A bundle of eleven free Claude skills that promises to "test a business before you launch it" has reached about 1,300 stars on GitHub. founder-skill, by Jake Schincariol, strings together a simulated board of directors, a competitor scout, a consumer panel of 100 buyer agents, a pricing strategist, a CFO that computes unit economics, and a launch planner, all as Claude skills under an MIT license. The GitHub README is unusually honest about what the simulation cannot do, and that honesty is the most useful part for anyone thinking of using it.
This guide covers what each skill does, how the pieces chain together, the four offline Python tools, install steps, and a claims-versus-README reading. It sits alongside other skill bundles we have covered, such as the same author's Replica skills that clone any app, the Superpowers framework and the 177 research skills collection.
TL;DR
| Question | Answer |
|---|---|
| What is it? | Eleven Claude skills plus four standard-library Python tools |
| Cost / license | Free, MIT, no signup, no API key |
| Install | Paste the repo link and say "install skill", or use the Claude Code plugin marketplace |
| Order | board, competitors, consumer, pricing, offer, cfo, marketing, brand, ops, launch, plan |
| Panel | 100 simulated buyers by default, one sub-agent each |
| Tools | unit_economics.py, panel.py, van_westendorp.py, compile.py; none touch the network |
| Popularity | About 1.3k stars and 174 forks at time of reading, one commit |
| Key warning | Simulated buyers are not customers; read the buy rate as an upper bound |
The eleven skills
| Command | Role | What it produces |
|---|---|---|
| /founder-board | Board of directors | Three sub-agents each applying one published framework (offers, monopoly, product) score the idea, name what would kill it and vote fund, fund if, or pass |
| /founder-competitors | Competitor scout | Direct and indirect competitors and substitutes from public sources, with prices, positioning and complaints, each fact linked |
| /founder-consumer | Consumer panel | A swarm of buyer agents, 100 by default, each with its own income, habits and objection, tallying who buys and why |
| /founder-pricing | Pricing strategist | An acceptable price range from the panel's answers, compared with competitors and margin |
| /founder-offer | Offer architect | Bonuses, guarantee, urgency and a name that answer the panel's objections, then re-tested |
| /founder-cfo | CFO | Contribution per sale, real profit margin, break-even per day, year 1 by month, cash needed, payback and what-ifs |
| /founder-marketing | Marketing director | Positioning, channels, a dated 30-day campaign, ten hooks and a maximum customer acquisition cost |
| /founder-brand | Brand designer | Name candidates with trademark, domain and handle checks to run, plus voice and a brief |
| /founder-ops | Operations manager | Suppliers, staffing, routines, tools, permits to check and a risk register |
| /founder-launch | Launch lead | A cheap real-world test first, then a dated countdown, run sheet and first 30 days |
| /founder-plan | Business planner | A compiled plan with a verdict: Profitable, Not yet, or Incomplete |
Each skill reads what the earlier ones wrote into a founder/ folder in your project, so you run them in sequence. You can also run any one alone; the README suggests /founder-consumer --quick on a half-formed idea as a fast way to decide whether to continue.
How to install
- Easiest: paste
https://github.com/Jakeschincariol/founder-skillinto Claude and say "install skill". - As a Claude Code plugin:
/plugin marketplace add Jakeschincariol/founder-skillthen/plugin install founder-skill@founder-skill. Plugin skills are namespaced, such as/founder-skill:founder-board. - Plain commands: clone the repo and copy the
skills/founder-*folders into~/.claude/skills/, or into your repo's.claude/skills/for a project-local install. - No Claude Code? Paste any single
SKILL.mdat the top of a chat and it runs as a mode, without sub-agents or the Python tools.
The tools need Python 3.8 or newer and nothing to pip install. For background on how skills load and run, read our complete guide to agent skills, and see the Claude Code documentation for plugin commands.
The four offline tools
All four are standard-library Python and none touch the network, per the README.
| Tool | What it does |
|---|---|
unit_economics.py | Reads one JSON file and reports contribution, break-even, margin at plan, year 1 by month, cash needed, payback and three what-ifs. Every number traces to an input and a missing field is an error, never a guess |
panel.py | Deals persona cards from your target customer (same seed, same cards), writes one brief per buyer, checks answers and tallies them. It never answers for a buyer |
van_westendorp.py | Reads four price answers per buyer and finds points of marginal cheapness and expensiveness, the optimal price and the indifference price |
compile.py | Assembles the plan and computes the verdict |
The Van Westendorp method is a standard survey technique for price ranges; its Wikipedia entry explains the four questions. Having the CFO and verdict in deterministic code instead of in a model's prose is the design choice that makes the output auditable: change an input and the output changes predictably.
