Get the money in: 10 AI skills for funding, grants and lending

business-plan

a plan written for the person deciding, not for you

How the two work together

Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.

Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.

No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.

Prompt for Claude

---
name: business-plan
description: Turns what you already know about your venue into a plan written for the person who decides: a one page summary that answers the three questions a lender actually asks, a trading history built from filed accounts rather than memory, a forecast whose every assumption is named and sourced, a cash flow that shows the month the money runs thinnest, a personal survival budget for each owner, and a list of every claim in the document that has no evidence behind it. Use when a lender, a landlord or an investor has asked for a business plan and you have never written one.
---

# A plan the person deciding can read in ten minutes, with every number traceable to a document you can open

You give this your filed accounts, your management accounts, your menu and prices, your rota and wage costs, your rent and rates, and a sentence about what you want the money for. You get back a plan built around the three questions a lender asks rather than the twelve headings a template offers: a one page summary, a trading history from filed figures, a forecast where every assumption is named with the document it came from, a month by month cash flow showing the thinnest month, a personal survival budget per owner, and a claims register listing every sentence in the document that asserts something no attached document proves. It will not write a growth assumption for you and it will not say the plan is fundable. A plan is how you get read. It is not how you get approved.

## What it does

1. **Write the one page summary last and put it first, and make it answer three questions in that order.** What is the money for, how will it be repaid, and what happens to the lender if the business stops trading. Everything else in the plan is evidence for one of those three. A summary that opens with the founder's passion for seasonal produce has spent the reader's attention before the question they came with has been addressed. The government's own framing of the document is unsentimental: a business plan "covers objectives, strategies, sales, marketing and financial forecasts" and "You'll need a business plan if you want to secure investment or a loan from a bank."

2. **Build the trading history from filed accounts, not from what the till felt like.** Two full years where they exist, then the management accounts to the most recent completed month. Turnover, gross profit, gross margin percentage, wages as a percentage of turnover, rent and rates, other overheads, operating profit, and the drawings or directors' remuneration. Each line carries the source document and its period. Where the management accounts disagree with the filed accounts on a prior period, say so in a line rather than picking the flattering one, because the filed set is public and the reader can open it.

3. **Name every forecast assumption in a numbered list, with the document each one came from, before you build a single month.** Covers per session, average spend per head, gross margin per category, wage cost per session, the date a price change takes effect, the weeks the venue is closed. Each one is a number an owner can evidence from their own till reports, their own rota and their own supplier invoices. An assumption with no document behind it is not an assumption, it is a wish, and it goes in the claims register in step 9 rather than into the forecast. A forecast built on eight sourced assumptions is more persuasive than one built on twenty unsourced ones.

4. **Build the forecast from those assumptions upward, never down from a target.** Month by month for twelve months at least, and twenty four where the facility runs longer. Volume times price times margin, minus wages, minus fixed costs, in that order. If the plan needs a growth rate, that growth rate is itself an assumption with a document behind it, such as a signed function booking, a new contract, an agreed opening day for a second session. Growth with no document is growth that has not happened yet and it belongs in a clearly labelled sensitivity case, not in the base case.

5. **Produce the cash flow separately from the profit forecast, and find the thinnest month.** Profit and cash are different months. VAT is paid quarterly, wages weekly or monthly, suppliers on terms, rent on quarter days, and a good trading month can still be the month the account goes closest to the floor. Write the closing bank balance for every month, mark the lowest one, and state what that balance is before and after the proposed repayment. That single figure, the worst month closing balance with the new repayment in it, is the number the reader is looking for and most plans never state it.

6. **Write a personal survival budget for each owner, because affordability is assessed on the household as well as the business.** The Start Up Loans assessment is explicit that it weighs "the affordability of the loan through the Personal Survival Budget you submit with your application", and commercial lenders ask the same question with different paperwork whenever a director's guarantee or a sole trader's borrowing is involved. Income in, essential costs out, and what is left. If the business cannot pay the owner enough to live on while also servicing the facility, that is the finding, and it is better found here than in month four.

