startup-stack

Prompt library

19 prompts that run on top of a filled-in stack. Copy one, paste it into an AI that can see your folder, and answer what it asks.

Copy a prompt, paste it into an AI that can see this folder, answer what it asks. Each one reads only the sections it needs and writes back into the stack.

They assume AGENTS.md is loaded. If your tool does not pick it up automatically, paste it first — it carries the rules that stop the model inventing numbers.

Group
Prompt 00 Bootstrap the stack Run this once, first. It turns a folder of raw material into a structured knowledge base. Build and maintain Prompt 01 Refresh the stack Folds new material into an existing stack without flattening the corrections you have made by hand. Build and maintain Prompt 02 Gap scan What is missing, unverified, stale or contradictory — ranked by how much damage it will do. Build and maintain Prompt 03 Customer and problem Sharpens who the customer is, what job they are hiring you for, and what they do today instead. Extracts the pattern across every interview you have… Sharpen a section Prompt 04 Market sizing and beachhead Builds TAM/SAM/SOM from countable units with every assumption on the page, and picks a beachhead you can actually walk into. Sharpen a section Prompt 05 Competitive intelligence Builds the full grid — including the four competitors founders reliably leave off — and tells you honestly where you lose. Sharpen a section Prompt 06 Product and roadmap Draws a hard line between what exists and what is a slide, builds a roadmap with dates and definitions of done, and checks the things that surface aw… Sharpen a section Prompt 07 Unit economics, costing and runway The most valuable prompt in this library. What one unit costs, what it costs at volume, how much cash you need to fulfil an order, and how many weeks… Sharpen a section Prompt 08 Build the list of 100 Turns "we should try selling to X" into a real list of a hundred named prospects, an outreach cadence, and a pitch you can say out loud. Sharpen a section Prompt 09 The marketing engine A paid-media step-test plan tied to your actual capacity, a creative testing schedule, the brief that holds an agency accountable, and the unglamorou… Sharpen a section Prompt 10 Pitch deck Slide-by-slide content in the order investors expect, drawn only from what is in the stack, with every gap marked rather than filled. Produce something Prompt 11 Fundraise readiness The staged funding roadmap, the dilution map, the non-dilutive money you may be eligible for and have not applied to, the data-room checklist, and th… Produce something Prompt 12 The weekly recap The pulse. Thirty minutes, every week, forever. Produce something Prompt 13 Mentor review An adversarial pass over the whole stack. What an experienced coach or investor would attack, in the order they would attack it. Get told what is wrong Prompt 14 Meeting to actions Transcript in → summary, minutes, owned action items with dates, and stack updates out. Produce something Prompt 15 Website to stack material A website in — yours or a competitor's — and cited, tagged, routed stack material out. Produce something Prompt 16 Session to record One coaching session in → the factual record, your attributed read, and the updated company record out. Run a portfolio Prompt 17 Pre-session brief Everything a coach needs before walking into a session, on half a page. Run a portfolio Prompt 18 Portfolio scan Every company's record read at once → what is going wrong often enough to teach, who needs attention, and what actually moved. Run a portfolio

Using them well

Answer the questions. These prompts stop and ask rather than assuming. That is the feature. A ten-second answer from you beats an assumption that propagates into six documents.

Do not skip steps. The sequence inside each prompt is the sequence a good coach uses. Jumping to the output produces something confident and unfounded.

Correct the output. The AI drafts; you validate. Nothing it writes is true because it wrote it.

Commit before and after.

git add -A && git commit -m "before: unit economics prompt"
# run the prompt
git diff                 # see exactly what changed
git add -A && git commit -m "07-money: unit economics from supplier invoices"

Writing your own

Every prompt in here follows the same shape, and yours should too:

  1. A **Requires:** line outside the fence — what must already be true before this is worth pasting. For the human holding the mouse.
  2. Role and scope — what the AI is doing, which sections to read, which to leave alone
  3. ## Step 0 — Check the ground — open the named files, check they hold content rather than [TBD], and if they do not: name what is missing, say which prompt fills it, and stop. Every prompt here has one, and each states its own tolerances — prompt 14 carries on happily against an empty stack because that is how a stack starts to fill; prompt 09 refuses, because a step test run against unknown capacity spends real money daily.
  4. Ask before assuming — the questions it must put to the founder rather than guess
  5. The work — numbered steps, in the order a practitioner would do them
  6. Output — exactly which files change and what goes in them

Do not restate AGENTS.md inside the fence. Every prompt opens by loading it, so repeating the tag table, the citation format or the no-invented-numbers rule spends context on something already in the window. Keep what encodes this prompt's file paths, output shape, and any judgement a model would get wrong left to itself — 02-gap-scan's "rank by damage if left open, not by ease of closing" is exactly the kind of line worth its words.

The shortest thing here is 17-pre-session-brief at 261 words, because it runs two minutes before a meeting. The longest is 15-scrape-a-site at 1,474, because it branches three ways on what your tool can actually do. Length should follow from the job, not from thoroughness.