A conversational AI agent that guides product managers and product leaders through structured business idea validation using proven methodologies like Lean Canvas and SVPG Product Discovery.
This agent helps you:
- Define the problem you're solving clearly with Socratic questioning
- Identify target customers specifically (no "everyone" or "small businesses")
- Validate your value proposition and differentiation with evidence
- Assess market opportunity and competition (with research capability)
- Design business model and pricing based on data
- Identify critical risks and assumptions to test first
- Generate a one-page validation canvas as output with cited sources
- ✅ Structured approach: Follows proven validation frameworks (Lean Canvas, SVPG, Socratic Questioning)
- ✅ Socratic questioning: Uses 5 question categories to uncover assumptions and clarify thinking
- ✅ Research capability: Can research market data, competitor pricing, and industry benchmarks when you need it
- ✅ Prevents hallucination: Only uses information you provide or confirmed research data
- ✅ Challenges assumptions: Pushes for clarity and evidence without being interrogative
- ✅ Brief and direct: Executive-level communication (2-3 sentences max)
- ✅ Actionable output: One-page canvas with next steps and cited sources
- ✅ Risk-focused: Identifies riskiest assumptions to test first
git clone <repository-url>
cd idea-validationRead /frameworks/lean-canvas.md to understand the methodology.
The agent guides you through 8 validation skills:
- Problem Definition - What problem are you solving?
- Customer Segments - Who has this problem?
- Value Proposition - What makes you different?
- Solution Overview - What will you build?
- Market Validation - How big is the opportunity?
- Business Model - How will you make money?
- Key Metrics - How will you measure success?
- Risks & Assumptions - What could go wrong?
After completing all skills, the agent generates a one-page validation canvas saved to /outputs/[your-idea]-validation-canvas.md.
Brief, Helpful Responses
Agent: "What problem does your idea solve?"
User: "We make project management easier."
Agent: "Too vague. What SPECIFIC problem in project management?"
Socratic Probing
User: "Sales teams waste time on admin work."
Agent: "What admin work specifically? How much time?"
User: "Updating CRM from email threads. About 5 hours per week."
Agent: "✓ Quantified. Who experiences this most acutely?"
Research Capability (NEW)
User: "I don't know the market size."
Agent: "Let me research that for you."
[Uses WebSearch]
Agent: "Found global CRM market is $65B (Gartner 2024). Your segment (email automation)
is ~$8B. Does this match your understanding?"
User: "Yes, sounds right."
Agent: "✓ Using $8B TAM. Moving on."
No Hallucination
Agent: "Need competitor pricing data. Should I research it?"
(Instead of making up numbers)
- 2-3 sentences max per response
- Socratic questioning to uncover assumptions and clarify thinking
- Researches market data when you lack critical information
- Challenges vagueness but guides toward better answers
- Marks unknowns as "TBD - Requires validation" or researches them
- Cites sources for all researched data
- Provides specific next action after completion
/
├── README.md # This file
├── agent/
│ ├── system-prompt.md # Core agent instructions
│ ├── behavior-guidelines.md # Detailed behavior rules
│ ├── agent-config.md # Configuration
│ ├── implementation-guide.md # How to implement
│ └── skills/ # 9 conversation skills
│ ├── 01-problem-definition.md
│ ├── 02-customer-segments.md
│ ├── 03-value-proposition.md
│ ├── 04-solution-overview.md
│ ├── 05-market-validation.md
│ ├── 06-business-model.md
│ ├── 07-key-metrics.md
│ ├── 08-risks-assumptions.md
│ └── 09-generate-output.md
├── frameworks/
│ ├── lean-canvas.md # Lean Canvas guide
│ ├── opportunity-assessment.md # SVPG framework
│ ├── socratic-questioning.md # Socratic methodology (NEW)
│ └── comparison.md # When to use which
├── templates/
│ ├── lean-canvas-template.md # Output template
│ ├── opportunity-assessment-template.md
│ └── output-onepager-template.md
├── outputs/ # Generated canvases
└── docs/
├── methodology-guide.md
├── getting-started.md
├── socratic-integration-guide.md # How Socratic questioning works (NEW)
└── agent-improvements-2026-01.md # Recent enhancements (NEW)
This agent combines three proven methodologies:
- Created by: Ash Maurya
- Purpose: One-page business model for startups
- When: Validating new business ideas
- Output: 9-block canvas covering problem, solution, customers, and business model
- Created by: Marty Cagan (Silicon Valley Product Group)
- Purpose: Evaluate product opportunities
- When: Existing products deciding what to build next
- Output: 10-question assessment with go/no-go recommendation
- Purpose: Uncover assumptions and clarify fuzzy thinking
- Method: 5 question categories (Problem Clarity, Solution Validation, Success Criteria, Constraints, Strategic Fit)
- When: Integrated throughout all validation conversations
- Impact: Transforms validation from form-filling into rigorous, evidence-based thinking
- See:
/frameworks/socratic-questioning.mdfor complete framework
The agent uses Lean Canvas as default output because:
- Most users are validating new ideas (pre-product)
- Covers full business model including revenue/costs
- One-page format is immediately actionable
- Widely recognized and understood
Socratic Questioning enhances all frameworks by ensuring:
- Specific, concrete problem statements (not abstract)
- Evidence-based claims (not assumptions)
- Clear success criteria with metrics
- Explicit scope boundaries
- Testable assumptions
See /frameworks/comparison.md for detailed comparison.
