convez.ai
AgenteProdutividadeIntermediário
sppillai

AI Product Discovery OS

sppillai/ai-discovery-plugin

Validação rigorosa de produto antes de codificar qualquer linha

Ver no GitHub
InicialPluginScore · de 0 a 100

Plugin Claude Code que automatiza a validação de produto seguindo o framework MIT Disciplined Entrepreneurship em 24 passos. Orquestra 4 agentes IA para gerar pesquisa de mercado, personas, modelos de precificação, projeções financeiras e deck de investidor — garantindo que você valide ideias antes de construir.

Instalação

$npx convez-ai sppillai/ai-discovery-plugin

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
2
Acompanhando
0
Saúde
28%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Founders validando ideias de produto antes de investir em desenvolvimento
  • Análise de viabilidade de mercado e definição de MVP com base em dados
  • Geração automática de pitch decks e projeções financeiras para investidores

Veredito

Ferramenta sólida para founders que precisa forçar disciplina no processo de discovery. Integração profunda com Claude Code e checkpoint workflow bem pensado, mas requer comprometimento real com entrevistas de clientes para ir além da IA — não substitui validação de mercado, apenas a estrutura.

README

A tradução para o português deste README ainda está na fila.

AI Product Discovery OS

Version License Claude Code Framework


The problem

Solo founders waste 3–6 months building products nobody wants. Not because they're bad at building — because they skip the hard questions. Who exactly is your customer? What does their life look like today? Would they actually pay for this? Most founders jump straight to building and find out the answer too late.

Existing AI tools make building faster. None of them make you validate first.

What this is

A Claude Code plugin that enforces rigorous product validation before you write a single line of code. You describe your idea, four AI agents run the full MIT Disciplined Entrepreneurship 24-step framework — generating market research, customer personas, pricing models, financial projections, and an investor deck — and at every step you either approve the output or go do the real-world work (customer interviews, pricing conversations, PMF surveys) that only you can do.

At the end you have a complete discovery report, validated assumptions, a defined MVP, and a launch plan. Built before you build anything.

Quick start

# 1. Install the plugin
claude plugin marketplace add sppillai/ai-discovery-plugin
claude plugin install ai-product-discovery

# 2. Install required skill packs
claude plugin marketplace add phuryn/pm-skills
claude plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill

# 3. Create a project folder and start
mkdir my-product-idea && cd my-product-idea
claude  # open Claude Code in this folder
# then run: /start-discovery

Where to use it

Requires Claude Code — works in all Claude Code surfaces:

SurfaceHow to accessNotes
Claude Code CLIclaude in your terminalRecommended — best for long sessions
Claude Desktop — CoworkCowork tab in the Claude desktop appUse "Work in a project" to point at your folder
Claude Desktop — CodeCode tab in the Claude desktop appSame as the CLI, built into the app
VS Code extensionInstall "Claude Code" from VS Code marketplaceRun /start-discovery in the Claude panel
JetBrains extensionInstall from JetBrains MarketplaceWorks in IntelliJ, WebStorm, PyCharm, etc.

Not compatible with: claude.ai web chat, OpenAI Codex, ChatGPT, GitHub Copilot, Cursor, or Windsurf. Plugins require the Claude Code CLI.


How it works

  1. Run /start-discovery and describe your product idea in 1–2 sentences
  2. Four AI agents work through all 24 steps — generating research, spreadsheets, Word docs, and PowerPoint decks
  3. After every step you get a checkpoint: approve the output, give feedback to revise, or paste in your real-world findings
  4. At 10 steps, the AI tells you exactly what real-world action to take and waits for you to return with results
  5. All deliverables are saved to your project folder as they're created

Resuming: open the project folder again in Claude Code — the plugin reads project-state.json and picks up exactly where you left off, including any pending human action.

Steps that require your real-world input

At these 10 steps, the AI creates a findings template and waits for you:

StepWhat you doWhy it can't be skipped
3 — End User ProfileList 5–10 people you know who match the user profileSeeds the real interview pipeline
6 — User ValidationConduct 5 customer discovery interviewsThe most important step — real signal beats AI assumption
11 — Pricing StrategyRun 3–5 willingness-to-pay conversationsGrounds the LTV/CAC model in real data
12 — Customer AcquisitionMap 10 potential early customers from your networkValidates go-to-market assumptions
15 — Business ModelShare the canvas with 2–3 advisorsTests revenue model before financial modeling
19 — Key AssumptionsReview and rank the assumption matrixAdd what the AI missed from your direct experience
20 — MVP DesignPick one experiment and commit to a start dateThe AI can't decide what you'll actually execute
22 — PMF ConfirmationRun the Sean Ellis survey with 40+ usersThe quantitative gate before building anything
23 — Development PlanDecide team structure and approve the tech stackGates the entire build timeline
24 — Launch PrepExecute the launch checklistThe AI drafts everything; you pull the trigger

The 24 steps

PhaseStepsWhat gets built
1 — Market Selection1–2Target market segment, beachhead selection, TAM analysis
2 — User Research3–6End user profile, personas, lifecycle map, interview guide
3 — Value Proposition7–8Value prop, full PRD, feature prioritization matrix
4 — Business Model9–18Revenue streams, pricing, LTV/CAC, GTM channels, financial model, investor deck
5 — Validation19–22Assumption matrix, MVP design, MVP scope, PMF confirmation
6 — Execution23–24Development roadmap, tech architecture, launch plan

What we added to the framework

The MIT DE framework tells you what to validate in the right order. What it doesn't enforce is the connective tissue between user pain and the features you commit to building. Without that thread, value propositions become generic, feature lists are driven by gut feel, and no one can answer "why are we building this?" when the PRD is written.

