testing-business-ideas
A startup-validation coach built from the Testing Business Ideas method (Bland & Osterwalder): turn an idea into testable hypotheses, sort t
Open source Open in the app JSON README (API)
About
A startup-validation coach built from the Testing Business Ideas method (Bland & Osterwalder): turn an idea into testable hypotheses, sort them by risk, pick experiments from the complete 44-experiment library, write Test Cards, and decide persevere / pivot / kill.
Details
- Kind
- Plugins
- Topic
- Developer tools
- Publisher
- nimitbhargava
- Origin
- gemini
- Category
- ferramentas
- Version
- 1.2.0
- Stars
- 1
- Last push
- 2026-07-21T03:22:31Z
- Repository state
- ativo
- Language
- Shell
- License
- NOASSERTION
- Added
- 2026-08-30 14:13:39
- Updated
- 2026-08-30 14:13:39
- Origin id
nimitbhargava/testing-business-ideas-skill
README
# Testing Business Ideas: an Agent Skill
An [Agent Skill](https://agentskills.io) that turns your coding agent into a startup
validation coach. It follows the open SKILL.md standard, so it works natively in
**Claude Code, OpenAI Codex, Gemini CLI**, and other compatible agents (Cursor,
GitHub Copilot, and more). Give it a business idea and it walks you through the
method from *Testing Business Ideas* by David Bland and Alex Osterwalder
(Strategyzer, 2020): break the idea into testable hypotheses, sort them by risk,
find the ones that would kill the idea if they're wrong, and pick concrete
experiments to run this week instead of building on faith.
It ships with the complete 44-experiment library, the assumptions-mapping and
hypothesis framework, a plain-English glossary, and fill-in Test Card / Learning Card
templates.
> Not affiliated with or endorsed by the authors or Strategyzer. This is an
> independent educational tool that summarizes and restructures the method for
> coaching use. If you find it useful, buy the book. It is very good.
## What it does
You bring an idea. The skill runs six steps:
1. **Surface assumptions** hiding inside your idea.
2. **Write testable hypotheses** ("We believe that...", precise and discrete).
3. **Map and prioritize** them by importance and evidence, so you test the dangerous
unknowns first, not the comfortable ones.
4. **Pick experiments** from the library, cheapest and fastest first, matched to your
business type.
5. **Design the test** with a Test Card and a pass/fail number set in advance.
6. **Read the results** into a persevere / pivot / kill decision.
The throughline: evidence from what people *do* beats what they *say*, and you should
reduce uncertainty before you build.
## Example
```
You: I want to build a subscription box of science kits for parents.
Claude: Here's the idea as a rough business model, and the assumptions it's
resting on... The riskiest one, important and with zero evidence, is
that parents will pay monthly rather than buy one-off. Let's test that
first, before anything gets built.
Hypothesis: "Millennial parents with kids 5-9 will pay $15/month for
curated science kits matched to their kid's grade level."
Cheapest strong test: a landing page with a real Subscribe button that
takes a card, driven by ~$200 of ads. Evidence strength 4/5 because
it's real payment intent, not a survey answer. Here's your Test Card...
```
## Install
Pick your agent. The same skill folder serves all of them.
### Quick install, any agent (skills.sh)
[skills.sh](https://skills.sh) (by Vercel Labs) installs skills into 70+ agents,
including Claude Code, Codex, Gemini CLI, and Cursor:
```bash
npx skills add nimitbhargava/testing-business-ideas-skill
```
### OpenAI Codex
Codex supports Agent Skills natively. Inside Codex, run the installer skill:
```
$skill-installer install https://github.com/nimitbhargava/testing-business-ideas-skill/tree/main/skills/testing-business-ideas
```
Or copy the folder manually to `~/.agents/skills/` (or `~/.codex/skills/` on older
builds). Invoke with `$testing-business-ideas` or just describe your idea.
The repo is also a Codex plugin marketplace (the `.codex-plugin/` and `.agents/`
manifests):
```
/plugin marketplace add nimitbhargava/testing-business-ideas-skill
/plugin install testing-business-ideas@nimit-skills
```
### Gemini CLI
Gemini CLI supports Agent Skills natively:
```bash
gemini skills install https://github.com/nimitbhargava/testing-business-ideas-skill.git --consent
```
The repo is also a Gemini CLI extension (this is what the root
`gemini-extension.json` is for):
```bash
gemini extensions install https://github.com/nimitbhargava/testing-business-ideas-skill --auto-update
```
### Claude Code
Two ways: install it as a plugin from the marketplace (one command, auto-updates), or
copy the skill folder in by hand. The repo doubles as its own Claude Code plugin
marketplace.
#### Option A: plugin marketplace (recommended)
Inside Claude Code, run:
```
/plugin marketplace add nimitbhargava/testing-business-ideas-skill
/plugin install testing-business-ideas@nimit-skills
```
That is it. The skill loads immediately and auto-fires when you describe a business
idea. In the `/` menu it appears under the plugin namespace (Claude Code namespaces
plugin skills, so it shows as `/testing-business-ideas:testing-business-ideas`); you
rarely need to type it, just describe your idea.
Update or remove it later with:
```
/plugin marketplace update nimit-skills
/plugin uninstall testing-business-ideas@nimit-skills
```
#### Option B: copy the skill folder
This gives you the clean bare command `/testing-business-ideas` (no plugin namespace).
