COMPANY.md
The assistant stops describing your company the way its training data describes companies in your category. Short, factual and mostly about boundaries. What you do, who you are, what you are allowed to claim and what would be a lie — because the fastest way to get a model to fabricate a certification or a customer count is to leave those fields empty.
# Company — <Company name>
## What we do
<One sentence a customer would recognise. Not a mission statement.>
## Stage and size
- Founded: <year>
- Team: <n people>
- Customers: <n, or "fewer than ten — say 'a handful' rather than a number">
- Funding: <bootstrapped / seed / Series A — and whether this is public>
- Markets: <UK and Ireland. Not US yet — do not imply otherwise.>
## Who we are for
<See ICP.md. Do not restate it here; a second copy of the ICP is a second copy to keep current.>
## What we may claim
- <Cut close time from nine days to two — one named customer, publishable with the name.>
- <Two logistics customers of similar size — numbers publishable, names under NDA.>
- <SOC 2 Type II: IN PROGRESS. Never claim it as complete.>
## What we must never say
- <That we integrate with SAP. We do not, and it is the most common wrong assumption.>
- <Any customer name other than <Name>, which is the only written consent we hold.>
- <Anything about our competitors' pricing. We do not have current figures.>
- <"Enterprise-grade" or "bank-level security". Neither is meaningful and both invite a question we cannot answer well yet.>
## How we talk about the category
<We call it multi-entity payroll reconciliation. Not "fintech", not "payroll automation" — the first is too broad to mean anything and the second is a category we lose in.>
## The honest weaknesses
<Say these plainly when relevant rather than hiding them: no US payroll support, no SAP integration, and the product needs a data extract that takes the customer about two hours to set up. An assistant that knows the weaknesses writes better messages than one that does not, because it stops promising things we cannot do.>
## Contact and links
- Site: <url>
- Who signs off on public claims: <name, role>- Claude Project knowledge
- ChatGPT Project files
- Agent folders
- All of it. This one is short on purpose.
Everything in <angle brackets> is example content, not instructions. Replace it. The file ships filled in rather than empty on purpose: an empty template gets abandoned, a filled one gets edited.
The reference behind this file
Including the "what we do not do" section, which is the part most companies leave out and the part that keeps an assistant honest.
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
Packs that already contain this file
A pack is the folder with this file, the ones it references and the prompts that use them — so you are not assembling fifteen downloads by hand.
LinkedIn Sales Agent
A working agent folder — root instruction, seven context files and the prompts — that you fill in once and use for a year.
11 files, 8 promptsLinkedIn Content Agent
Drafts in your voice, built from real material, with a self-critique attached before you read them.
6 files, 5 promptsLinkedIn Agency System
A per-client structure where one client’s proof point cannot end up in another client’s post.
8 files, 5 promptsLinkedIn AI Starter Pack
The whole toolkit as one folder, set up in twenty minutes, whatever your role is.
12 files, 12 promptsOne file is a start. The folder is the point.
Files reference each other — CLAUDE.md declares the reading order, the prompts expect the files by name. The packs ship them together, already wired up.