This plugin adds a structured pain-to-feature chain across Steps 3, 5, 6, 7, and 8:

StepWhat's addedWhy it matters
3 — End User ProfileCurrent state process map — a step-by-step map of what the user does TODAY, with pain severity (1–5) and emotional state at each stepCreates the "before picture" that everything downstream anchors to
5 — Lifecycle Use CaseBefore & After transformation sheet added to the journey map ExcelMakes the value the product delivers concrete: each step of today's process mapped against what changes with the product
6 — User ValidationInterview findings loop back to update the process map — pain severity is re-scored based on real confirmation rate, new pains are added, every row marked as AI-assumed or interview-confirmedReplaces AI assumptions with real evidence before the value prop is written
7 — Value PropositionOpportunity map built from the top-severity pain points before the value prop is drafted — each pain becomes an opportunity with Impact × Confidence priority scoreValue prop is written from the highest-ranked opportunity, not generic benefit language
8 — Product SpecificationEvery feature in the PRD links to a row in the opportunity map — features without a link are flagged "Needs validation"Every feature can answer: which specific pain does this solve, how severe is it, and what process step does it change

The result is a PRD where features are not invented — they are derived from real user pain, validated by interviews, and prioritised by opportunity impact.


What you get

Every step produces real files in your project folder:

your-project/
├── project-state.json
├── IDEA.md
├── PHASE-1-MARKET-SELECTION/
│   ├── step-1-target-market-segment.md
│   ├── step-2-beachhead-market.md
│   └── deliverables/
│       ├── market-segmentation-matrix.xlsx
│       └── tam-analysis.xlsx
├── PHASE-2-USER-RESEARCH/
│   ├── step-3-end-user-profile.md
│   ├── step-3-network-map.md          ← your findings template
│   ├── step-6-interview-findings.md   ← your findings template
│   └── deliverables/
│       ├── current-state-process-map.xlsx    ← today's process with pain severity
│       ├── current-state-process-map-v2.xlsx ← updated after interviews
│       ├── personas.docx
│       ├── user-journey-map.xlsx             ← adoption journey + before/after
│       └── interview-guide.docx
├── PHASE-3-VALUE-PROPOSITION/
│   └── deliverables/
│       ├── opportunity-map.xlsx       ← pain→opportunity with priority scores
│       ├── PRD.docx
│       └── feature-prioritization.xlsx       ← features linked to opportunities
├── PHASE-4-BUSINESS-MODEL/
│   ├── step-11-pricing-conversations.md  ← your findings template
│   ├── step-12-network-map.md            ← your findings template
│   ├── step-15-advisor-feedback.md       ← your findings template
│   └── deliverables/
│       ├── tam-sam-som.xlsx
│       ├── ltv-cac-model.xlsx
│       ├── business-model-canvas.docx
│       ├── burn-rate-analysis.xlsx
│       ├── financial-projections.xlsx
│       └── investor-deck.pptx
├── PHASE-5-VALIDATION/
│   ├── step-20-experiment-commitment.md  ← your findings template
│   ├── step-22-pmf-survey.md             ← your findings template
│   └── deliverables/
│       ├── assumption-matrix.xlsx
│       ├── mvp-scope.docx
│       ├── mvp-prioritization.xlsx
│       └── pmf-validation-report.docx
├── PHASE-6-EXECUTION/
│   ├── step-23-build-decisions.md    ← your findings template
│   ├── step-24-launch-log.md         ← your findings template
│   └── deliverables/
│       ├── technical-roadmap.xlsx
│       ├── product-architecture-deck.pptx
│       ├── go-to-market-plan.docx
│       └── launch-checklist.xlsx
├── DELIVERABLES-SUMMARY/
│   ├── executive-summary-deck.pptx
│   ├── complete-discovery-report.docx
│   └── next-steps.md
└── PIVOT-1/                          ← created if you pivot
    ├── PIVOT-RATIONALE.md
    ├── project-state.json
    └── [same phase structure]

Pivoting

When interviews, PMF results, or any step reveals a wrong assumption:

  • Minor pivot — re-runs specific steps in-place, saves old files as -v1.md
  • Major pivot — creates a PIVOT-1/ folder, carries over steps still valid, and restarts from the earliest invalidated step with a PIVOT-RATIONALE.md capturing what you learned

At Step 22, if the PMF score is below 40%, the Supervisor proactively asks whether to pivot the problem, solution, or segment.


