PM-Skills Compatibility with AI Assistants Beyond Claude: Implementation Guide

PM-Skills works with Gemini, OpenCode, Cursor, and Kiro because its product management frameworks are stored as plain markdown files that any AI assistant can interpret, while only the automation layer is Claude-specific.

The phuryn/pm-skills repository provides a marketplace of product management frameworks designed for AI augmentation. While originally built with Claude integration in mind, the architecture deliberately separates knowledge storage from execution logic, enabling seamless PM-skills compatibility with AI assistants across different platforms.

Understanding the Architecture: Skills vs. Commands

The repository organizes functionality into two distinct layers that determine cross-platform compatibility.

Skills represent the core knowledge base. These are plain-text markdown files stored under each plugin's skills/ directory (e.g., pm-product-discovery/skills/brainstorm-ideas-new/SKILL.md). Because they contain universal framework descriptions without platform-specific syntax, any AI assistant capable of reading markdown can execute them.

Commands provide workflow automation. These are Claude-specific slash-command definitions located in commands/ folders (e.g., pm-product-discovery/commands/discover.md). This layer orchestrates multiple skills into end-to-end workflows, but only Claude interprets these definitions natively.

Setting Up PM-Skills for Other AI Assistants

To use the framework with non-Claude assistants, copy the skill files into your target platform's expected directory structure.

Gemini CLI Configuration

Gemini CLI reads skills from the global ~/.gemini/skills/ directory or project-level .gemini/skills/ folders.

# Copy all skills from repository root to Gemini's global skills folder

for plugin in pm-*/; do
  cp -r "$plugin/skills/"* ~/.gemini/skills/ 2>/dev/null
done

OpenCode Setup

OpenCode expects skills in a project-level .opencode/skills/ directory.

# From the repository root

for plugin in pm-*/; do
  mkdir -p .opencode/skills/
  cp -r "$plugin/skills/"* .opencode/skills/ 2>/dev/null
done

(Adapted from the README's "Other AI assistants (skills only)" section【/cache/repos/github.com/phuryn/pm-skills/main/README.md†L123-L130】.)

Cursor and Kiro Integration

Both assistants follow similar patterns:

• Cursor: Copy skills to .cursor/skills/
• Kiro: Copy skills to .kiro/skills/

The same shell loop pattern applies, substituting the target directory for your specific assistant.

Invoking Skills on Different Platforms

Once copied, invoke skills using your assistant's native syntax. The skill file contains the complete prompt describing the framework, so the assistant returns structured guidance immediately.

Using OpenCode:

opencode skill product-vision \
  "Create a compelling vision for an AI-driven project-management SaaS."

Using Gemini CLI:

gemini skill brainstorm-ideas-new \
  "Generate 5 high-impact ideas for a new B2B analytics platform."

Recreating End-to-End Workflows Without Claude

While Skills transfer directly, Commands require manual reimplementation on other platforms. You can recreate Claude's automated workflows by chaining skill calls in your preferred scripting language.

The following Python pseudo-code illustrates how to replicate the discover command workflow using generic skill invocations:

def discover(idea):
    ideas = opencode.skill("brainstorm-ideas-new", idea)
    assumptions = opencode.skill("identify-assumptions-new", ideas)
    prioritized = opencode.skill("prioritize-assumptions", assumptions)
    experiments = opencode.skill("brainstorm-experiments-new", prioritized)
    return {
        "ideas": ideas,
        "assumptions": assumptions,
        "prioritized": prioritized,
        "experiments": experiments,
    }

# Example execution

result = discover("AI-powered meeting summarizer")

This approach preserves the core logic from files like pm-product-discovery/commands/discover.md while adapting the execution mechanism to your platform's specific API or CLI interface.

Summary

• Skills are assistant-agnostic: Plain markdown files in pm-*/skills/ directories work on any platform that reads markdown.
• Commands are Claude-specific: Workflow automation requires reimplementation for Gemini, OpenCode, or other assistants.
• Setup requires only file copying: Move skill directories to platform-specific locations (e.g., ~/.gemini/skills/, .opencode/skills/).
• Workflows can be recreated: Chain individual skill calls using your target assistant's programmable interface to replace Claude's native command orchestration.

Frequently Asked Questions

Can I use PM-Skills with OpenCode without modifying the files?

Yes. Because skills are stored as standard markdown files (e.g., pm-product-strategy/skills/product-vision/SKILL.md), you can copy them directly into .opencode/skills/ without modification. The OpenCode CLI reads the content and executes the framework as written.

What is the difference between skills and commands in the PM-Skills repository?

Skills are markdown documentation containing PM frameworks and prompts, stored in skills/ subdirectories. Commands are Claude-specific workflow definitions stored in commands/ folders that automate multi-step processes. Skills work universally; commands only work with Claude's slash-command feature.

Do I need to install the entire repository to use PM-Skills with Gemini?

No. You only need the skills/ directories from the relevant plugin folders (e.g., pm-product-discovery/skills/). The repository README provides a shell script that iterates through all plugins and copies only the skill files to your Gemini skills directory, ignoring the Claude-specific command files.

How do I recreate an automated workflow like /discover on a non-Claude platform?

Extract the workflow logic from the command file (e.g., pm-product-discovery/commands/discover.md) and reimplement it using your target assistant's scripting capabilities. The workflow typically chains multiple skills—such as brainstorming, assumption identification, and experiment design—into a sequential process that you can script in Python, Bash, or your assistant's native automation language.

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