Structure of a PM Skills Plugin: A Complete Guide to the Claude Marketplace Architecture

A PM Skills plugin follows a standardized three-tier directory structure containing a .claude-plugin/plugin.json metadata file, a skills/ directory with individual SKILL.md files, and a commands/ directory housing slash-command definitions.

The PM Skills repository (phuryn/pm-skills) operates as an AI-powered marketplace of Claude plugins designed for product management workflows. Understanding the structure of a PM Skills plugin is essential for contributors and users who want to extend the marketplace or leverage its capabilities for strategic planning, discovery, and execution tasks.

Directory Layout and Core Components

Each plugin in the marketplace resides in its own top-level folder following the pm-<domain>/ naming convention (e.g., pm-go-to-market, pm-product-discovery). The repository organizes nine distinct plugins this way, and all share an identical internal architecture.

pm-<domain>/
├─ .claude-plugin/
│   └─ plugin.json           ← plugin metadata and discovery info
├─ skills/
│   └─ <skill-name>/
│       └─ SKILL.md          ← reusable prompt template
└─ commands/
    └─ <command-name>.md     ← slash-command workflow definition

Plugin Metadata Configuration

The .claude-plugin/plugin.json file declares the plugin's identity, version, and author credentials. Claude reads this file during the discovery phase to register the plugin's capabilities.

In pm-go-to-market/.claude-plugin/plugin.json, the metadata specifies the plugin name, version, and descriptive tags that help Claude match user queries to available skills. Similarly, pm-ai-shipping/.claude-plugin/plugin.json (version 2.0.0) demonstrates how semantic versioning tracks feature updates across the marketplace.

Skill Definitions

Skills are self-contained markdown files stored in skills/<skill-name>/SKILL.md. These files encode proven product management frameworks—such as SWOT analysis, pricing strategies, or PRD templates—as structured prompts that Claude can execute.

For example, pm-product-strategy/skills/pricing-strategy/SKILL.md contains the prompt template for generating tiered pricing models, while pm-execution/skills/create-prd/SKILL.md defines the narrative structure and required sections for product requirement documents. Each skill operates independently, allowing multiple commands to reference the same underlying capability.

Command Workflows

Commands represent multistep workflows that chain together individual skills. Stored as commands/<command-name>.md, these files define slash commands (e.g., /discover, /plan-launch) and specify the sequence of skills to execute.

The file pm-product-discovery/commands/discover.md illustrates this pattern by listing four sequential skills: brainstorm-ideas-new, identify-assumptions-new, prioritize-assumptions, and brainstorm-experiments-new. When a user invokes /discover, Claude parses this command file and orchestrates the skill chain automatically.

How the Plugin Architecture Works

The PM Skills marketplace follows a four-stage execution flow that transforms user input into structured product management deliverables.

1. Installation – Users add the marketplace via Claude Cowork or the CLI, which pulls all nine plugins and their metadata into the local environment.

2. Skill Loading – When Claude detects a query matching a skill's semantic signature (e.g., "pricing strategy" matching pm-product-strategy/skills/pricing-strategy/SKILL.md), it loads the corresponding markdown file without requiring explicit command invocation.

3. Command Execution – Slash commands trigger sequential skill execution. For instance, /plan-launch invokes the gtm-strategy skill, beachhead-segment, ideal-customer-profile, growth-loops, and competitive-battlecard in sequence, aggregating outputs into a cohesive go-to-market plan.

4. Result Delivery – Claude returns assembled outputs (tables, strategic frameworks, or draft documents) directly in the chat interface. Users can persist results by requesting file exports or copying structured markdown.

Practical Examples of Plugin Usage

Running a Discovery Workflow

Invoke the discovery command with a product concept:

/discover AI-powered meeting summarizer for remote teams

Claude executes the skill chain defined in pm-product-discovery/commands/discover.md, returning a package containing idea variations, risky assumptions, prioritized validation experiments, and lean test proposals.

Installing a Single Plugin via CLI

Install only the Go-to-Market plugin to keep the environment lightweight:

claude plugin install pm-go-to-market@pm-skills

This command pulls only the specified plugin's plugin.json, skills, and commands, rather than the entire marketplace.

Using Individual Skills

Query a specific skill directly without invoking a full command workflow:

pricing-strategy — Build a tiered pricing model for a SaaS product with competitive analysis

Claude loads pm-product-strategy/skills/pricing-strategy/SKILL.md and returns a markdown table with tier names, price points, feature bundles, and competitive positioning rationale.

Data Analytics Commands

For SQL generation tasks, the command file pm-data-analytics/commands/write-query.md wraps the sql-queries skill, allowing users to request database queries through natural language descriptions.

Summary

• Standardized structure: Every PM Skills plugin uses pm-<domain>/.claude-plugin/plugin.json, skills/<name>/SKILL.md, and commands/<name>.md.
• Separation of concerns: Skills contain reusable prompt templates, while commands define workflow orchestration.
• File-based configuration: No compilation required—plugins are defined through markdown and JSON files that Claude interprets at runtime.
• Composable architecture: Multiple commands can reference the same skill, maximizing reuse across the nine-plugin marketplace.

Frequently Asked Questions

What is the difference between a skill and a command in PM Skills?

A skill is a single reusable prompt template stored in skills/<skill-name>/SKILL.md that performs one specific product management task (e.g., generating a competitive battlecard). A command is a workflow definition stored in commands/<command-name>.md that chains multiple skills together to accomplish complex, multi-step objectives (e.g., /discover running four distinct discovery skills sequentially).

How do I install a specific PM Skills plugin instead of the entire marketplace?

Use the Claude CLI with the specific plugin identifier: claude plugin install pm-go-to-market@pm-skills. This pulls only the targeted plugin's metadata, skills, and commands rather than all nine plugins in the phuryn/pm-skills repository, keeping your local environment minimal and focused.

What file format does Claude expect for skill definitions?

Claude expects skill definitions to be Markdown files named exactly SKILL.md and located within a subdirectory under the skills/ folder (e.g., skills/pricing-strategy/SKILL.md). These files must contain the prompt template, usage instructions, and optional output format specifications that guide Claude's response generation.

Can I create my own PM Skills plugin for a custom domain?

Yes. Create a new top-level directory following the pm-<domain> naming convention, then populate it with the standard three-component structure: .claude-plugin/plugin.json for metadata, skills/ subdirectories containing SKILL.md files, and commands/ containing markdown workflow definitions. Refer to pm-toolkit/.claude-plugin/plugin.json and pm-ai-shipping/.claude-plugin/plugin.json as canonical examples of the required JSON schema.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}