Agents

AgentResponsibilitySteps
SupervisorOrchestrates workflow, manages pivots, generates final reportAll
PM Market StrategistMarket research, TAM, business model, pricing1–2, 9–18
UX Research AgentUser interviews, personas, journey mapping, PMF3–6, 22
Architect AgentProduct spec, MVP definition, technical planning7–8, 20–21, 23

At each step, agents surface a 💬 Expert Insight block with a real quote from a relevant Lenny's Podcast episode — attributed to the guest, applied directly to your situation.

External skill packs

The agents invoke these open-source Claude Code skill packs at the right steps. Install them as part of setup:

Skill packWhat it adds
phuryn/pm-skills49 PM skills — discovery, strategy, pricing, finance, go-to-market
nextlevelbuilder/ui-ux-pro-max-skillDesign system generator for 161 product types (Steps 8 and 23)
ChatPRD/lennys-podcast-transcripts269 Lenny's Podcast transcripts, fetched live at each step

Installation

Step 1 — Install the plugin

claude plugin marketplace add sppillai/ai-discovery-plugin
claude plugin install ai-product-discovery

Step 2 — Install required skill packs

claude plugin marketplace add phuryn/pm-skills
claude plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill

Step 3 — Verify

claude plugin list

You should see ai-product-discovery@ai-discovery, pm-skills, and ui-ux-pro-max-skill all listed as enabled.

Updating

claude plugin marketplace update ai-discovery
claude plugin update ai-product-discovery@ai-discovery

The @ai-discovery suffix is the marketplace name Claude Code assigns to this repo — always use it when updating.

Local install

git clone https://github.com/sppillai/ai-discovery-plugin
claude plugin install /path/to/ai-discovery-plugin

Requirements

  • Claude Code (CLI, desktop, VS Code, or JetBrains extension)
  • anthropic-skills for Excel, Word, and PowerPoint generation
  • Brave Search MCP for real market data (optional but strongly recommended)

About MIT Disciplined Entrepreneurship

Disciplined Entrepreneurship was created by Bill Aulet, Managing Director of the Martin Trust Center for MIT Entrepreneurship. It is a structured, 24-step framework for building successful startups — developed from decades of research and teaching at MIT and refined through thousands of entrepreneurs worldwide.

The framework forces founders to answer hard questions in the right order: Who exactly is your customer? What is their life like today? Why would they pay for your solution? What's the business model? How do you get to them? Each answer builds on the last, and skipping steps is how startups waste months building things nobody wants.

Documented in Disciplined Entrepreneurship: 24 Steps to a Successful Startup (Wiley, 2013) and the Disciplined Entrepreneurship Workbook (Wiley, 2017).

This plugin is an independent implementation of the DE framework. It is not affiliated with or endorsed by MIT or Bill Aulet. All intellectual credit for the framework belongs to Bill Aulet and the Martin Trust Center for MIT Entrepreneurship.


Attribution

External Claude Code skill repos

RepoAuthorLicenseWhat it contributes
phuryn/pm-skills@phurynMIT49 PM skills spanning discovery, strategy, delivery, finance, AI readiness, and career
deanpeters/Product-Manager-SkillsDean PetersCC BY-NC-SA 4.047 PM skills (fork of phuryn/pm-skills) with pedagogic framing
nextlevelbuilder/ui-ux-pro-max-skill@nextlevelbuilderMITDesign system generator for 161 product types
ChatPRD/lennys-podcast-transcriptsChatPRD—Indexed archive of 269 Lenny's Podcast episodes

Lenny's Podcast

The agents fetch episode transcripts from Lenny's Podcast — hosted by Lenny Rachitsky — at each step of the framework. All transcript content belongs to Lenny Rachitsky and the respective guests.

Frameworks and methodologies

The external skill repos draw on the following established frameworks:

Product strategy & market research — Jobs to Be Done (Clayton Christensen, Tony Ulwick, Bob Moesta) · Opportunity Solution Tree (Teresa Torres) · Business Model Canvas (Alexander Osterwalder) · Lean Canvas (Ash Maurya) · Porter's Five Forces (Michael E. Porter)

User research & discovery — Discovery Interview Methodology (Steve Blank) · Empathy Mapping (Dave Gray) · Customer Journey Mapping

Validation & experimentation — Pretotyping (Alberto Savoia) · Product-Market Fit Survey (Sean Ellis) · Build–Measure–Learn (Eric Ries)

Pricing — Van Westendorp Price Sensitivity Meter · Value-Based Pricing (Madhavan Ramanujam)

Growth & go-to-market — Growth Loops (Brian Balfour) · North Star Metric (Sean Ellis) · OKRs (Andy Grove / John Doerr)

Execution — Pre-mortem (Gary Klein) · MoSCoW prioritization · Agile Sprints (Ken Schwaber)

Ver o original no GitHub

Bom para

Relacionados