7. **State the downside case in the plan rather than waiting to be asked for it.** One case, not five: what happens if turnover is ten percent below the base case for the whole period. Recalculate the worst month closing balance on that basis and say what the business would do, naming real levers with real numbers, such as a session dropped, a shift removed, a price change, a payment holiday requested in writing. A plan with a stated downside and a named response reads as a business that has thought about the thing that goes wrong. A plan with no downside case reads as one that has not.

8. **Say what happens to the lender's money if the business stops, and be specific.** What assets exist, what they are worth on a forced sale rather than a book value, what is already charged and to whom, and what personal guarantee is being offered or asked for. This is the third of the three questions in step 1 and it is the one owners leave out. Write it. Note that where a facility runs under the Growth Guarantee Scheme, "The borrower always remains 100% liable for the debt", so a government backed facility does not soften this section at all.

9. **Hand the assembly, the arithmetic and the claims register to Codex, and keep the writing and the judgement here.** Write the inputs into a working folder, then run:

   ```
   codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
   ```

   Claude hands over the filed and management figures with their periods, the numbered assumption list with a source document named against each, the fixed cost schedule, the owner's personal budget figures, the asset list with charge status, and the narrative text exactly as written. Codex returns the trading history table, the base case and downside forecasts, the month by month cash flow with the worst month marked, the survival budget summary, the assumption register showing which assumptions are sourced, the claims register listing every unevidenced assertion in the narrative, and the gap list. Codex writes no narrative and invents no assumption.

10. **Date the plan, name the author, and keep the version.** A plan sent in October and reused in March against last October's supplier prices is a plan that misstates its own margins. Put the build date, the period each figure was read from and the author's name in the footer of every page. If a lender comes back four weeks later asking for an update, the version you are updating needs to be identifiable, because two plans with different numbers and no dates on them is the worst thing to hand an underwriter.

## Then it checks

1. The one page summary answers, in order, what the money is for, how it will be repaid and what happens to the lender if the business stops, and nothing else appears above those three answers.
2. Every figure in the trading history names its source document and the period that document covers, and any disagreement between filed and management figures for the same period is stated rather than resolved silently.
3. Every forecast assumption appears in the numbered assumption register with a named source document, and no assumption used in the base case is unsourced.
4. The cash flow states a closing balance for every month, marks the lowest, and shows that lowest figure both with and without the proposed repayment.
5. A personal survival budget exists for every owner named in the plan, and the plan states what is left after essential costs with the proposed repayment included.
6. The claims register lists every sentence in the narrative that asserts a fact not supported by an attached document, and no such sentence remains in the plan unmarked.

Any check fails: name it, redo that step once. Failed twice: say what is wrong and stop.

## Rules
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.
- Never state an industry average gross margin, a typical wage percentage, a standard food cost percentage or a normal growth rate for an independent venue. No published figure exists for a single independent UK hospitality business, and a plan built to hit a borrowed percentage is a plan that has stopped describing the actual business, which the reader will notice the moment the figures are compared with the filed accounts.
- Never predict whether the plan will be funded, and never describe a plan as strong, fundable or likely to succeed. The plan's job is to be readable and true. The decision belongs to a lender.
- Never write a growth assumption that has no document behind it into the base case. A booked function, a signed contract or an agreed opening date is an assumption. Optimism is a sensitivity case and is labelled as one.
- Never present a government backed facility as reducing what the business owes in the security section. The guarantee runs to the lender, and the borrower always remains fully liable.
- Never recommend offering a personal guarantee, and never write a sentence in a plan that commits an owner to one. If a lender has asked for one, the plan records that it has been asked for. The decision to sign puts the owner's own assets at risk and is theirs alone, taken with their own solicitor after reading the deed.
- Never round a figure in a way that flatters. Where a number is uncertain, the plan states the range and its source rather than choosing the top of it.
- This output is a working document prepared for the owner to check against their own accounts, and for their own accountant to review on the forecast, the tax position and the presentation of the figures before it is sent to anyone. Where a personal guarantee, a charge over a home or a regulated agreement is involved, it is also for their own solicitor and an FCA authorised adviser. It is not advice, it is not a forecast anyone warrants, and it is not a statement that the plan is accurate or achievable.

