All questions

PMF Detective: the P/MF Detective method, inside Claude Code

Short answer: PMF Detective is a free, open-source Claude Code plugin I built. Before you write a line of code, it walks you through who it's for (ICP), why they'd care (value proposition), the smallest thing to build (MVP), and how you'll test it (a validation plan with GO / ITERATE / PIVOT thresholds). Everything is saved as files your AI coding agent uses as product context.

AI made building cheap. It didn't make being wrong cheap. Most builders skip straight to code, and Claude Code will happily build whatever you ask for.

PMF Detective makes you answer the hard questions first, one question at a time.

Install

In Claude Code, run:

/plugin marketplace add adidacta/pmf-detective
/plugin install pmf-detective@pmf-detective

Then type /plan-pmf to start the guided flow, or just ask: "Help me define my ICP."

Or let Claude Code do it

Paste this into Claude Code, inside your project folder.

Help me install and run the PMF Detective plugin for Claude Code: https://github.com/adidacta/pmf-detective

1. Tell me the exact commands to install it (it's a plugin marketplace: /plugin marketplace add adidacta/pmf-detective, then /plugin install pmf-detective@pmf-detective), and wait while I run them.
2. Once it's installed, start the guided flow with /plan-pmf and walk me through it one question at a time.
3. Save everything to the pmf/ folder in this repo so we can build from it later.

No Claude Code? Run the method in any AI chat

Paste this into ChatGPT, Claude or any assistant that can read web pages.

I don't use Claude Code. Read the PMF Detective method by Adi Shmorak from https://github.com/adidacta/pmf-detective (start with the README and the files in skills/), then run it with me here as a guided interview:
1. ICP: help me write three customer hypotheses, drill each into the real pain, and pick the strongest.
2. Value proposition: a Callout (who it's for and the pain) and a Magnet (the future they want).
3. MVP: work backwards from the Aha! moment to the smallest thing to build, plus an out-of-scope list.
4. Validation plan: my three riskiest assumptions, how to test them, and GO / ITERATE / PIVOT thresholds set before any data comes in.
Ask one question at a time and wait for my answer. At the end, give me everything as clean markdown I can save.

What it does

  1. ICP: three customer hypotheses, each drilled to the real pain with 5 Whys and researched in parallel. You pick the strongest.
  2. Value proposition: a Callout (identity, context and pain that makes them say "that's me") and a Magnet (the future they want). Four angles to choose from.
  3. MVP scope: work backwards from the Aha! Moment to the shortest path, the features it needs, and an out-of-scope list.
  4. Validation plan: your three core assumptions, the test method, and GO / ITERATE / PIVOT thresholds set before the data lands.

Then it routes you to a way to test: a landing page it generates for you, outreach with Mom Test interview questions, or a build-and-test path.

What you get

pmf/
├── icp.md              who your customer is
├── value-prop.md       why they should care
├── mvp.md              aha moment, features, out of scope
├── validation-plan.md  how you'll test, and the thresholds
└── outreach-plan.md    Mom Test questions and channels

Commit the folder. Every teammate and every AI session then builds from the same foundation, and you update it as you learn.

Who it's for

Founders and builders using Claude Code who have an idea, or an early product, and want to validate it before investing more. It's the same method I use with clients, packaged so you can run it yourself.

Free and open source (MIT). View it on GitHub.

Ran it and want a second pair of eyes on the result? Send me your pmf/ files.