🎯 PMF Kit

Discover and validate product-market-fit faster with AI agents.

A spec-driven toolkit for systematic PMF discovery and validation of AI SaaS products, built on the foundations of spec-kit.

PMF-Kit is a specialized variant of Spec-Kit, adapted for product-market-fit (PMF) discovery and validation of AI SaaS products.

While Spec-Kit enables spec-driven software development with AI agents, PMF-Kit applies the same methodology to the domain of product management, market research, and customer validation—helping founders and PMs discover PMF through structured, AI-assisted workflows instead of ad-hoc experimentation.

Choose your preferred installation method:

Run directly without installing—always uses the latest version:

uvx --from git+https://github.com/agentii-ai/pmf-kit.git pmf init my-product
uvx --from git+https://github.com/agentii-ai/pmf-kit.git pmf check

Note: This project is improving rapidly. We recommend uvx to always get the latest features and fixes.

Running pmf init shows an interactive wizard to select your AI assistant

Install once and use everywhere (may require periodic updates):

uv tool install pmf-cli --from git+https://github.com/agentii-ai/pmf-kit.git

The pmf executable is installed globally for use anywhere

Then use the tool directly:

pmf init my-product
pmf check

To update to the latest version:

uv tool install pmf-cli --force --from git+https://github.com/agentii-ai/pmf-kit.git

pmf init my-ai-product
cd my-ai-product

This creates a project with PMF-specific templates, constitution, and agent commands.

After setup completes, you'll see all available slash commands and next steps

Open your AI assistant (Claude Code, Cursor, Windsurf, etc.) in the project directory. You'll see /pmfkit.* commands available:

/pmfkit.constitution    # Establish PMF-specific principles
/pmfkit.pmf         # Define what you're trying to learn
/pmfkit.clarify         # Resolve ambiguities in your hypothesis
/pmfkit.plan            # Create research execution plan
/pmfkit.tasks           # Generate actionable research tasks
/pmfkit.implement       # Execute PMF discovery workflow

Claude Code automatically detects all /pmfkit.* slash commands in your project

/pmfkit.pmf Validate willingness to pay for AI-powered contract review among solo lawyers, following Harvey's market approach

This generates a PMF specification with:

• Sharp personas (role, company, tools, environment)
• Jobs-to-be-done with current workarounds
• Hero workflows (intent → AI work → artifact)
• PMF success metrics (activation, engagement, AI-specific, business)
• Constraints, risks, and distribution hypotheses

PMF-Kit works with all agents supported by Spec-Kit:

PMF-Kit provides project templates for 18 AI coding agents, automatically downloaded when you run pmf init. Each template includes:

• PMF-Kit Constitution v1.0.0 with 7 PMF-specific principles
• Workflow templates: spec.md, plan.md, tasks.md for research documentation
• 9 slash commands: /pmfkit.pmf, /pmfkit.plan, /pmfkit.tasks, /pmfkit.implement, /pmfkit.clarify, /pmfkit.analyze, /pmfkit.checklist, /pmfkit.taskstoissues, /pmfkit.constitution
• Scripts: Bash or PowerShell variants for automation
• Memory system: constitution.md for project-specific principles

Templates are available for all 18 agents in both bash and PowerShell variants (36 total):

• Claude Code • Cursor Agent • Windsurf • Google Gemini
• GitHub Copilot • Qoder • Qwen • OpenCode
• Codex • KiloCode • Auggie • CodeBuddy
• AMP • Shai • Amazon Q • Bob • Roo

See GitHub Releases for downloadable template archives with SHA-256 checksums.

Define WHAT you're trying to learn and WHY:

• Target personas (role/skill, company, tools, environment)
• Top 3 jobs-to-be-done (customer problems)
• 1-2 hero workflows (end-to-end AI-native flows)
• Success metrics (activation, engagement, AI-specific, business)
• Constraints & risks
• Distribution hypotheses

Resolve ambiguities before committing to research:

• Validate persona sharpness
• Clarify JTBD and hypotheses
• Define success metrics precisely
• Identify unstated assumptions

Define HOW you'll execute discovery:

• Research methodology (interviews, surveys, experiments)
• Sample sizes and recruitment strategy
• Evidence collection instruments
• Analysis approach
• Validation checkpoints & PDCA cycles

Generate actionable research tasks:

• Recruit participants from target segment
• Conduct hero workflow research
• Analyze behavioral data
• Validate hypotheses
• Document learnings & pivot/persevere decisions

Execute research systematically with AI assistance:

• Collect customer evidence (interviews, behavioral data)
• Measure PMF signals (retention, TTFW, completion rates)
• Validate go/no-go criteria per phase
• Document decisions and evidence

PMF-Kit is built on 7 core principles that guide all discovery work:

Define hypotheses and success criteria before running experiments.

