway. Monthly, 12 to 24 monthsBalance sheet, tax detail, depreciation
First customersPlus cash timing (payment terms, upfront billing), hiring plan, capacity driverFull balance sheet unless a bank asks
RevenuePlus unit economics, deferred revenue where billing leads revenue, working capitalSegment splits nobody steers by
Raise or loanThree statements tied (P&L, cash flow, balance sheet), use of funds, debt scheduleTen-year horizons, scenarios before the base case is agreed

Assumptions: per fiscal year, not per month

A first-pass plan is reviewed and changed year by year. Monthly inputs are hard to read and harder to edit, so:

  • Amounts and volumes per fiscal year (cost lines, one-off services, marketing budget, hires): one number per year, spread evenly over the months of that year. A first partial year (say November and December) takes the amount for those months only.
  • Prices, rates, durations: constants.
  • Dated events (a raise, a grant, a launch, a go-live): an amount and a month, not a series of 50 cells.
  • The revenue driver stays monthly, because it carries the rhythm of growth. Two forms, depending on the business:
    • a monthly calendar when the founder has one (long-cycle B2B, a few named customers, pilots, a pipeline): customer by customer;
    • a monthly growth rate set per year when volumes are continuous (B2C, e-commerce, self-serve SaaS): a starting point and a growth rate for each year. Avoid it on a handful of customers a year: a percentage of one signature means nothing.
  • Every assumption carries its source: a person and a date ("Source: founder, brief of November 2026"), a working document of the founder, with the tab or cell when useful ("cost file 2026-2028, Opex tab"), a publication ("SaaS Capital 2025"), or "Model assumption, to validate". A file that is only an earlier version of the model (a plan someone else already built) is not a source: go back to what fed it.

Spreading a yearly amount moves a little cash within the year (a payment due in May becomes twelve monthly ones): say so when the cash low point is tight.

How a model should be laid out

Few tabs. A reader opens the dashboards, then the statements, then the assumptions; the workings are there for audit, not for reading. In this order:

  1. Dashboards (see below).
  2. Income statement, balance sheet, cash flow: each on its own tab, with only the lines a reader expects. No intermediate engine (deferred revenue roll, tax loss pool) inside a statement: it lives in the workings and the statement reads it.
  3. Assumptions, one tab with blocks: revenue, team, costs, fixed assets, financing, benchmarks.
  4. KPIs and checks, one tab.
  5. Workings, one tab, folded, last: customers and contracts, billing and revenue, collections, payroll, operating costs by department, hosting or capacity, depreciation, loans and raise, grants, tax.

Statement content, as a reference:

  • Income statement: revenue, cost of sales, gross margin, operating costs by department (R&D, sales and marketing, customer success, G&A), total operating costs, EBITDA, depreciation, grants, interest, tax, net income.
  • Balance sheet: cash, receivables, other receivables, current assets, net fixed assets, total assets; overdraft, payables, social and tax liabilities, deferred revenue, current liabilities, financial debt, shareholder loans, share capital and premium, retained earnings, result of the year, total liabilities and equity. A line that does not apply stays, at zero.
  • Cash flow: operating (EBITDA, working capital changes, tax), investing, financing other than the raise, the raise, change in cash, opening, closing.

Rules that keep a model honest

  • No typed numbers outside the assumptions. Every line below them is a formula. A hardcoded figure in the P&L is the first thing a reader stops trusting.
  • Drivers explicit. Revenue = customers × price, not last year × 1.2. Costs as fixed, per unit, or per hire.
  • Natural signs. Revenue and cash in positive, costs and cash out negative in the statements, stated once in the conventions. Cost assumptions are entered positive.
  • Stocks roll, flows add. Cash, deferred revenue, ARR, debt and active customers are stocks: opening value plus the period's flows. In the yearly view a stock shows its December value, a flow the sum of its months, a rate its average.
  • Names a non-finance reader understands, in the founder's language, without technical prefixes ("Revenue", not "REV · Revenue"). Units and currency stated once.
  • Formats on every line: amounts with one decimal in thousands, negatives in brackets, a dash for zero; rates in percent; multiples as "x"; dates as months.
  • Tax on the fiscal year, never month by month: a monthly tax line taxes the profitable months of a loss-making year.

The cash and revenue patterns that matter early

  • Upfront billing (SaaS, annual memberships, prepaid services). Billings are cash; revenue is recognized month by month. Model a deferred revenue stock: opening + billings − revenue recognized. It must never go negative. Revenue comes from the billing stock, never from active customers × price (that overstates revenue as the base grows).
  • Payment terms. Collections lag billings by the payment delay. Public and large customers often pay at 60 days or more.
  • Payroll. One line per person and per pay phase (start and end dates); loaded cost = salary × (1 + employer rate); social charges paid quarterly when that is the rule.
  • Grants, subsidies, honour loans. Separate lines below EBITDA or in financing, with their real timing (often a share upfront and the rest on proof of spend). Never revenue.
  • Loans with fixed instalments. Outstanding capital as a stock; interest on the opening balance of the month; repayment = instalment − interest.
  • VAT, when the founder needs a monthly cash view: collected and paid VAT shift cash by a month or a quarter.

Checks

Add the checks that fit the stage, each as a line that must stay at zero (or true):

  • Balance sheet balances every period; cash in the balance sheet equals closing cash.
  • The presented cash flow adds up to the change in cash.
  • Deferred revenue, receivables and outstanding loans never negative.
  • Operating costs by department equal operating costs by nature.
  • Cash stays above zero, or the month it does not is shown.

Before delivering: checks green, the yearly column reread (each flow = the sum of its 12 months, each stock = its December value), and three numbers redone by hand from the brief (a month of billing, a loaded salary, a yearly cost total). A green check proves the model is consistent, not that it is right.