## Built from
- Department for Business and Trade and GOV.UK, "Write a business plan", https://www.gov.uk/write-business-plan, no publication date shown on the page, read 23 September 2026: the definition of what a plan covers and the statement that one is needed to secure investment or a bank loan, in step 1.
- The Start Up Loans Company, "Am I eligible?", https://www.startuploans.co.uk/about-the-loan/am-i-eligible, no publication date shown on the page, read 23 September 2026: the three document assessment of business plan, cash flow forecast and personal survival budget, in steps 1 and 6.
- The Start Up Loans Company, "What is a Start Up Loan?", https://www.startuploans.co.uk/what-is-a-start-up-loan, no publication date shown on the page, read 23 September 2026: the free templates for business plan, cash flow forecast and personal survival budget, which set the minimum structure used in steps 5 and 6.
- British Business Bank, "Growth Guarantee Scheme", https://www.british-business-bank.co.uk/finance-options/debt-finance/growth-guarantee-scheme, no publication date shown on the page, read 23 September 2026: that the borrower always remains 100 percent liable for the debt, used in the security section in step 8.
- Companies House, HM Revenue and Customs and GOV.UK, "Accounts and tax returns for private limited companies", https://www.gov.uk/prepare-file-annual-accounts-for-limited-company, no publication date shown on the page, read 23 September 2026: the nine month filing deadline that fixes which filed year is the latest one available to quote in step 2.
- Companies House and GOV.UK, "Get information about a company", https://www.gov.uk/get-information-about-a-company, no publication date shown on the page, read 23 September 2026: that filed accounts are publicly downloadable as document images, which is why step 2 forbids silently preferring the management figures.
- Experian, Equifax and TransUnion, "Credit Reference Agency Information Notice (CRAIN)", version 1.2, https://www.experian.co.uk/legal/crain/, adopted 02 December 2024, read 23 September 2026: the statement that the agencies do not tell lenders whether to offer credit, used in the rule forbidding any claim that a plan is fundable.

Prompt for Codex

# business-plan

## You are given
The financial and operational inputs for one UK hospitality business writing a plan for a lender. The last two sets of filed accounts and the management accounts to the most recent completed month, each as a file with its period stated. A numbered list of forecast assumptions written by the owner, each with the assumption name, its value, its unit, the document it came from, that document's name and that document's date. The fixed cost schedule: rent, business rates, insurance, utilities, licences, software and any other recurring cost, each with an amount, a frequency and the document it came from. The payment timing rules the owner supplied: VAT quarter end dates and payment dates, wage payment dates, supplier payment terms per supplier, rent quarter days, and the dates of any existing loan or finance repayments with amounts. The proposed new facility with its amount, rate, term and monthly repayment. The personal survival budget figures for each owner: income sources, essential outgoings by category, and the owner's name. The asset list with, for each asset, a description, the owner's stated current value, the basis of that value, whether it is charged and to whom. The narrative text of the plan exactly as the owner wrote it, in a single file. The closure dates and any weeks the venue does not trade. All judgement calls, including every assumption value, the growth position, the asset valuations and the wording of the narrative, have already been made by the owner and are supplied to you.

