Cash Flow Forecast . Agency OS

What it does

Builds a forward, timing-based cash forecast. It takes the bank balance you give it today and rolls it forward week by week, over a thirteen-week horizon by default or monthly if you want to see further out. The critical design choice: it models when money actually lands in your account, not when you sent the invoice. Service businesses get paid long after they bill, which is how a business can be profitable on paper and still miss payroll in week six. The forecast marks the low point in the horizon, flags any week the balance goes negative, and runs best and worst scenarios alongside the base case. If you triggered it with a decision, a hire or a purchase, it shows the forecast with and without that outflow so the answer to "can I afford it" is a line you can point at, not a feeling.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

> can I afford this hire?

> will I make payroll?

> cash flow forecast

> what happens if my biggest client pays a month late?

> should I make this purchase now or wait?

When to reach for it

you are about to commit to a new salary, a big purchase or a deposit and want to see the cash consequence first
payroll feels tight some months and you want to know exactly which week is the dangerous one
a large invoice is outstanding and you need to know what happens if it slips
revenue looks fine on the profit report but the bank balance keeps surprising you
you are heading into a known slow season and want to see whether the balance survives it

When NOT to use it

If you actually want	Use this instead
a backward look at last month: what came in, what went out, what changed	`close-month`
to chase the overdue invoices that are wrecking the forecast in the first place	`collections`
to know which clients actually make you money after delivery cost	`profit-by-client`
bookkeeping, reconciliation, tax filing or anything an accountant signs	your bookkeeper or accounting software, this skill is neither

Before you start

What you need	Why
Your actual bank balance as of today	the whole forecast rolls forward from this figure; if the workspace snapshot already holds a current cash number the skill starts there instead of asking again	Required
Expected inflows: each amount plus the date the money is expected to actually arrive	the arrival date, not the invoice date, is what the model runs on; an inflow with no expected date cannot be placed and gets excluded and flagged	Required
Outflows with their dates: payroll, rent, subscriptions, tax, ad spend, anything recurring or already committed	outflows are usually the predictable half; missing one silently flatters every week after it	Required
The decision you are weighing, if there is one: the hire's cost and start date, or the purchase amount and timing	this is what turns a generic forecast into a with-and-without comparison that answers your actual question	Optional
A current `metrics-snapshot.md` in the workspace	gives the skill a sourced starting balance and a place to append its forward-estimate block without touching the backward numbers	Optional

The honest refusal

If you cannot supply expected pay-dates for your inflows, the skill stops and tells you so. It will not invent a date, it will not assume every invoice pays on standard thirty-day terms, and it will not interpolate. A forecast built on invented timing is fiction wearing a spreadsheet, and this skill refuses to produce one. Bring the open invoices with the dates you genuinely expect them to pay, and it runs.

How it runs

Provenance firstOpens with a header stating the starting balance and where it came from (the workspace snapshot or you directly), the full inflow and outflow list it was given, and the horizon it will roll.
Roll the balance forwardFor each week, closing balance equals opening balance plus the inflows landing that week minus the outflows due that week. The closing figure carries into the next week, all the way down the horizon.
Mark the low pointFinds the lowest closing balance in the run and names the week it happens. Any week that closes negative is flagged explicitly, because that is the week the plan breaks.
Run the scenariosAlongside the base case it runs a worst case, for example your largest expected inflow slipping by a month, and a best case. Each scenario is labeled as an estimate and carries its assumption in one line, so you can see exactly what would have to be true.
Answer the decisionIf a hire or a purchase triggered the run, the forecast appears twice: with the new outflow and without it. The difference between the two low points is the real cost of saying yes.
Give the verdict in plain languageEnds with a sentence a human can act on: you clear payroll every week, the tight point is this week at this amount, and here is the single event that would push you negative.

What you get

A provenance header: starting balance and its source, the inflow and outflow list, the horizon
The week-by-week closing-balance table, or the compact version: starting balance, the low point and its week, any negative weeks, the ending balance
Best and worst scenario lines, each labeled as an estimate with its assumption stated
The with-and-without comparison when a specific decision triggered the run
A plain-language verdict naming the tight week and what would break it
A clearly labeled forward-estimate block appended to the workspace snapshot, leaving the backward numbers untouched

Honest limits

Read this before you rely on it

An expected payment with no expected date is excluded from the forecast and flagged, never guessed. The skill does not assume thirty-day terms or any other default. If the excluded amount is large, it tells you how much the picture improves once you pin the date down.
Every forward number is a labeled estimate, not a fact. The scenarios state their assumptions in one line each so you can judge them, and you should.
It starts from the bank balance you give it. If that figure is stale or wrong, every week downstream inherits the error. Check the balance the same day you run the forecast.
It is not bookkeeping and not accounting software. It will not reconcile transactions, categorise spending or produce anything your accountant files.
The forecast is a snapshot of what you told it. When an invoice pays early, a client churns or a cost appears, the old run is out of date. Rerun it rather than trusting last month's.

Where people go wrong

The mistake	Do this instead
Feeding it invoice dates instead of expected pay-dates	The date you billed is history. Use the date the money will actually land, and when you honestly do not know, say so and let the skill exclude it rather than smuggling in a guess.
Reading only the ending balance	A run can end comfortably and still go negative in the middle. The low point and its week are the numbers that matter; the ending balance is trivia.
Treating profit as cash	A profitable month with slow-paying clients can still miss payroll. Suppose you invoice 30,000 in a month but 25,000 of it lands after the payroll date: the profit report smiles while the account runs dry. That timing gap is the entire reason this skill exists.
Running it once and filing it away	Rerun it whenever the inputs change: an invoice pays, a client signs, a cost lands. A rolling forecast only works if it actually rolls.
Ignoring the worst case because the base case clears	The worst case is usually one late payment away. If a single slipped invoice sends a week negative, you do not have a safe plan, you have a lucky one.
Committing to the hire before running the with-and-without view	Ask the question with the decision in the run. Seeing the low point drop from a comfortable cushion to nearly nothing is a very different conversation from a gut feel about affordability.

Worth knowing

The numbers in any example here are invented for illustration. Your forecast runs entirely on your own balance, your own invoices and your own commitments, which is exactly why it can only ever be as honest as the dates you feed it.