Skip to content

Platforms

PM-Skills works across the AI agent ecosystem. This page covers every supported platform with step-by-step setup instructions. For the README’s quick-start summary, see README.md.

Using the pm-skills sub-agents on a non-Claude client? 5 of the 6 sub-agents ship with dispatch skills (skills/utility-pm-{role}/) for cross-client access; the sixth, pm-skill-router, is an internal tooling instrument with no dispatch skill and is not invoked by users (see the compatibility matrix). Native @-mention registration is live on Claude Code; the dispatch mechanism is validated on Codex CLI; Cursor / Windsurf / Copilot CLI / Gemini CLI remain experimental. See the canonical Sub-Agent Compatibility Matrix for per-client status before invoking pm-critic, pm-skill-auditor, pm-changelog-curator, pm-release-conductor, or pm-workflow-orchestrator on a non-Claude platform.

PlatformNative?Recommended path
Claude CodeYesPlugin marketplace
Claude.ai / Claude DesktopYesZIP upload
CursorYesGit clone (AGENTS.md auto-discovery)
WindsurfYesGit clone (AGENTS.md auto-discovery)
Gemini CLIYesGit clone (AGENTS.md auto-discovery)
GitHub CopilotYesGit clone (AGENTS.md auto-discovery)
VS Code (Cline / Continue)YesGit clone + extension
OpenCodeYesGit clone + skill source config
Any MCP clientUniversalpm-skills-mcp (maintenance mode)
ChatGPT / other LLMsManualCopy skill content into prompt

Two install paths, depending on your Claude Code version.

Section titled “Modern Claude Code: plugin marketplace (recommended)”
/plugin marketplace add product-on-purpose/agent-plugins
/plugin install pm-skills@product-on-purpose

Already installed via the old pm-skills-marketplace? It keeps working - no action needed. To move to the new home, see the v2.21.0 release notes.

After install, all 68 skills resolve from any directory. Slash commands like /pm-skills:deliver-prd, /pm-skills:define-opportunity-tree, /pm-skills:foundation-okr-writer, /agenda work immediately. Verify with /plugin list.

To track a local working copy instead of the published version (for forkers and contributors):

/plugin marketplace add /path/to/your/local/pm-skills
/plugin install pm-skills@pm-skills-marketplace
Terminal window
npx skills add product-on-purpose/pm-skills

Installs all 68 skills into Claude Code’s default skills directory. No clone, no sync. Works identically across most agents (see Cursor, Copilot, Cline).

For Claude clients that do not support the marketplace flow:

  1. Clone the repo or extract the release ZIP.
  2. During plugin setup, select the manifest at .claude-plugin/plugin.json.
  3. Complete install in your client; reload if prompted.
.claude-plugin/plugin.json
git clone https://github.com/product-on-purpose/pm-skills.git
cd pm-skills

Test with a real skill:

/pm-skills:deliver-prd "Search feature for our e-commerce platform"

You should see a complete PRD with problem statement, success metrics, user stories, scope, and technical considerations.


Both use ZIP upload to Project Files.

  1. Download the latest release ZIP from Releases:
    • File: pm-skills-v2.22.0.zip (or the latest version shown there)
  2. Upload to your Claude environment:
    • Claude.ai: Project Settings > Add Files > Upload ZIP
    • Claude Desktop: Settings > Capabilities > Upload ZIP
  3. Use skills by name in your conversation: “Use the prd skill to create requirements for [your topic]”

The ZIP contains all 68 skills, slash commands, workflows, sub-agent definitions, activation hooks, library samples, and a QUICKSTART.md with detailed instructions.


Cursor auto-discovers PM-Skills via AGENTS.md once the repo is in your workspace.

Terminal window
# Clone into your project workspace
git clone https://github.com/product-on-purpose/pm-skills.git
# Or add as a git submodule (recommended for team use)
git submodule add https://github.com/product-on-purpose/pm-skills.git

Open Cursor with the workspace that includes pm-skills. Cursor’s AI assistant will discover all 68 skills via AGENTS.md. Invoke by name:

“Use the hypothesis skill to test my assumption about checkout abandonment.” “Use the opportunity-tree skill for our Q3 retention initiative.”

For programmatic access, you can also configure pm-skills-mcp; see Any MCP client.


Same pattern as Cursor. Windsurf auto-discovers via AGENTS.md.

Terminal window
git clone https://github.com/product-on-purpose/pm-skills.git

Open the folder in Windsurf. The AI assistant discovers and can invoke all 68 skills. No sync helper needed; Windsurf reads AGENTS.md directly.


Same pattern as Cursor and Windsurf. Gemini CLI auto-discovers PM-Skills via AGENTS.md once the repo is in your workspace.

