PM AI Partner

12 AI skills, 6 workflow commands, 3 automation hooks, and 17 MCP servers built for Product Managers.

npx pm-ai-partner@latest

Works with Claude Code, Cursor, and Codex. Mac, Windows, and Linux.

"Finally — AI skills that understand PM work, not just engineering."

Why This Exists

AI coding assistants are built for engineers. PMs get generic chatbot responses — "have you considered your stakeholders?" — instead of tools that match how we actually work.

PM AI Partner gives Claude the context to be useful for PM tasks:

• Thinking through strategy — not generating buzzword soup
• Translating code to product insights — not dumping raw implementations
• Writing briefs that engineering can scope from — not vague wishlists
• Challenging your ideas constructively — not agreeing with everything

What You Get

12 Agent Skills

Tell Claude which "hat" to wear. Skills auto-activate based on context or invoke them explicitly.

| Skill | Command | What It Does | |-------|---------|-------------| | Thought Partner | /pm:thought-partner | Brainstorm, explore options, surface assumptions | | Technical Analyst | /pm:technical-analyst | Translate code and systems into PM-friendly insights | | Writer | /pm:writer | Draft briefs, updates, emails — concise and structured | | Devil's Advocate | /pm:devil-advocate | Stress-test proposals, anticipate objections | | Builder | /pm:builder | Build tools, scripts, dashboards, prototypes | | Data Analyst | /pm:data-analyst | Write queries, analyze metrics, extract insights | | Product Brief | /pm:product-brief | Create structured PRDs and feature specs | | Meeting Prep | /pm:meeting-prep | Prepare talking points, anticipate questions | | Stakeholder Update | /pm:stakeholder-update | Status reports and executive summaries | | Strategic Clarity | /pm:strategic-clarity | Define team identity, boundaries, and charter | | Hypothesis Tester | /pm:hypothesis-tester | Design experiments, evaluate A/B tests, interpret results | | Competitor Analyst | /pm:competitor-analyst | Competitive analysis, positioning maps, feature gap assessment |

6 Workflow Commands

Multi-step workflows for recurring PM tasks.

| Command | What It Does | |---------|-------------| | /pm:plan-week | Weekly planning — pulls context from GitHub, Linear, Slack via MCP | | /pm:write-brief [feature] | Guided product brief creation with quality checklist | | /pm:prep-meeting [topic] | Meeting prep with talking points, Q&A, and role-play | | /pm:audit-codebase [system] | Codebase exploration with auto-detection, Mermaid diagrams, capability mapping | | /pm:competitor-scan [competitor] | Competitive analysis — positioning maps, feature gaps, response framework | | /pm:metrics-review | Structured metrics review — what moved, why, and what to do |

3 Automation Hooks

Hooks fire automatically — no manual invocation needed.

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

17 MCP Servers

Connect your AI assistant to external tools. The installer offers a curated catalog of PM-relevant MCP servers — pick what you need and the config is generated automatically.

Zero-config (work immediately):

| Server | What It Does | |--------|-------------| | Memory | Persistent knowledge graph — track decisions across sessions | | Fetch | Pull web pages as markdown into chat | | Sequential Thinking | Enhanced step-by-step reasoning for complex problems | | Filesystem | Read/write local files with directory scoping | | Puppeteer | Browser automation — screenshots, page navigation, form filling | | Git | Git log, diff, status, branch operations (requires Python/uvx) |

PM Essentials (one credential each):

| Server | What It Does | Credential | |--------|-------------|------------| | GitHub | Issues, PRs, code search | GITHUB_PERSONAL_ACCESS_TOKEN | | Linear | Issues, projects, sprints (remote — OAuth) | Authenticates via browser | | Atlassian | Jira issues + Confluence pages (remote — OAuth) | Authenticates via browser | | Google Drive | Team docs, sheets, slides | GOOGLE_APPLICATION_CREDENTIALS | | Slack | Channels, threads, search | SLACK_BOT_TOKEN | | Brave Search | Web search for research | BRAVE_API_KEY | | Notion | Pages, databases, wikis | OPENAPI_MCP_HEADERS | | Sentry | Error reports, stack traces, issue trends | SENTRY_AUTH_TOKEN |

Data & Analytics (credentials + setup):

| Server | What It Does | Credential | |--------|-------------|------------| | PostgreSQL | SQL queries against Postgres | Connection string | | SQLite | Query local SQLite databases | File path | | BigQuery | Google BigQuery analytics | GCP credentials JSON |

Servers you skip are added as disabled with placeholder values — enable them later by adding your credentials to the config file. Remote servers (Linear, Atlassian) authenticate via browser on first use.