## Produce
Write into a `./business-plan-output/` folder:

1. `trading-history.csv` with these columns in this order: `period_label`, `period_start`, `period_end`, `source_document`, `source_type`, `turnover_gbp`, `gross_profit_gbp`, `gross_margin_percent`, `wages_gbp`, `wages_percent_of_turnover`, `rent_and_rates_gbp`, `other_overheads_gbp`, `operating_profit_gbp`, `drawings_or_directors_remuneration_gbp`, `disagrees_with_filed`. `source_type` is exactly one of `filed accounts`, `management accounts`, `draft accounts`. `disagrees_with_filed` is `yes`, `no` or `not applicable`, and every `yes` is listed in `gaps.md` with both figures shown side by side and neither one discarded.
2. `assumptions.csv` with columns: `assumption_no`, `assumption_name_verbatim`, `value`, `unit`, `applies_to_months`, `source_document`, `source_document_date`, `sourced`, `used_in_base_case`, `used_in_downside_case`. `sourced` is `yes` only where a source document name and date were both supplied. Any row with `sourced` as `no` and `used_in_base_case` as `yes` is a defect and is listed first in `gaps.md`.
3. `forecast-base.csv` and `forecast-downside.csv`, both with columns: `month_label`, `month_start`, `month_end`, `trading_days`, `covers`, `average_spend_gbp`, `turnover_gbp`, `cost_of_sales_gbp`, `gross_profit_gbp`, `gross_margin_percent`, `wages_gbp`, `fixed_costs_gbp`, `other_costs_gbp`, `operating_profit_gbp`, `assumptions_used`. `assumptions_used` lists the `assumption_no` values behind that row, comma separated. The downside file applies only the single supplied downside adjustment and nothing else; where no downside adjustment was supplied, the file is not written and the omission is recorded in `gaps.md`.
4. `cash-flow.csv` with columns: `month_label`, `opening_balance_gbp`, `receipts_gbp`, `supplier_payments_gbp`, `wage_payments_gbp`, `vat_payments_gbp`, `rent_payments_gbp`, `existing_finance_payments_gbp`, `proposed_new_repayment_gbp`, `other_payments_gbp`, `closing_balance_gbp`, `closing_balance_excluding_new_repayment_gbp`, `is_lowest_month`. Exactly one row carries `is_lowest_month` as `yes`, chosen on `closing_balance_gbp`. Timing follows the supplied payment rules exactly and is never smoothed.
5. `survival-budget.csv` with columns: `owner_name`, `income_source`, `income_amount_gbp`, `income_frequency`, `cost_category`, `cost_amount_gbp`, `cost_frequency`, `monthly_income_total_gbp`, `monthly_essential_costs_total_gbp`, `monthly_surplus_gbp`, `surplus_after_proposed_repayment_gbp`. `cost_category` is exactly one of `housing`, `utilities`, `food`, `transport`, `insurance`, `childcare`, `existing_credit_repayments`, `other_essential`. One block per owner, with the totals repeated on each row of that owner's block.
6. `security-position.csv` with columns: `asset_ref`, `asset_description_verbatim`, `owner_stated_value_gbp`, `valuation_basis_verbatim`, `valuation_date`, `already_charged`, `charge_holder_verbatim`, `available_as_security`, `is_a_home`, `personal_guarantee_requested`, `personal_guarantee_decision`. `already_charged`, `available_as_security` and `is_a_home` are `yes`, `no` or `not stated`. `personal_guarantee_decision` is always written as `owner decision, not recorded here` and is never completed with anything else.
7. `claims-register.csv` with columns: `line_no`, `sentence_verbatim`, `claim_type`, `evidence_document_supplied`, `evidence_file_name`, `evidence_date`, `substantiated`. `claim_type` is exactly one of `trading figure`, `market or demand claim`, `competitor claim`, `award or accolade`, `contract or booking claim`, `capability claim`, `plain description`. `substantiated` is `yes` or `no`. Rows marked `plain description` need no evidence. Every sentence in the supplied narrative is read and classified; none is skipped.
8. `summary-facts.md` recording, on separate lines and nothing else: the amount of the proposed facility; the purpose sentence verbatim; the proposed monthly repayment; the lowest closing balance in `cash-flow.csv` with its month, shown both including and excluding the new repayment; the same figure from `forecast-downside.csv` where that file exists; the monthly surplus after the proposed repayment for each owner; the total of assets marked `available_as_security`; the count of unsourced base case assumptions; and the count of unsubstantiated claims. Every line names the file the figure came from.
9. `gaps.md` a numbered list of: every base case assumption with `sourced` as `no`; every trading history row with `disagrees_with_filed` as `yes` with both figures shown; every month in the cash flow whose closing balance is negative; every owner with no survival budget block; every asset with no valuation basis or no valuation date; every claim with `substantiated` as `no`; any missing downside adjustment; and any supplied text containing an em dash character.

