How Skills and Commands Work Together in the PM Skills Architecture

Skills are modular building blocks that encode domain knowledge, while commands are user-triggered workflows that orchestrate multiple skills into complete product management tasks.

The phuryn/pm-skills repository implements a modular architecture that cleanly separates domain intelligence from procedural workflows. Understanding how skills and commands work together in the PM Skills architecture enables developers to compose reusable AI-driven product management tools without duplicating logic. This separation allows the marketplace to grow organically, with new skills added independently of existing commands.

Core Architecture: Skills vs. Commands

What Are Skills?

Skills are standalone markdown files located under pm-*/skills/.../SKILL.md that encapsulate specific frameworks, analyses, or knowledge domains. Each skill functions as a reusable intelligence module that the system auto-loads when detecting relevant conversation context, or that users can force explicitly using slash-prefixed syntax (e.g., /brainstorm-ideas-new or /plugin-name:skill-name).

According to the repository's README, skills encode product-management knowledge as modular building blocks that remain language-agnostic and work across Claude, Gemini, OpenCode, and other AI agents.

What Are Commands?

Commands are orchestration scripts defined in markdown files under pm-*/commands/...md. They represent end-to-end workflows triggered by leading-slash syntax (e.g., /discover, /strategy). Unlike skills, commands do not contain domain logic directly; instead, they act as scripts that chain together one or more skills to achieve multi-step processes, passing the output of each skill as input to the next.

The Interaction Model: Chaining and Data Flow

When a user invokes a command, the architecture executes its constituent skills in strict sequence. The system passes the output of one skill as the input to the next, creating a seamless pipeline that transforms raw input into structured deliverables.

For example, when running /discover AI-powered meeting summarizer for remote teams, the command triggers four skills in order: brainstorm-ideas-new → identify-assumptions-new → prioritize-assumptions → brainstorm-experiments-new. Each step receives the context produced by the previous skill, ultimately returning a complete discovery output. After execution, commands suggest the next logical workflow, creating guided user experiences while reusing the same skill definitions across many commands.

Implementation Examples from Source Code

Skill Definition: brainstorm-ideas-new

The brainstorm-ideas-new skill demonstrates the single-responsibility pattern. Located at pm-product-discovery/skills/brainstorm-ideas-new/SKILL.md, it contains focused instructions for generating product ideas without workflow logic:

# Brainstorm Ideas (New Products)

You are an expert product strategist.  
Generate 5 distinct ideas for a brand‑new SaaS product that solves a common pain point.

This skill is language-agnostic and can be consumed by any AI agent that reads markdown instructions.

Command Definition: /discover

The /discover command orchestrates multiple skills into a complete discovery cycle. Defined in pm-product-discovery/commands/discover.md, it chains four distinct skills:

# /discover

Chains four skills to run a full discovery cycle:

1. brainstorm-ideas-new
2. identify-assumptions-new
3. prioritize-assumptions
4. brainstorm-experiments-new

When invoked, this command file instructs the system to execute each skill sequentially, creating a multi-step process while the individual skills remain reusable by other commands.

Direct Skill Invocation

Users can bypass command orchestration and load skills directly using /skill-name syntax. For example, /product-vision forces the product-vision skill to load immediately from pm-product-strategy/skills/product-vision/SKILL.md, useful when needing specific frameworks without executing a full workflow chain.

Key Files and Directory Structure

The repository organizes skills and commands to maintain strict separation of concerns:

• pm-product-discovery/skills/brainstorm-ideas-new/SKILL.md: Encodes the brainstorming framework as a reusable skill.
• pm-product-discovery/commands/discover.md: Defines the /discover workflow orchestration.
• pm-toolkit/skills/review-resume/SKILL.md: Demonstrates cross-domain skill reusability outside product discovery.
• pm-toolkit/commands/review-resume.md: Shows simple command-to-skill forwarding for the resume review workflow.
• README.md: Contains the "How It Works (Skills, Commands, Plugins)" section explaining the high-level architecture and interaction model.

Summary

• Skills are modular markdown files containing domain-specific knowledge and frameworks, stored under pm-*/skills/.
• Commands are user-triggered workflows defined under pm-*/commands/ that chain multiple skills into end-to-end processes.
• Data flows sequentially through skill chains, with each skill's output feeding into the next skill's input.
• Skills are language-agnostic and work across AI platforms, while commands use Claude-specific slash syntax.
• The architecture enables organic growth: developers add new skills without changing existing commands, and compose new commands from any combination of existing skills.

Frequently Asked Questions

How do skills load automatically in the PM Skills architecture?

Skills auto-load when the model detects that their domain knowledge is relevant to the current conversation context. Users can also force immediate loading by typing a slash-prefixed skill name (e.g., /brainstorm-ideas-new), which bypasses the auto-detection logic and immediately injects the skill's framework into the conversation.

Can PM Skills commands be used with AI models other than Claude?

Commands are Claude-specific implementations that rely on slash-command syntax. While other AI agents can read the command markdown files, they must translate the workflow definitions into their own native orchestration language. However, the underlying skills are language-agnostic markdown files that work with Claude, Gemini, OpenCode, and other AI systems without modification.

What is the difference between /discover and /brainstorm-ideas-new syntax?

/discover invokes a command that orchestrates multiple skills into a complete workflow, whereas /brainstorm-ideas-new invokes a single skill directly. Use command syntax when you need multi-step processes with sequential execution, and skill syntax when you need a specific framework or analysis tool without the surrounding workflow orchestration.

How do I create a new command that uses existing skills?

Create a new markdown file under the appropriate pm-*/commands/ directory (e.g., pm-product-strategy/commands/my-command.md). Define the command name with a leading slash in the H1 heading, then list the skills to invoke in sequence. The system automatically chains these skills, passing context between them according to the standard output-to-input flow defined in the PM Skills architecture.

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