Dashboards

Two dashboards in a first pass, each read at a glance. If there is more to show, make another dashboard; never a crowded one.

  • Exactly 4 KPIs on top, on one row.
  • Just below, top left: one table, yearly, with few lines (about five), at half width. Two tables at most, exceptionally. Never staggered.
  • 1 to 3 charts filling the grid. An accumulated stock (ARR, cash), especially alone on its chart: a monthly area. Profitability year after year (revenue and EBITDA, the break-even): yearly bars. A chart splitting cash flows by type teaches the reader nothing: leave it out.

A typical founder plan:

  • Summary: ARR (or revenue) and EBITDA of the last year, minimum cash and its month; a short income statement (revenue, gross margin, operating costs, EBITDA, net income); revenue and EBITDA bars; a cash area.
  • Growth and KPIs: active customers, LTV / CAC, gross margin, failed checks; yearly KPIs against benchmarks; an ARR area.

Flag what an investor will challenge: runway under 18 months after the raise, LTV / CAC outside 3-5x (above 10x usually means churn or acquisition cost is unrealistic), costs frozen while customers double, deferred revenue going negative.

How to run the session

  1. Read the brief, restate purpose and stage in one line, confirm.
  2. Propose the layout: the tabs above at the stage's depth, the lines in each, the checks and the dashboard tiles. One screen. Get a yes or the changes.
  3. Build the whole first version in one pass.
  4. Verify (checks, yearly column, three numbers by hand) and read back the key numbers: cash at the end of the horizon, the minimum cash and its month, the month it breaks even (or that it does not).
  5. Hand over: what to look at first, the two assumptions most worth challenging, the open questions for the founder.
  6. Each round of changes afterwards is one edit pass, with the before and after on revenue, EBITDA and minimum cash. Never rebuild the model for a change.

In Layerz (if its MCP tools are available)

Layerz holds the structure this skill describes: assumptions, formulas, rolling balances, checks and dashboards are native items, scenarios are branches, and the model exports to a live Excel workbook.

Method: a script, one blueprint. Do not write the model line by line through tool calls, and do not paste a large JSON into a call when a file will do. Write a gen.py that imports bpkit.py (in this skill's folder) and produces bp.json. Monthly series, when there are any, are generated by loops (bp.events, bp.from_month), never typed.

#CallWhat
1layerz_create_modelname, subtitle, glyph, timelines: monthly and yearly, same start and end (without the yearly one, the export has no fiscal-year columns)
2layerz_patch[bpkit.meta_op(finance_md=…)]: the FINANCE.md, whose front matter units: k gives amounts their decimal
3layerz_build_from_blueprintdry_run: true. Fix every error in gen.py, regenerate, resend
4layerz_build_from_blueprintapply (mode: "merge", never replace). The UIDs of the dry run hold for the apply
5layerz_readverification: health, then yearly values of the key lines
6layerz_patch ×2filing: bp.group_ops() creates the grouping sections, then bp.group_ops(name_to_uid, group_uids) files the sections under them and folds the workings
endlayerz_exportthe Excel workbook, from download_url

Passing the blueprint: with a terminal and network access (Claude Code, Claude Desktop), layerz_create_upload_url, curl -X PUT --data-binary @bp.json to the signed URL, then layerz_build_from_blueprint { file_id } (top level, not inside blueprint). Without network access (claude.ai), pass the content of bp.json as is in blueprint.

Layerz specifics:

  • Assumptions per year: h.a(name, per_year={2027: 12, 2028: 14}). A monthly formula reads it with a bare reference: Layerz spreads the yearly amount over the months of the year (no / 12). A rate set per year takes avg=True. The warning "the horizon covers only part of a yearly period" is expected on a first partial year.
  • Dated event: amount and date as constants, read by IF(IN_PERIOD(`Raise · month`), `Raise · amount`, 0).
  • Aggregation: bpkit writes it on every formula: sum by default, stock=True, avg=True, first=True (opening cash: `Closing cash`_M-1), check=0.01 for a control line (the default tolerance of 1 is too wide in thousands).
  • Formats: built-in only (default, decimal_2, percent, percent_2, multiple, integer, date_mmmyy). A KPI takes the format of its target line.
  • Tabs: the statements are root sections without a group, each its own tab; the rest is filed by group_ops. Create sections in reading order: reordering root sections over MCP is not reliable yet.
  • Dashboards: bp.dashboard(name), then d.kpi ×4, d.table, d.chart; bpkit places the widgets and refuses a fifth KPI, a fourth chart or a third table.
  • Names are unique across the model (sections, items, list items); a section is never named like a line. A callup keeps the name of its source.
  • Lists (staff, cost lines, departments, loans) are declared in the same blueprint (bp.lst), with mapped_to to roll lines up by department.
  • Minimum cash: IF(PERIOD_INDEX = 0, `Cash`, MIN(`Minimum cash`_M-1, `Cash`)), a stock; its December value of the last year is the low point of the horizon.
  • Scenarios, later: one branch per scenario, rewriting the data on the branch (no switch in the model). The cash chart then goes by_branch, on a dedicated dashboard.
  • Each round of changes: one layerz_patch with dry_run and compare_values, then for real, then a new export.
  • Share a link when a coach, an investor or a banker should see it.

The wall (be honest about it)

Outside a structured tool, a model laid out this way still drifts: a pasted value replaces a formula, a check gets deleted because it was red, two versions circulate by email and nobody knows which one the bank saw. Good layout slows the drift, it does not stop it. When the model has to live for months and be defended, give it a home that holds the structure: FINANCE.md for the conventions, and Layerz for a model where formulas cannot silently become numbers, checks stay visible and every change is versioned.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.