Agent: What problem does your idea solve?
User: We're building a project management tool.
Agent: What specific problem does it solve that existing tools don't?
User: Dev teams can't track dependencies across projects easily.
Agent: ✓ Got it. Problem: tracking cross-project dependencies for dev teams.
Who specifically experiences this?
User: Engineering managers at mid-size tech companies.
Agent: How mid-size? How many engineers? What industry?
User: 50-200 engineers. B2B SaaS companies.
Agent: ✓ Clear. Target: Engineering managers at B2B SaaS companies, 50-200 engineers.
Moving to value proposition.
...
[After completing all skills]
Agent: Your validation canvas is ready: dependency-tracker-validation-canvas.md
**Riskiest Assumption**: Engineering managers will pay $99/month
**Next Step**: Interview 10 engineering managers to validate problem and pricing
The agent generates a comprehensive one-pager:
# Lean Canvas: Cross-Project Dependency Tracker
## Problem
1. Dev teams lose hours tracking dependencies manually
2. Delays cascade when dependencies aren't visible
3. No single source of truth for cross-project status
## Customer Segments
**Primary**: Engineering Managers at B2B SaaS companies (50-200 engineers)
**Early Adopters**: EMs currently using spreadsheets or Jira add-ons
## Unique Value Proposition
Real-time dependency mapping across all projects without manual updates
## Solution
1. Auto-detect dependencies from Git/Jira
2. Visual dependency graph
3. Automated alerts when blockers arise
...
## Riskiest Assumption
Engineering managers will pay $99/month per team
## Next Steps
1. Interview 10 engineering managers
2. Test pricing with 5 prospects
3. Build basic dependency detection prototypeSee /agent/implementation-guide.md for detailed instructions on:
- How to load agent components
- State management
- Skill progression logic
- Output generation
- Platform-specific implementations (Claude, GPT, etc.)
Your AI platform needs:
- System prompt support (to load agent instructions)
- Conversation history (to maintain context)
- File generation (to save output)
- State persistence (to track skill progression)
Edit templates in /templates/:
lean-canvas-template.md- Default outputopportunity-assessment-template.md- SVPG formatoutput-onepager-template.md- Executive summary
Edit files in /agent/skills/ to:
- Add/remove questions
- Change validation criteria
- Adjust behavior for that skill
- Create new template in
/templates/ - Document framework in
/frameworks/ - Update agent config to support it
- Be specific - Generic answers get challenged
- Quantify when possible - Numbers are better than adjectives
- Mark unknowns - It's OK to say "TBD, need to research"
- Test assumptions - Use the output to run experiments
- Don't let agent hallucinate - Enforce evidence-based responses
- Maintain brevity - Max 2-3 sentences per response
- Challenge vagueness - Push for specific, actionable answers
- Highlight risks - Focus on what could invalidate the idea
We welcome contributions:
- Skills: Improve questions or validation criteria
- Templates: Add new output formats
- Frameworks: Document additional methodologies
- Documentation: Clarify or expand guides
- Lean Canvas - Official Lean Canvas site
- SVPG - Silicon Valley Product Group articles
- "Running Lean" by Ash Maurya - Book on Lean Canvas
- "Inspired" by Marty Cagan - Book on SVPG product discovery
- "The Mom Test" - How to talk to customers
- "Testing Business Ideas" - Validation experiments
- "Lean Startup" - Eric Ries' foundational book
- "Value Proposition Design" - Customer-centric validation
- Socratic Method - Wikipedia overview
/frameworks/socratic-questioning.md- Complete framework in this repo/docs/socratic-integration-guide.md- Practical integration guide
Q: How long does validation take? A: 20-60 minutes for a complete canvas, depending on how prepared you are.
Q: Can I skip skills? A: Yes, though we recommend completing all for comprehensive validation.
Q: What if I don't know the answer? A: Mark it as "TBD - Requires validation" and make it a test priority.
Q: Can I update my canvas later? A: Yes! The canvas is a living document. Update as you learn.
Q: Do I need to validate everything before building? A: No. Validate the riskiest assumptions first, then iterate.
Q: Is this for B2B or B2C? A: Both. The frameworks work for any business model.
Q: Can the agent help me research market data? A: Yes! If you don't know critical information (market size, competitor pricing, industry benchmarks), the agent can research it using WebSearch and cite sources. It will NOT research your specific problem or solution—those require your domain expertise.
[Specify your license]
- Documentation: See
/docs/folder - Issues: GitHub Issues
- Questions: Discussion Forum
Ready to validate your idea? Start by reading /docs/getting-started.md then engage with the agent!