PM AI Partner Framework

The Problem

Platform PMs face a unique challenge: the truth lives in code, but strategy lives in docs.

What Goes Wrong

Inherited teams with unclear identity
Strategy claims capabilities that don't exist
Same questions asked repeatedly
Codebase understanding takes months
Docs compete with operations

Traditional Approaches

Read docs → Form opinions → Write more docs → Hope you didn't miss something → Iterate slowly → Context lives in people's heads

        // The Opportunity: AI changes what's possible

        Without System → AI writes generic docs, context lost, ad-hoc prompting

        With System    → AI explores you decide, knowledge compounds, repeatable workflows

Four Sources of Truth

PM work is about reconciling different sources — AI helps explore, humans decide.

💻

Code

"What CAN happen?"

APIs, contracts, implementations, patterns

📄

Docs

"What's EXPECTED?"

Strategy, RFCs, handovers, context

📊

Data

"What IS happening?"

Metrics, analytics, A/B tests, research

🧠

Judgment

"What SHOULD happen?"

Intuition, values, experience, priorities

        // Where conflicts happen — the interesting part

        Code ≠ Docs    → "We claim to do X but code shows Y"

        Data ≠ Docs    → "Strategy says X matters, metrics show Y"

        Code ≠ Data    → "Feature exists but isn't being used"

        Judgment ≠ Data → "I think X, but data shows Y"

The Plugin

One command installs 12 skills, 6 commands, and 3 hooks into your editor.

$ npx pm-ai-partner@latest

Workflow Commands

Multi-step workflows for recurring PM tasks — not single prompts.

Command	What It Does
`/pm:plan-week`	Reviews context, sets priorities, flags risks
`/pm:write-brief`	Guided PRD creation with quality checklist
`/pm:prep-meeting`	Talking points, anticipated questions, role-play
`/pm:audit-codebase`	PM-focused code exploration — capabilities, gaps, implications
`/pm:competitor-scan`	Competitive analysis with positioning maps and action recommendations
`/pm:metrics-review`	Structured metrics review — headlines, deep dives, action items

Automation Hooks

Run in the background — no manual invocation needed.

Hook	Trigger	What It Does
Quality Gate	Write to `product-catalog/`	Checks doc against quality bar (evidence, clarity, audience)
Draft Status	Write to `sandbox/`	Reminds to add status markers to drafts
Commit Reminder	`git commit`	Prompts to capture lessons in CLAUDE.md

12 Agent Skills

Different Claude "modes" for different jobs — invoke by name or let them auto-activate.

🤝

Thought Partner

Brainstorm, explore options, surface assumptions

🔍

Technical Analyst

Translate code and systems into PM-friendly insights

✍️

Writer

Draft briefs, updates, emails — concise and structured

😈

Devil's Advocate

Stress-test proposals, anticipate objections

🛠️

Builder

Create tools, scripts, dashboards, prototypes

📊

Data Analyst

Write queries, analyze metrics, extract insights

📋

Product Brief

Structured PRDs and feature specs

🎯

Meeting Prep

Talking points, anticipated questions

📢

Stakeholder Update

Status reports and executive summaries

🧭

Strategic Clarity

Define team identity, boundaries, and charter

🧪

Hypothesis Tester

Design experiments, evaluate A/B tests, interpret results

🏁

Competitor Analyst

Map landscapes, assess positioning, identify real gaps

Workflow Patterns

Repeatable processes for common PM tasks.

Strategic Clarity

Define charter, value proposition, and improvement hypotheses

Planning Cycle

Review → Prioritize → Commit → Communicate for quarterly and sprint planning

Codebase Discovery

Map capabilities, find ownership, understand contracts

Architecture Analysis

Understand systems, propose changes, evaluate tradeoffs

Document Production

Create PRDs, briefs, stakeholder docs with iteration

Tool Building

Create scripts, dashboards, HTML artifacts

Meeting Preparation

Structure agendas, anticipate questions, draft talking points

Data Investigation

Question → SQL query → analyze results → connect to strategy

Doc Restructuring

Diagnose confusion, separate concerns, add structure, sync references

Core Principles

What makes this framework work.

Human Judgment is Final

AI explores and proposes; humans decide and prioritize. Never delegate decisions that require context or values.

Structured Knowledge Beats Chat

Persistent documents compound value. Every conversation should produce or update an artifact.

Context is Everything

Well-organized workspace context makes AI 10x more useful. Invest in structure upfront.

Explicit Agent Modes

Telling AI what role to play produces better output than generic requests.

Git as AI Memory

Your repo is persistent; chat disappears. Commit after each AI session. Diffs show what changed; branches let you experiment.

Review the Diff

Before sharing AI-assisted work, look at what actually changed. The diff is your ownership tool.

Get Started

Install the plugin or fork the full framework.

Option 1: Install the Plugin (30 seconds)

        # One command — installs skills, commands, and hooks

        $ npx pm-ai-partner@latest

        # Choose your runtime: Claude Code, Cursor, or Codex

        # Choose scope: global (all projects) or local (current project)

Option 2: Fork the Framework (30 minutes)

For the full methodology with workspace architecture, templates, and case studies.

        # Fork and clone

        $ git clone git@github.com:YOUR-USERNAME/PM-AI-Partner-Framework.git

        # Customize CLAUDE.md, set up workspace, configure MCP tools

        # See getting-started.md for the full guide

Supported Platforms

Claude Code

Skills install as a Claude Code plugin with manifest, commands, and hooks.

Cursor

Skills install to ~/.cursor/skills/ with slash command support.

Codex

Skills install to ~/.codex/skills/ for OpenAI's CLI agent.

MCP Tools

Extend with BigQuery, Google Drive, Code Search, and more.

The Problem

What Goes Wrong

Traditional Approaches

Four Sources of Truth

Code

Docs

Data

Judgment

The Plugin

Workflow Commands

Automation Hooks

12 Agent Skills

Thought Partner

Technical Analyst

Writer

Devil's Advocate

Builder

Data Analyst

Product Brief

Meeting Prep

Stakeholder Update

Strategic Clarity

Hypothesis Tester

Competitor Analyst

Workflow Patterns

Strategic Clarity

Planning Cycle

Codebase Discovery

Architecture Analysis

Document Production

Tool Building

Meeting Preparation

Data Investigation

Doc Restructuring

Core Principles

Human Judgment is Final

Structured Knowledge Beats Chat

Context is Everything

Explicit Agent Modes

Git as AI Memory

Review the Diff

Get Started

Option 1: Install the Plugin (30 seconds)

Option 2: Fork the Framework (30 minutes)

Supported Platforms

Claude Code

Cursor

Codex

MCP Tools

Resources

Presentations

🎤 Framework Overview

🔍 OpenClaw Deep Dive

Documentation

📦 npm Package

🔌 Plugin Details

🚀 Getting Started

📖 Full Framework

⚡ Quick Reference

🔄 Workflow Guides

🎯 Cursor Skills

📁 GitHub Repo