A Claude Code plugin for designing and analyzing MVPs (Maximally Valuable Problems). Reframes the traditional "Minimum Viable Product" around hypothesis testing and revealed preference rather than feature minimization and product.
Guides a structured conversation to produce a testable hypothesis grounded in revealed preference, a pre-committed threshold test, and a concrete experiment design. Outputs a markdown file.
Triggers: "create an MVP", "plan an MVP", "validate my idea", "test a hypothesis", "design an experiment for my idea"
Guides post-experiment analysis of results against the pre-committed threshold test. Focuses on what the outcome means and what to do next — not on whether the outcome feels good. Challenges confirmation bias directly.
Triggers: "analyze MVP results", "review experiment results", "did the MVP pass", "post-experiment analysis"
Add the plugin to your Claude Code configuration:
claude plugins add /path/to/mvps/plugins/mvps
plugins/mvps/
├── .claude-plugin/plugin.json
├── references/
│ ├── mvp-file-format.md
│ └── mvp-analysis-format.md
└── skills/
├── mvp-creation/
│ └── SKILL.md
└── mvp-analysis/
└── SKILL.md
MVP = Maximally Valuable Problem. The MVP is not the product. It is a test of whether the problem exists and drives behavior.
Revealed preference over stated preference. Design every experiment to observe what people do, not what they say. Survey data and interview quotes are not evidence of demand.
Pre-committed threshold tests. A hypothesis without a pass/fail threshold set before the experiment is useless. The threshold prevents post-hoc rationalization.
50% is the optimal pass rate. A team that passes every test is setting weak thresholds. A team that fails every test is disconnected from reality. A roughly even split means the team is testing at the frontier of its knowledge.
BSD 3-Clause. See LICENSE.