Install

One Command

npx pm-ai-partner@latest

The installer prompts you to choose:

1. Runtime — Claude Code, Cursor, Codex, or all
2. Scope — Global (all projects) or local (current project only)
3. MCP Servers — optionally browse and install from a catalog of 17 servers

Non-Interactive

# Claude Code — global
npx pm-ai-partner --claude --global

# Cursor — global
npx pm-ai-partner --cursor --global

# All runtimes — global
npx pm-ai-partner --all --global

# Current project only
npx pm-ai-partner --claude --local

# Include MCP servers (all zero-config servers)
npx pm-ai-partner --claude --global --mcp

# Specific MCP servers by ID
npx pm-ai-partner --claude --global --mcp=memory,fetch,github

# MCP only (skip skills, just configure servers)
npx pm-ai-partner --mcp-only --claude --global

# List currently configured MCP servers
npx pm-ai-partner --mcp-list --claude --global

Claude Code Plugin (Alternative)

# If the plugin is published to a marketplace:
/plugin install pm-ai-partner

Manual Install

git clone https://github.com/ahmedkhaledmohamed/PM-AI-Partner-Framework.git
cd PM-AI-Partner-Framework/plugin
node bin/install.js

Verify

After installing, restart your editor and run:

/pm:thought-partner What should I focus on this quarter?

How It Works

Skills = Agent Modes

Each skill gives Claude a different "personality" optimized for a PM task. Instead of generic AI responses, you get structured, PM-aware output:

Without PM AI Partner:

"You should consider stakeholder alignment and create a roadmap..."

With /pm:devil-advocate:

What's strong: Clear user value, low engineering lift.

Key challenges:

1. Quality consistency — what's your plan when AI descriptions are wrong?
2. Success metric is fuzzy — how will you know this worked in 3 months?

Likely objections:

• Engineering: "Why not improve search instead?"
• Legal: "Who owns AI-generated text?"

Suggested mitigations: [table with specific actions]

Hooks = Automated Quality

Hooks run in the background. When you write a document to product-catalog/, the quality gate automatically checks that it meets the bar:

• Accurate: Claims backed by evidence
• Clear: Readable without context
• Owned: Has author and date
• Evidenced: Links to supporting data
• Audience-appropriate: Tone matches the reader

Commands = PM Workflows

Commands are multi-step workflows. /pm:plan-week doesn't just ask "what are your priorities?" — it:

1. Reviews your existing context (CLAUDE.md, sandbox/, product-catalog/)
2. Assesses current state and in-flight work
3. Identifies top priorities based on team context
4. Maps priorities to concrete actions
5. Flags risks for the week
6. Sets measurable success criteria

The Methodology Behind It

PM AI Partner is built on a simple idea:

PM work is reconciling four sources of truth. AI helps you explore each faster. Judgment stays human.

| Source | Question | AI Role | |--------|----------|---------| | Code | What CAN happen? | Explore, map | | Docs | What's EXPECTED? | Synthesize | | Data | What IS happening? | Analyze | | Judgment | What SHOULD happen? | Challenge |

The skills map to this framework — Technical Analyst explores code, Data Analyst queries data, Writer synthesizes docs, Devil's Advocate challenges judgment.

For the full methodology, workspace architecture, and workflow patterns, see the PM AI Partner Framework.

Case Study: OpenClaw Deep Dive

A worked example showing the framework in action — a PM-style deep dive into OpenClaw completed in ~3 hours:

1. Technical Analyst — Mapped architecture (Gateway, Brain, Hands, Memory, Heartbeat)
2. Data Analyst — Analyzed growth (0 → 220K stars in 84 days)
3. Devil's Advocate — Surfaced security risks (6 CVEs, ~900 malicious skills, $16M crypto scam)
4. Builder — Generated HTML presentation and PPTX deck

Read the full case study →

Configuration

Workspace Setup

For the best experience, create a CLAUDE.md in your project root with your team context:

# My Workspace
**Team:** [Your Team]
**Mission:** [One sentence]
**Systems:** [What you own]
**Current Focus:** [This quarter's priorities]

The more context you provide, the better the skills perform. Templates are available in the full framework.

Workspace Structure

The framework works best with this layout:

your-workspace/
├── CLAUDE.md           # Team context for AI
├── sandbox/            # Drafts & exploration (private)
└── product-catalog/    # Polished artifacts (shareable)

The quality gate hook activates for product-catalog/ files. The draft status hook activates for sandbox/ files.

Updating

npx pm-ai-partner@latest

Re-running the installer updates all skills, commands, and hooks. MCP server configs are never overwritten — existing servers are preserved.

Contributing

Improvements welcome — especially:

• New PM-specific skills or commands
• Better examples from real PM work
• Bug reports and feature requests
• Hooks for additional PM workflows

See CONTRIBUTING.md.

Full Framework

This plugin is part of the PM AI Partner Framework — a complete methodology for using AI as a structured PM thinking partner. The framework includes:

• 9 detailed workflow patterns (Strategic Clarity, Planning Cycle, Codebase Deep-Dive, Tool Building, Communication Prep, PM-Eng Bridge, Data Investigation, Doc Restructuring, Analysis Pipeline)
• Workspace architecture (sandbox → product-catalog graduation)
• CLAUDE.md templates for team context
• 17 MCP server configs (Memory, GitHub, Linear, Atlassian, Slack, BigQuery, and more)
• Case studies with worked examples
• Presentations and demo materials

Explore the full framework →

Author

Ahmed Khaled Mohamed

• GitHub
• LinkedIn
• Email

License

MIT License. See LICENSE for details.