AI Native PM Agent

An AI product coach that asks "Is this direction really worth pursuing?" — with structured methodology backing every decision, from spark of inspiration to production deployment.

📖 中文版 README

Why Do You Need This?

90% of AI product teams die in the same traps:

• Direction Trap: Spend 3 months building an AI feature, only to find users won't pay for it
• Needs Trap: Fake needs look too much like real ones — AI makes prototyping near-free, but also lets you build the wrong thing faster
• Boundary Trap: AI crosses the line and does things it shouldn't, triggering compliance risks
• Hallucination Trap: Pre-launch accuracy looks like 95%, post-launch reality says otherwise
• Cost Trap: Token bills explode, business model falls apart

This Agent doesn't write code for you — it makes you pause at every critical decision point and verify with structured methods.

Why Not Just Use Traditional Product Methodology?

Traditional product frameworks (Lean Startup, Jobs-to-be-Done, Design Thinking) were built for a world where prototyping was expensive and AI didn't exist. They break down in the AI era because:

This methodology is AI Native from the ground up: it starts with boundary design before capability design, validates with certainty rather than confidence, and prices on risk reduction rather than feature count. Every stage assumes AI is in the loop — and designs for what happens when it goes wrong.

See What It Does in 30 Seconds

Core Capabilities: 95 executable Skills across 8 stages + Stage auto-routing + Conflict detection + Evidence chain tracking

A Concrete Example

Scenario: You want to build an "AI Contract Review Assistant"

Needs Discovery (P0)

The Agent validates the need with tool cards:

• Micro-Needs Five Questions: Lawyers review contract clauses daily — the pain is small but constant
• Real-Needs Validation: Long-standing problem + compensation behaviors (manual annotation) + structural root cause (liability risk)
• Four-Layer Decomposition: Surface: "automate review" → Situation: "lawyers bear liability risk" → Cost: "$200/hour per review"
• Agent Boundary Checklist: AI can flag risky clauses, but cannot determine contract validity

Output: Needs validated + Agent boundary design

Direction Framing (P1)

The Agent asks you:

• Where does contract data come from? (Availability)
• Does it involve client confidentiality? (Desensitization)
• Who is responsible for review results? (Authorization)
• Is the output format standardized? (Structured)
• What happens when new regulations emerge? (Sustained supply)

Output: Direction Brief — clear go/no-go with conditions

Business Model (P6)

The Agent prices using the Certainty Premium formula:

• Fear Level: Lawyers' biggest fear is missing a risky clause → High
• Error Cost: Missing one clause could mean millions in liability → Extremely high
• Substitution Cost: Manual review at $200/hour → Medium
• Recommended Model: Insurance Mode (charge per successful review, compensate for misses)

Output: Pricing strategy — $5/review, 10x compensation for missed clauses

Audit & Release (P9)

The Agent checks:

• Reliability: Identification accuracy, hallucination rate
• Safety: Sensitive information handling
• Boundaries: Which clause types require human review
• Cost: Token cost per review vs. pricing

Output: Release boundary document — Auto-execute zone / Human handoff zone / Disabled zone

Quick Start

Option 1: One-Click Install (Recommended)

# Install all 95 skills + orchestrator
curl -fsSL https://raw.githubusercontent.com/gmaxxxie/ai-native-product-agent-skills/main/install.sh | bash

# Start a product project
hermes run "I want to build an AI customer service product, help me start from direction framing"

Option 2: Install from GitHub URL

Install individual skills on demand:

# Orchestrator (entry point)
hermes skills install \
  https://raw.githubusercontent.com/gmaxxxie/ai-native-product-agent-skills/main/orchestrator/SKILL.md \
  --name ai-native-pm-agent

# Any individual skill
hermes skills install \
  https://raw.githubusercontent.com/gmaxxxie/ai-native-product-agent-skills/main/skills/ai-native-direction-framing/SKILL.md \
  --name ai-native-direction-framing

Option 3: Clone & Local Install

git clone https://github.com/gmaxxxie/ai-native-product-agent-skills.git
cd ai-native-product-agent-skills
bash install.sh   # copies all skills to ~/.hermes/skills/ai-native-pm/

Option 4: Use with Other AI Agents

These skills work as structured prompts — they're not tied to any specific agent framework.

The simplest way: just tell your AI agent to install from this repo.

💡 Tip: Claude Code, Codex, and OpenCode can all git clone the repo and read SKILL.md files directly. Just give them the repo URL and tell them to install — they'll figure out the rest.