## Rules
- Codex calculates, tabulates and classifies. It never writes narrative, never invents an assumption, never chooses a growth rate, never values an asset and never decides whether a claim is fair. It reports what was supplied and what was not.
- Never estimate a missing figure. An assumption with no value stops the months it feeds, those months carry empty cells, and the omission is recorded. A part built forecast reports as incomplete rather than as a number.
- Never smooth a payment. VAT, rent and wages fall in the months the supplied rules put them in, even where that produces an ugly cash flow. The ugly month is the point of the exercise.
- Never write an industry average, a typical margin, a standard wage percentage or a normal growth rate into any file. None exists for a single independent UK hospitality business and none is to be recorded.
- Never write any statement, in any file, about whether the plan is fundable, strong, or likely to be approved.
- Never complete `personal_guarantee_decision` with anything other than the fixed wording given above. The decision puts the owner's own assets at risk and is not recorded by a tool.
- Never carry a figure forward from an earlier version of the plan or from a prior year to fill a current gap.
- Never move an unsourced assumption into the base case, and never silently promote a downside figure into the base case because the base case looks poor.
- Use British English, GBP written as `GBP` or the word pounds, and DD Month YYYY dates. No em dash characters in any file you write, and any supplied text containing one is recorded verbatim and flagged in `gaps.md`.
- Every narrative sentence, asset description, valuation basis and assumption name is copied exactly, including its punctuation and capitalisation.
- Every file ends with this line: this is a working document prepared for the owner to check against their own accounts, and for their own accountant to review on the forecast, the tax position and the presentation of the figures, and for their own solicitor and an FCA authorised adviser where a personal guarantee, a charge over a home or a regulated agreement is involved. It is not advice and it is not a warranty that any forecast will be achieved.

## Return
The absolute path of every file written, the row count of each CSV, the number of months forecast in the base case and in the downside case, the lowest closing balance with its month shown both including and excluding the proposed repayment, the same figure from the downside case or the word absent where no downside adjustment was supplied, the count of assumptions and how many are sourced, the list of any unsourced assumption used in the base case by number and name, the count of trading history rows disagreeing with the filed accounts with both figures for each, the monthly surplus after the proposed repayment for each owner by name, the total value of assets marked available as security with the valuation bases they rest on, the count of claims classified and how many are unsubstantiated, and the `gaps.md` item count.

Built from the best public work on this

Sources for business-plan

Everything below was opened and read on 23 September 2026. Nothing is cited that could not be loaded.

1. Department for Business and Trade and GOV.UK, "Write a business plan"

https://www.gov.uk/write-business-plan, no publication date shown on the page, read 23 September 2026.

The government's own description of the document is short enough to quote whole and useful because of what it leaves out: "A business plan is a written document that describes your business. It covers objectives, strategies, sales, marketing and financial forecasts. You can use it to: clarify your business idea, spot potential problems, set out your goals, measure your progress."

Then the sentence that sets the audience: "You'll need a business plan if you want to secure investment or a loan from a bank." Followed by: "It can also help to convince customers, suppliers and potential employees to support you."

Two different documents are being described in those two sentences and most owners write the second by accident when they were asked for the first. A plan written to convince a supplier leads with the story. A plan written for a lender leads with repayment. The skill splits them by putting the three lender questions at the top of the summary and forbidding anything above them.