The worked example
The README walks through an iced matcha latte café in Toronto at $6.50 a cup, with all inputs included so you can run the tools yourself. According to the README, the CFO step produced these example figures: each cup leaves $4.76 after its own costs, break-even is 186 cups a day, the margin at the planned 383 cups a day is 38%, year 1 makes $93,926 on operations after a slow ramp, the $25,280 startup spend is earned back in month 8, and the founder needs $34,092 in cash before it pays for itself. These are outputs of the example inputs, not a forecast for any real café.
Claims versus the README's own caveats
The repo description reads like a strong promise, so it matters what the fine print says.
| Marketing line | What the README actually says |
|---|---|
| "A board trained on Hormozi, Thiel and Jobs" | The board applies short summaries, in the author's words, of ideas from three books. It never speaks as those people, quotes them, or claims endorsement; the project is not affiliated with them |
| "100 buyer agents run your business through 100 scenarios" | The panel is simulated buyers, not customers; language models lean agreeable, so the buy rate is an upper bound; confirm big calls with real people; never use panel answers as testimonials |
| "Finds your real profit margins" | The numbers are only as good as the inputs; every estimate is marked as an estimate; check costs with suppliers and structure, payroll and tax with an accountant |
| "Know how to launch, market, and make money" | Not financial, legal or tax advice; permits, trademarks and claims depend on location |
| Research | Uses public sources with links; no invented competitors, prices, reviews or quotes; no scraping against a site's terms |
The strongest practical design choice is that /founder-launch begins with a real-world test, such as a pop-up with real buyers, precisely because the panel cannot replace one. We could not run the skills ourselves for this write-up, so our read is based on the README and file layout, not on tested output.
How to run it well
A few habits make the output more useful than a single pass of prompts.
- Write the idea file honestly.
/founder-boardwritesfounder/idea.mdfrom what you tell it, and everything downstream reads that file. Vague input produces vague objections. - Edit the customer file by hand. The panel is only as realistic as the target customer description. Replace guessed incomes and habits with numbers from real conversations or public statistics.
- Keep the seed fixed when you compare. The panel tool deals the same cards for the same seed, so changing the price or offer and re-running shows a like-for-like change instead of random noise.
- Replace estimates with quotes. The CFO marks estimates; swap each one for a supplier quote and rerun the unit-economics tool to see how the break-even moves.
- Treat the verdict as a checklist. A "Not yet" verdict names what failed. Fix that input, rerun, and keep the history in Git.
Who should use it, and who should not
Good fit:
- Early founders who want a structured checklist of questions before spending money.
- Teams that want a repeatable, file-based planning workflow they can version in Git.
- Educators teaching unit economics, who can use the example JSON as a live demo.
Poor fit:
- Anyone who plans to treat the simulated buy rate as demand data. It is an upper bound at best.
- Businesses where regulation, safety or local rules dominate. The skills point to what to check but do not replace advice.
- Teams that need market data beyond public sources. There is no paid research feed.
Risks of any simulated-customer workflow
Synthetic panels are fast and cheap but share failure modes: the model's picture of your customer comes from its training data, which may be stale or stereotyped; agreeable models understate objections; persona cards seeded from your own description reflect your assumptions; and results can look precise (a buy rate of 41%) when they are not. Use panels to surface objections and wording, then validate with a handful of real conversations. If you give agents shell access to run the tools, apply the usual care: review the Python before running, keep the project folder scoped, and consider an agent security layer such as AgentBeam from the explainx.ai team.
Related reading
- Complete guide to agent skills
- Replica skills: eleven Claude skills to clone any app
- Superpowers agentic skills framework
- 177 research skills (K-Dense)
- huashu-motion launch-film skill
- Ponytail: a lazy-senior-dev skill
- Official: founder-skill on GitHub · Agent skills on explainx.ai
Details are from the repository README at the time of writing; we did not run the skills.