Per-Stage Usage

Each stage is an independent Skill you can call individually:

Cross-Stage Combos (One-Stop)

Complete Skill List (95 Skills)

P0 — Needs Discovery Layer (17 Skills)

P1–P2 — Direction & Experiment Layer (8 Skills)

P3–P4 — System Building Layer (5 Skills)

P5–P6 — Business Model Layer (5 Skills)

P7 — Growth Strategy Layer (6 Skills)

P8 — User Experience Layer (4 Skills)

P9–P11 — Audit, Operations & Team Layer (9 Skills)

P12 — Contemplation Layer (10 Skills)

From: Contemplation — Product Judgment, User Understanding, and Decision Correction in the AI Era

P13 — Judgment & Intuition Layer (12 Skills)

From: Intuition — Judgment and Intuition in the AI Era

P14 — Aesthetic Authority Layer (9 Skills)

From: AI Beaty — Aesthetic Authority in the Age of AI

Cross-Book Combo Skills (3 Skills)

Project Structure

ai-native-pm-agent-skills/
├── README.md / README_CN.md         # This document (EN / 中文)
├── ARCHITECTURE.md                  # System architecture design
├── skill-registry.yaml              # Skill registry (95 skills registered)
├── orchestrator/SKILL.md            # Main orchestrator: stage routing + conflict detection
├── install.sh                       # One-click install script
├── assets/                          # Hero banner, pipeline flow, industry matrix, methodology books
├── skills/
│   ├── p0-needs-orchestrator/        # P0 Needs Discovery orchestrator
│   ├── p0-product-needs/             # P0 unified needs discovery
│   ├── p0a-micro-needs-detector/     # P0a micro-needs detection
│   ├── p0b-real-needs-validator/     # P0b real vs fake needs
│   ├── p0c-needs-decomposer/        # P0c four-layer decomposition
│   ├── p0d-needs-archaeologist/     # P0d deep needs archaeology
│   ├── p0e-good-question-generator/ # P0e good questions six dimensions
│   ├── p0f-agent-boundary-designer/ # P0f AI boundary design
│   ├── p0g-diverse-recommendation-rewriter/    # P0g diversity rewrite
│   ├── p0g-diversity-rewrite-checklist/       # P0g diversity checklist
│   ├── p0h-ai-product-triple-balance/         # P0h triple balance
│   ├── p0h-triple-balance-assessor/           # P0h balance assessor
│   ├── ai-native-direction-framing/        # P1 direction framing
│   ├── p2-experiment-engine/        # P2 experiment overview
│   ├── ai-native-experiment-engine/     # P2a experiment setup
│   ├── p2b-product-form-exploration/ # P2b product form
│   ├── p2c-process-redesign/       # P2c process redesign
│   ├── p2d-convergence-decision/    # P2d convergence decision
│   ├── p2e-shadow-validation/       # P2e shadow validation
│   ├── ai-native-system-building/          # P3 system building
│   ├── ai-native-agent-skill-design/       # P4 agent & skill design
│   ├── p5-memory-system/           # P5 memory system
│   ├── p6-context-engineering/      # P6 context engineering
│   ├── p7-knowledge-rag/            # P7 RAG & knowledge
│   ├── p6-business-model/           # P6 business model overview
│   ├── p6a-certainty-premium-calculator/    # P6a certainty premium
│   ├── p6b-arbiter-mode-designer/  # P6b arbiter mode
│   ├── p6c-insurance-mode-designer/ # P6c insurance mode
│   ├── p6d-prediction-arbitrage-designer/   # P6d prediction arbitrage
│   ├── p7-marketing-growth/         # P7 marketing overview
│   ├── p7a-data-flywheel-builder/   # P7a data flywheel
│   ├── p7b-intent-prediction-designer/      # P7b intent prediction
│   ├── p7c-predictive-retention-designer/   # P7c predictive retention
│   ├── p7d-marketing-productizer/   # P7d marketing productizer
│   ├── p7e-customer-loop/           # P7e customer loop
│   ├── ai-native-ux-design/               # P8 UX design overview
│   ├── p8a-rax-risk-assessor/       # P8a RAX risk assessment
│   ├── p8b-trust-tier-designer/     # P8b trust tier design
│   ├── p8c-progressive-disclosure/  # P8c progressive disclosure
│   ├── ai-native-audit-release/           # P9 audit & release
│   ├── ai-native-production-ops/          # P10 production operations
│   ├── p10a-value-discovery-loop/  # P10a value discovery loop
│   ├── p10b-aiops-case/            # P10b AIOps case
│   ├── p10c-customer-service-case/  # P10c AI customer service case
│   ├── p10d-saas-case/              # P10d SaaS case
│   ├── p11-product-team/           # P11 product team design
│   ├── p12-contemplation-orchestrator/     # P12 Contemplation orchestrator
│   ├── p12a-contemplation-right-view/      # P12 right view
│   ├── p12a-contemplation-view-correction/ # P12 view correction
│   ├── p12a-contemplation-prerequisite-check/   # P12 prerequisite
│   ├── p12a-contemplation-right-thinking/    # P12 right thinking
│   ├── p12a-contemplation-right-speech/      # P12 right speech
│   ├── p12a-contemplation-right-action/      # P12 right action
│   ├── p12a-contemplation-right-livelihood/   # P12 right livelihood
│   ├── p12a-contemplation-right-effort/       # P12 right effort
│   ├── p12a-contemplation-right-mindfulness/  # P12 right mindfulness
│   ├── p13-intuition-orchestrator/    # P13 Intuition orchestrator
│   ├── p13a-judgment-metacognition/   # P13a judgment metacognition
│   ├── p13b-systemic-thinker/        # P13b systemic thinking
│   ├── p13c-product-psychology/       # P13c product psychology
│   ├── p13d-intuition-training/      # P13d intuition training
│   ├── p13e-nine-step-framework/     # P13e nine-step overview
│   ├── p13f-first-half-judgment/     # P13f first half
│   ├── p13g-mid-judgment/            # P13g mid judgment
│   ├── p13h-validation-market/       # P13h validation & market
│   ├── p13i-judgment-traps/          # P13i judgment traps
│   ├── p13j-organizational-judgment/  # P13j organizational judgment
│   ├── p13k-intuition-evolution/      # P13k intuition evolution
│   ├── p14-beauty-orchestrator/       # P14 Beauty orchestrator
│   ├── p14a-beauty-redefinition/     # P14a aesthetic redefinition
│   ├── p14b-beauty-ai-roles/         # P14b AI's role in aesthetics
│   ├── p14c-beauty-selection/         # P14c selection over generation
│   ├── p14d-beauty-narrative/        # P14d narrative aesthetics
│   ├── p14e-beauty-human-edge/        # P14e human edge
│   ├── p14f-beauty-commercial/        # P14f commercial value
│   ├── p14g-beauty-system/            # P14g training system
│   ├── p14h-beauty-preface/          # P14h core proposition
│   ├── combo-needs-to-direction/     # Combo: pain point → direction
│   ├── combo-business-to-growth/     # Combo: pricing → growth
│   └── combo-ux-to-audit/            # Combo: UX → audit
└── scripts/
    ├── init_product_context.py       # Product context initialization
    ├── test_orchestrator.py         # Orchestrator tests
    └── final_validation.py          # Final validation