Support all PMF claims with direct customer evidence—not opinions or assumptions.

Follow build-measure-learn cycles with independent, testable increments.

Use the simplest approach that achieves the learning objective.

Integrate insights from product, engineering, sales, marketing, and customer success.

Enable multiple kit variants (pmf-kit, pd-kit, marketing-kit) to coexist without conflicts.

Serve as a reference implementation for creating domain-specific kit variants.

See memory/constitution.md for full details.

PMF-Kit includes comprehensive reference materials to guide your discovery:

• refs/0_overview.md - Overview of PMF discovery for AI SaaS products
• refs/1_principles_for_constitution.md - PMF-specific principles and patterns
• refs/2_define_for_specify.md - How to structure sharp PMF specifications
• refs/3_project_management_for_plan.md - Research planning methodology
• refs/4_pm_tasking_for_tasks.md - PMF discovery task patterns
• refs/instructions.md - How to create your own kit variants

PMF-Kit is designed to coexist with Spec-Kit and other kit variants:

# Install Spec-Kit for software development
uv tool install specify-cli --from git+https://github.com/github/spec-kit.git

# Install PMF-Kit for product-market-fit discovery
uv tool install pmf-cli --from git+https://github.com/agentii-ai/pmf-kit.git

# Both tools work independently
specify check    # Shows Spec-Kit configuration
pmf check        # Shows PMF-Kit configuration

# Create projects with different kits
specify init my-feature           # Software feature project
pmf init my-product             # PMF discovery project

In your AI agent, both command namespaces are available:

• /speckit.* commands for software development workflows
• /pmfkit.* commands for PMF discovery workflows

pmf init <PROJECT_NAME>
pmf init my-product --ai claude
pmf init . --here --force        # Initialize in current directory
pmf init my-product --ai cursor --script ps   # PowerShell scripts

Options:

• --ai - Specify AI assistant (claude, cursor, windsurf, gemini, etc.)
• --script - Script variant (sh for bash/zsh, ps for PowerShell)
• --here - Initialize in current directory
• --force - Skip confirmation when directory has files
• --no-git - Skip git initialization
• --ignore-agent-tools - Skip tool availability checks

pmf check

Verifies PMF-Kit installation and checks for required tools (git, claude, cursor, windsurf, etc.).

pmf init ai-code-assistant
/pmfkit.pmf "Validate demand for AI-powered coding assistance among backend engineers, similar to Cursor's approach"

Expected artifacts:

• Personas: Backend engineers at 50-500 person SaaS
• JTBD: "When writing boilerplate code, I want AI to generate it, so I can focus on logic"
• Hero workflow: Intent (write comment) → AI generates → review → merge
• Success metric: Time-to-first-completion, edit distance, retention

pmf init video-generation
/pmfkit.pmf "Validate demand for text-to-video generation among YouTube creators"

Expected artifacts:

• Personas: Content creators with 10k-100k subscribers
• JTBD: "When I need to produce weekly videos, I want to cut editing time by 80%"
• Hero workflow: Upload footage → AI edits with templates → review → export
• Success metric: Time-to-first-render, edit satisfaction, repeat usage

pmf init contract-ai
/pmfkit.pmf "Validate willingness to pay for AI-powered legal analysis among solo practitioners and small law firms"

• PMF-Kit Specification - Full feature specification
• PMF-Kit Implementation Plan - Technical implementation details
• Spec-Kit Repository - Upstream project for software development
• Spec-Driven Development Methodology - Core methodology

The next major feature for PMF-Kit is an automated optimization module that evaluates and improves template quality using multi-judge LLM evaluation and optimization algorithms.

