startup-skill
/startup-skill
Testuje pomysł na startup lub biznes przed wydaniem pieniędzy: doradcy, konkurenci, symulowani kupujący, liczby i werdykt.
- Autor
- psabbagh ↗
- Repozytorium
- startup-skill ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 08.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 11
- Skrypty
- 7
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 08.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 7 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: tests/_paths.py, tests/test_buyers.py, tests/test_competitors.py, tests/test_funnel.py, tests/test_numbers.py, tests/test_pack.py, tests/test_plan.py
Co robi i kiedy
Testuje pomysł na startup lub biznes przed wydaniem pieniędzy: doradcy, konkurenci, symulowani kupujący, liczby i werdykt.
Wyzwalacze z opisu
Argumenty
/startup-skill <your business idea, or leave blank to use the conversation> [--quick]
Oryginalny opis (EN)
Tests a startup or business idea before any money is spent, start to finish: a board of advisors scores it, real competitors are researched with links, 100 simulated buyers say whether they would pay, a channel plan forecasts where the customers come from, the numbers show whether it makes money, and a business plan ends with a computed verdict, Profitable or Not yet. Use when someone says "test my startup idea", "test my business idea", "is this idea any good", "would this business work", "validate my idea", "run the startup skill", or describes a business they are thinking of starting and as
Jak działa
- Before you start, tell the user what will happen
- Run the steps
- At the end
Podgląd SKILL.md
---
name: startup-skill
description: >-
Tests a startup or business idea before any money is spent, start to finish: a
board of advisors scores it, real competitors are researched with links, 100
simulated buyers say whether they would pay, a channel plan forecasts where the
customers come from, the numbers show whether it makes money, and a business plan
ends with a computed verdict, Profitable or Not yet. Use when someone says "test my startup idea", "test
my business idea", "is this idea any good", "would this business work", "validate
my idea", "run the startup skill", or describes a business they are thinking of
starting and asks what you think of it.
argument-hint: "<your business idea, or leave blank to use the conversation> [--quick]"
---
# startup-skill
Six steps, each its own skill, run in this order. Each one reads what the earlier
ones wrote, in a `startup/` folder in the user's project.
```
startup-board -> startup-competitors -> startup-buyers -> startup-marketing -> startup-numbers -> startup-plan
the board who already does it 100 buyers where customers the numbers the verdict
come from
```
What the user typed after the command: `$ARGUMENTS`. If it is blank, take the idea
from the conversation. `--quick` runs 20 buyers instead of 100.
## Before you start, tell the user what will happen
In a few plain lines:
1. **The board**: several advisors, each applying one published playbook, score the
idea and vote.
2. **The competitors**: real research on the web, every fact linked: who already
solves this, what they charge, and what their customers complain about. It needs web
search (in Codex, start it with `codex --search`).
3. **The buyers**: 100 simulated buyers (20 with `--quick`), each a separate
sub-agent, say whether they would pay and why. This is the longest step and uses
the most of their plan. Each buyer writes one small file, so let it write without
asking: in Claude Code, accept-edits mode (Shift+Tab); in Codex, the auto approval
mode (or `codex --full-auto`).
4. **The marketing**: where the first customers will come from: two or three channels,
what each step converts, what it costs and how much of their time it takes. That
becomes the customer forecast.
5. **The numbers**: you will ask them for real numbers: the price, the costs, what they
will pay themselves.
6. **The plan**: one document with a verdict that is computed, not written: Profitable,
or Not yet with the numbers that fail.
## Run the steps
For each step, run that skill. In Claude Code, invoke it with the Skill tool
(`startup-board`, then `startup-competitors`, `startup-buyers`, `startup-marketing`,
`startup-numbers`, `startup-plan`). In
Codex, and on any host without that tool, read the step's `SKILL.md` in the sibling
folder next to this skill's folder (`../startup-board/SKILL.md` and so on, usually
under `~/.codex/skills/` in Codex) and follow it. Both hosts run the board and the
buyers as separate agents: Claude Code with its Agent tool, Codex with its
multi-agent feature.
- **Step 1, `startup-board`.** It writes `startup/idea.md` first and asks for anything
missing. After the vote, show the user the result in two lines and carry on, unless
the board's strongest version of the business is different enough that the user may
want to change the idea first. Ask.
- **Step 2, `startup-competitors`.** If the host has no web search, say so, skip it, and
tell the user the buyer profile and the price will rest on their knowledge alone.
- **Step 3, `startup-buyers`** (pass `--quick` if the user asked for it). The user
confirms the target customer before any buyer is dealt. Do not skip that. If few
buyers would buy for reasons a better offer could answer (proof, risk, price, effort,
missing information), offer one round of the skill's "Improve the pitch and re-test"
before moving on. If the reasons are "no need" or "happy with what I use today",
say so plainly instead: that is the market, not the pitch.
- **Step 4, `startup-marketing`.** Ask the user what they will really do each month to
find customers. Never fill the channels with activity they did not agree to.
- **Step 5, `startup-numbers`.** Ask for everything only the user knows in one message,
then let the marketing plan fill in the customer forecast (`funnel.py --apply`).
- **Step 6, `startup-plan`.** It computes the verdict and writes the plan.
If a step cannot run (for example, the host has no sub-agents for the buyers), say so
plainly and stop. Never skip a step silently, and never fill a step's output from
guesses.
## At the end
Give the user, in this order:
1. The verdict in one line, and the three numbers behind it.
2. What the buyers and the board agreed was the biggest problem.
3. The one thing to do this week: usually the real-world test in `startup/real-test.md`.
4. Where everything is: `startup/plan.md`.
Then say once: the buyers are simulated and lean agreeable, so their yes is the best
case; real people and real quotes confirm it. Not financial, legal or tax advice.
Pliki w paczce (11)
- LICENSE 1 KB
- README.md 9 KB
- SKILL.md 5 KB
- agents/openai.yaml 1 KB
- tests/_paths.py 1 KB
- tests/test_buyers.py 16 KB
- tests/test_competitors.py 5 KB
- tests/test_funnel.py 9 KB
- tests/test_numbers.py 8 KB
- tests/test_pack.py 11 KB
- tests/test_plan.py 10 KB
Wersje
W bazie nie ma innych skilli o nazwie startup-skill.
Migawka bazy z 13.11.2026 · wersja paczki e5abca46debe (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100startup-plan
Łączy wyniki pakietu startupowego w jeden plan biznesowy, a werdykt wylicza z danych zamiast wpisywać go ręcznie. - 100startup-validator
Sprawdza pomysł na startup: problem, dowody popytu, konkurencję i gotowość płacenia, a na końcu daje werdykt go, pivot lub kill. - 100validate-idea
Zbiera pomysł na startup w plik z grupą docelową, hipotezą problemu i progiem zaliczenia, ustalonym przed danymi. - 90pressure-test
Testuje pomysł biznesowy jak szczery do bólu współzałożyciel według Startup Playbook Sama Altmana: ocenia pięć obszarów, wskazuje trzy słabości i pierwszy krok. - 90validate-idea
Ocenia pomysł na startup lub produkt przed budową: problem, klienta, gotowość do płacenia, konkurencję i tani eksperyment. - 100validate-idea
Testuje pomysł biznesowy na dowodach przed budową: falsyfikowalne założenia, ocena zebranych dowodów, tanie eksperymenty (wywiady, smoke testy, przedsprzedaż) i werdykt z liczbami.