Eight Books Behind the Methodology

All 95 Skills are derived from eight methodology books. Each book's tool cards and concept cards have been converted into executable Skills:

Quality Stats (as of May 2026)

📖 Available on Amazon (8 published):

• Micro-Needs for AI Products — Needs discovery, micro-needs detection, real-needs validation, needs decomposition, agent boundary design
• AI Native Product Methodology — Direction framing, experiment engine, system building, audit & release, production operations
• JUDGMENT: How to Make Better AI Product Decisions — Certainty Premium, business model design, pricing strategy
• Contemplation: Product Judgment, User Understanding, and Decision Correction in the AI Era — Right view, prerequisite checks, judgment correction, decision mindfulness
• Aesthetic Authority: Why Human Judgment and Taste Matter in the Age of AI — Data flywheel, intent prediction, predictive retention, marketing-as-product

✅ All 8 books published on Amazon — 95 executable Skills derived from the complete methodology library.

Industry Scenarios

Design Principles (Why We Designed It This Way)

1. Problem before solution — Validate that the problem is real before building features
2. Boundaries before capabilities — Define what AI shouldn't do before designing what it can
3. Evidence before decisions — Replace "I think it works" with Shadow validation
4. Orchestration before automation — Keep human confirmation at critical decision points
5. Iteration before perfection — Optimize through failure analysis, not first-time perfection
6. View correction before action — Check if you're asking the right question before answering
7. Judgment before intuition — Make the reasoning explicit before trusting gut feel
8. Aesthetic authority over feature completeness — What you choose not to build defines the product

Contributing

Issues and PRs are welcome! Priority areas:

• New scenarios: Add industry cases with complete input-output examples
• Boundary designs: How to draw AI boundaries in high-risk scenarios
• Failure cases: Failed experiment analyses are more valuable than success stories
• New tool cards: Convert book concept cards into executable Skills

⭐ Star History

License

MIT License

"Problem before solution. Boundaries before capabilities. Evidence before decisions. Orchestration before automation."