• Multi-Judge Consensus: 3 LLM judges (GPT-4o strict, Claude balanced, Gemini lenient) with Bradley-Terry aggregation
• 8-Dimensional Rubric: Correctness, Coherence, Instruction-Following, Completeness, Specificity, Clarity, Actionability, Policy-Adherence
• Quantified Improvements: Target +15-25% quality improvement with statistical significance (p < 0.05)
• CLI + Agent Support: pmf optimize <target> and /pmfkit.optimize slash command

# CLI usage
pmf optimize .pmf/templates/spec-template.md --mode=full

# Agent command
/pmfkit.optimize specs/my-feature/spec.md

See specs/003-workflow-optimization for full specification and implementation plan.

pmf-kit/
├── .claude/commands/          # Claude Code slash commands
│   ├── constitution.md
│   ├── specify.md
│   ├── plan.md
│   ├── tasks.md
│   ├── implement.md
│   ├── clarify.md
│   ├── analyze.md
│   └── checklist.md
├── memory/
│   └── constitution.md        # PMF-Kit principles (v1.0.0)
├── templates/
│   ├── spec-template.md       # PMF specification template
│   ├── plan-template.md       # PMF research planning template
│   ├── tasks-template.md      # PMF task breakdown template
│   ├── checklist-template.md  # PMF quality validation template
│   └── commands/              # Agent command templates
├── scripts/
│   ├── bash/                  # Bash automation scripts
│   └── powershell/            # PowerShell automation scripts
├── refs/                      # Reference documentation
│   ├── 0_overview.md
│   ├── 1_principles_for_constitution.md
│   ├── 2_define_for_specify.md
│   ├── 3_project_management_for_plan.md
│   ├── 4_pm_tasking_for_tasks.md
│   └── instructions.md        # How to create kit variants
└── specs/
    └── 001-pmf-kit-variant/   # Feature specification & implementation docs

• Linux/macOS/Windows
• Supported AI coding agent
• uv for package management
• Python 3.11+
• Git

Run pmf check to verify all prerequisites and see which AI agents are available:

pmf check

The pmf check command shows all detected tools and AI agents

PMF-Kit demonstrates how to adapt spec-driven methodology to any domain. Want to create a variant for product design, marketing, or business writing?

See refs/instructions.md for a comprehensive guide on:

• How to fork and adapt spec-kit for your domain
• How to define domain-specific principles
• How to create templates and reference materials
• How to enable multi-kit coexistence

Example variants:

• pd-kit - Product design and UX workflows
• marketing-kit - Go-to-market and growth campaigns
• writing-kit - Technical and business writing
• ops-kit - Operations and project management

All variants are published at kits.agentii.ai.

PMF-Kit is built on the excellent work of the Spec-Kit project from GitHub. We preserve 100% of Spec-Kit's architecture and infrastructure while adapting templates and methodology for PMF discovery.

Spec-Kit Credits:

• Den Delimarsky
• John Lam

For issues, questions, or feedback:

• GitHub Issues: Report on PMF-Kit
• Spec-Kit Issues: Report on Spec-Kit
• Kit Variants: Visit kits.agentii.ai

This project is licensed under the terms of the MIT open source license. See LICENSE for details.

Note: PMF-Kit extends Spec-Kit's MIT license. For Spec-Kit license details, see Spec-Kit LICENSE.

• Systematic: Replace vibe-based PMF discovery with structured, hypothesis-driven research
• AI-Assisted: Leverage AI agents for specification, planning, and evidence analysis
• Validated: Reference templates based on proven PMF patterns from Cursor, Runway, Harvey, Writer, and other successful AI products
• Evidence-Driven: Focus on customer evidence and measurable PMF signals, not opinions

• Reproducible: Spec-driven workflows are more transparent and collaborative than ad-hoc processes
• Extensible: PMF-Kit serves as a reference for creating domain-specific kit variants
• Community-Friendly: All templates and reference materials are open source and MIT-licensed
• Professional: Built on proven Spec-Kit infrastructure, adapted by experienced product leaders

Ready to discover product-market-fit with confidence?

pmf init my-product

Let's build products customers love, with evidence guiding every decision.