Notably, GOV.UK does not itself publish a template. It points outward: "Get detailed information about how to write a business plan on the Start Up Donut website. Download a free business plan template on The King's Trust website. You can also download a free cash flow forecast template or a business plan template on the Start Up Loans website to help you manage your finances." The skill follows the Start Up Loans route in source 2 below, because that is the one attached to an actual lending decision.

Where the skill departs: the page treats the plan as a general purpose management document. The skill narrows it to the lending use, and says so, because a plan trying to do both jobs does neither well and a reader with forty applications on the desk will not hunt for the repayment section.

2. The Start Up Loans Company, "Am I eligible?" and "What is a Start Up Loan?"

https://www.startuploans.co.uk/about-the-loan/am-i-eligible and https://www.startuploans.co.uk/what-is-a-start-up-loan, no publication date shown on either page, both read 23 September 2026.

The eligibility page contains the most useful single sentence in any public UK source about what a lender is doing when it reads a plan: "Our overall loan assessment will also take into account the strength and viability of your business, demonstrated through the Business Plan and Cash Flow Forecast and the affordability of the loan through the Personal Survival Budget you submit with your application."

That sentence names three documents doing three different jobs. The business plan says the business is real and understood. The cash flow forecast says the money arrives in time to make the payments. The personal survival budget says the human being at the centre of it can live while that happens. Commercial lenders ask the same three questions with different paperwork, and structuring the plan around them is the single change that most improves how it reads.

The product page confirms the templates exist and are free: "We will provide free templates and can personally help you create your Business Plan, Cash Flow Forecast and Personal Survival Budget to submit with your application." It also supplies the facts the plan's repayment section needs when this is the facility being applied for: a "Fixed interest rate of 7.5% p.a." over a "1 - 5 year loan repayment term", with "No application or set-up fees", and the loan "unsecured, so there's no need to put forward any assets or guarantors to support an application."

Where the skill departs: the eligibility criteria themselves live in the start-up-loan skill in this pack, and the excluded purposes list, which includes "debt repayment", is not repeated here. This skill takes only the three document structure and the affordability framing.

3. British Business Bank, "Growth Guarantee Scheme"

https://www.british-business-bank.co.uk/finance-options/debt-finance/growth-guarantee-scheme, no publication date shown on the page, read 23 September 2026.

Cited for the sentence that the security section of a plan must not soften: "The scheme provides the lender with a 70% government-backed guarantee against the outstanding balance of the facility after it has completed its normal recovery process. The borrower always remains 100% liable for the debt."

Owners writing a plan for a scheme backed facility routinely write a security section implying that the government stands behind the debt. It does not stand behind the borrower's liability at all. A plan that says otherwise is a plan whose most important paragraph is wrong, and it will be corrected by the lender in a way that costs credibility everywhere else in the document.

The page also records that "Personal guarantees can be taken at the lender's discretion, in line with their normal commercial lending practices", and that "Principal Private Residences cannot be taken as security within the Scheme". The skill records both in the security section as facts, and refuses to turn either into a recommendation.

Where the skill departs: the scheme's eligibility tests, facility sizes and term lengths belong in the funding-options skill in this pack. Here the page supplies one sentence and one boundary.

4. Companies House, HM Revenue and Customs and GOV.UK, "Accounts and tax returns for private limited companies"

https://www.gov.uk/prepare-file-annual-accounts-for-limited-company, no publication date shown on the page, read 23 September 2026.

The filing deadline table decides, mechanically, which year of accounts is the latest one an owner can point at: "File annual accounts with Companies House" is due "9 months after your company's financial year ends", and first accounts are due "21 months after the date you registered with Companies House".

This matters for a plan because the trading history section has to be honest about what is filed and what is not. A plan quoting a year that has not yet been filed is quoting management figures, and it has to say so. The same page sets out what statutory accounts contain, "a balance sheet, which shows the value of everything the company owns, owes and is owed on the last day of the financial year" and "a profit and loss account, which shows the company's sales", which is the minimum set of lines the trading history table reproduces.