**Personal (available in every project):**
```bash
git clone https://github.com/nimitbhargava/testing-business-ideas-skill.git
cp -r testing-business-ideas-skill/skills/testing-business-ideas ~/.claude/skills/
```
**Project-scoped (available only in one repo, shareable with your team):**
```bash
git clone https://github.com/nimitbhargava/testing-business-ideas-skill.git
cp -r testing-business-ideas-skill/skills/testing-business-ideas /path/to/your/project/.claude/skills/
```
**Or use the installer** (defaults to a personal install):
```bash
git clone https://github.com/nimitbhargava/testing-business-ideas-skill.git
cd testing-business-ideas-skill
./install.sh # installs to ~/.claude/skills/
./install.sh /path/to/project # installs to that project's .claude/skills/
```
Restart Claude Code (or start a new session) after a manual install so it picks up the
skill. With Option B the folder name becomes a slash command automatically: type
`/testing-business-ideas`, optionally with your idea inline
(`/testing-business-ideas a subscription box of science kits for parents`).
### Claude apps (claude.ai and Desktop)
There is no slash-command syntax on claude.ai or the Desktop app, and `~/.claude/skills/`
on your machine is Claude Code only. To use this skill there, add it as a skill through
the Customize menu, then it auto-invokes when you describe an idea. For cloud sessions
(routines and similar), commit the `skills/testing-business-ideas/` folder into your
repo's `.claude/skills/` so the session loads it.
### API and your own agents
The same folder is a valid [Agent
Skill](https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview). Upload
it as a skill in the Claude Developer Platform, or load `SKILL.md` as a system prompt
with the reference files attached if you are wiring it into your own agent. The content
is plain Markdown and JSON, no dependencies, no build step.
## Usage
Two ways to trigger it in Claude Code:
- **Slash command:** `/testing-business-ideas` (Option B install) or the namespaced
`/testing-business-ideas:testing-business-ideas` (plugin install), optionally with
your idea after it.
- **Just talk:** describe an idea and Claude fires the skill on its own. For example:
- "Help me validate this business idea: ..."
- "What's the riskiest assumption in this and how do I test it?"
- "What experiment should I run to find out if people will pay for this?"
- "I ran the test and got these results, should I persevere or pivot?"
You do not need to have read the book. The skill carries the method.
## What's inside
```
testing-business-ideas-skill/
├── .claude-plugin/
│ ├── marketplace.json Claude Code plugin marketplace
│ └── plugin.json Claude Code plugin manifest
├── .codex-plugin/
│ └── plugin.json OpenAI Codex plugin manifest
├── .agents/
│ └── plugins/marketplace.json Codex plugin marketplace
├── gemini-extension.json Gemini CLI extension manifest
├── README.md
├── LICENSE
├── install.sh
└── skills/
└── testing-business-ideas/ <- the skill (Option B copies this into ~/.claude/skills/)
├── SKILL.md the coaching workflow Claude follows
├── reference/
│ ├── method.md the framework: risks, hypotheses, evidence, decide
│ ├── experiment-library.md all 44 experiments, scannable, with selection rules
│ ├── glossary.md plain-English definitions of the vocabulary
│ ├── mindset.md the 8 pitfalls, leadership, funding stages
│ └── testing-business-ideas.json structured source data
└── templates/
├── test-card.md design a test before running it
└── learning-card.md turn results into a decision
```
The skill uses progressive disclosure: `SKILL.md` stays short, and Claude loads a
reference file only when a step needs it. That keeps the skill cheap to run and easy
to read.
## Coverage and provenance
The skill covers the whole book:
- **All 44 experiments** (pp.106-306), with cost, setup time, run time, evidence
strength, tested risks, and type **transcribed from the book's own rating tables,
profile pages, and selection charts** (vector-extracted from the page data and
cross-checked visually), using the book's printed scale calibration. The few places
the book contradicts itself are adjudicated by 2-of-3 majority across those sources
and listed in the JSON's `library_notes`.
- **All 10 of the book's named case studies** (Product Hunt, Intuit, Palm, NSA
CubeSat, American Family Insurance, Realtor.com, Zoku, Buffer, Thrive Smart
Systems, Topology Eyewear).
- **The full method**, including the closing Mindset chapter (the eight experiment
pitfalls, leading through experimentation, and organizing/funding testing).
- Prose descriptions are paraphrased summaries, not book text. Spot-checked numeric
thresholds (interview ranking accuracy, feature-stub funnel targets, brochure and
referral conversion targets) were verified against the book and confirmed.
Page numbers point to the book for exact procedures and figures.
## Contributing
Issues and pull requests welcome, especially: tightening the coaching prompts in
`SKILL.md` and fixing any place the summaries drift from the book. Keep the tone
coach-not-cheerleader and keep `SKILL.md` short.
## License
MIT, see [LICENSE](LICENSE). The skill implementation is MIT-licensed. The underlying
methodology, frameworks, and experiment names belong to David Bland, Alex
Osterwalder, and Strategyzer, and are referenced here for education under fair use.
## Privacy
This skill is plain text: no code execution, no network calls, no data
collection of any kind. See [PRIVACY.md](PRIVACY.md).