Where the skill departs: the page covers the Company Tax Return and the Corporation Tax payment deadline as well, and the skill does not model tax in the forecast. Tax on a forecast is the accountant's work, and a skill producing a corporation tax estimate inside a lender plan would be producing a number the owner cannot defend.

5. Companies House and GOV.UK, "Get information about a company"

https://www.gov.uk/get-information-about-a-company, no publication date shown on the page, read 23 September 2026.

Cited for one consequence. Anybody can obtain, free, "document images" of the filed accounts, along with "current and resigned officers", "mortgage charge data" and "insolvency information".

The lender reading the plan can therefore open the filed accounts while reading it. That is why step 2 of the skill forbids silently preferring a management figure over a filed one for the same period, and instead requires the disagreement to be stated. A discrepancy the owner names is a bookkeeping timing difference. The same discrepancy found by the reader is a question about whether any of the numbers can be trusted.

Where the skill departs: the wider Companies House record, and the charge reconciliation built from it, belong in the lender-pack skill in this pack. Here the page supplies only the reason for one drafting rule.

6. Experian, Equifax and TransUnion, "Credit Reference Agency Information Notice (CRAIN)", version 1.2

https://www.experian.co.uk/legal/crain/, adopted 02 December 2024, read 23 September 2026.

Cited for the sentence that governs what this skill refuses to say about its own output: "we do not tell lenders whether to offer credit to consumers; we just provide services that help those lenders make decisions about consumers. An organisation's own data, knowledge, processes and practices will also play a significant role in those decisions."

The last clause is the one that matters for a business plan. Even with perfect information about the applicant, the decision depends on the lender's own appetite that quarter, which is invisible from outside. A plan can be complete, true and well presented and still be declined for reasons that have nothing to do with the plan. Telling an owner otherwise sets them up to read a normal commercial decline as a verdict on their business.

Where the skill departs: everything about what a credit file contains and how to read it is the credit-profile skill in this pack. CRAIN appears here only to justify one rule.

Best public prompt we found for this job

The closest public artefact is the `cash-flow-snapshot` skill in Anthropic's `knowledge-work-plugins` repository, raw source at https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/small-business/skills/cash-flow-snapshot/SKILL.md, loaded successfully on 23 September 2026. The repository star count is 25,463, read from https://api.github.com/repos/anthropics/knowledge-work-plugins on 23 September 2026.

Its best instinct is one this skill copies directly. On the opening balance it says:

**Always establish the starting cash balance** - "will I make payroll" is a question about the balance, not the net. Pull it in Step 2, or ask: "What's in the business account right now, and as of what date?" **Never assume one.** If nobody knows, head the output "no opening balance - net change only" and drop cash-on-hand from the risk flags.

Naming the output as incomplete rather than filling the hole is exactly the discipline the assumption register and the gaps file exist to enforce. The same skill also refuses to infer a payout lag where a connector cannot read one, telling itself to ask the owner and "never infer a lag". That is the behaviour that separates a forecast from a fiction.

What we deliberately did not copy. `cash-flow-snapshot` produces "percentage-variance confidence bands" from each customer's payment variance. For a hospitality venue where most sales settle on the day, a confidence band computed from a handful of corporate invoices would be a statistically meaningless number sitting in a document a lender is reading closely, and a number that cannot be defended in a meeting is worse than no number. It is also built around ledger connectors, MYOB, NetSuite, QuickBooks, Xero and Zoho, where the input here is filed accounts and till reports.

Honest answer on the wider search: no good public prompt exists for writing a UK lender facing business plan. The public material is templates, and templates are the problem rather than the solution: they produce twelve headings of equal weight when the reader has three questions, and they have no concept of an assumption register, which is the only part of a forecast a careful reader actually interrogates. We looked through the 240 skill index of `knowledge-work-plugins`, including the whole `small-business` and `finance` sets, and through general prompt libraries. Nothing found requires a source document against every assumption, and that requirement is the entire difference between a plan and a wish list.

Want this running in your business?

I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.