Terminal window
git clone https://github.com/product-on-purpose/pm-skills.git

Open Gemini CLI in the folder that includes pm-skills. Gemini discovers and can invoke all 68 skills via AGENTS.md. Invoke by name:

“Use the hypothesis skill to test my assumption about checkout abandonment.”

For programmatic access, you can also configure pm-skills-mcp; see Any MCP client.


Copilot Chat auto-discovers PM-Skills via AGENTS.md when the repo is in your workspace.

Terminal window
# Clone into your project
git clone https://github.com/product-on-purpose/pm-skills.git
# Or add as a submodule
git submodule add https://github.com/product-on-purpose/pm-skills.git
  1. Open a workspace that includes pm-skills (repo root or submodule).
  2. Use Copilot Chat in agent or workspace context.
  3. Invoke skills by name in chat: Use the hypothesis skill for checkout abandonment.

Copilot will read the skill, follow the template, and produce structured output.


PM-Skills works with the Cline and Continue extensions.

  1. Install the extension in VS Code.

  2. Clone PM-Skills into your workspace:

    Terminal window
    git clone https://github.com/product-on-purpose/pm-skills.git
  3. The extension will discover skills via AGENTS.md.

  4. In the extension chat: “Use the hypothesis skill to test my assumption about [your topic].”

Manual approach (any VS Code AI extension)

Section titled “Manual approach (any VS Code AI extension)”
  1. Open any SKILL.md file from skills/{phase-skill}/ (for example skills/deliver-prd/SKILL.md).
  2. Copy the content into your AI chat.
  3. Ask: “Follow these instructions to create a PRD for [your topic].”

OpenCode loads PM-Skills directly from the repository.

Terminal window
git clone https://github.com/product-on-purpose/pm-skills.git
cd pm-skills
  1. Configure OpenCode to use this folder as a skills source.

  2. If your OpenCode flow expects .claude/skills/, run the sync helper once:

    Terminal window
    ./scripts/sync-claude.sh # macOS / Linux
    .\scripts\sync-claude.ps1 # Windows
  3. Invoke skills by name: “Use the prd skill to create requirements for [your topic].”


For MCP-compatible clients (Claude Desktop, Cursor with MCP enabled, custom MCP setups), use the pm-skills-mcp companion server.

MCP server is in maintenance mode as of 2026-05-04. The catalog is frozen at the v2.9.2 build. New users on most platforms are better served by the file-based paths above. Use MCP only if your tooling specifically requires the protocol.

Add to your MCP client config (typically claude_desktop_config.json or your client’s equivalent):

{
"mcpServers": {
"pm-skills": {
"command": "npx",
"args": ["pm-skills-mcp"]
}
}
}

After restart, the MCP server’s embedded catalog becomes available as programmatic MCP tools. The pm-skills-mcp catalog is frozen at the v2.9.2 build’s 40-skill embedded set (plus 11 workflow tools and 8 server-utility tools, 59 tools total) per the 2026-05-04 maintenance-mode decision. The current pm-skills file-based catalog is 68 skills; the new v2.15.0 Foundation Sprint and Design Sprint families are NOT embedded in MCP, and the file-based install path is required for parity with the full current catalog. Some utility skills (for example update-pm-skills) are designed for file-system contexts and have reduced applicability via MCP transport.

See the pm-skills-mcp README for client-specific setup details.


ChatGPT, OpenAI Codex, and other LLMs without native skill support can still use PM-Skills via copy-paste.

  1. Clone or download PM-Skills:

    Terminal window
    git clone https://github.com/product-on-purpose/pm-skills.git

    Or download the latest release ZIP.

  2. Open the skill you need. For example, for a PRD: skills/deliver-prd/SKILL.md.

  3. Copy the full content into your ChatGPT (or other LLM) conversation.

  4. Ask: “Follow these instructions to create a PRD for [your topic].”

The skill content is self-instructing: it tells the LLM what to do, what shape to produce, and what quality bar to meet. The model does not need to be skill-aware.

For higher-quality output, also paste the matching references/EXAMPLE.md so the LLM has a worked example to mirror.


Slash commands don’t appear after install

Section titled “Slash commands don’t appear after install”
  • Claude Code: run /plugin list to confirm pm-skills is installed. If not, re-run the install command. If yes, run /reload-plugins or restart the session.
  • skills CLI install: confirm skills landed in your agent’s default skills directory; the CLI prints the path on completion.
  • Git clone install: ensure your agent reads AGENTS.md (Cursor, Copilot, Windsurf, Gemini CLI do automatically when the repo is in workspace).

If you see schema errors during /plugin marketplace add, you are likely on a pre-v2.13.1 cached version. Pull the latest:

Terminal window
git -C /path/to/pm-skills pull
/plugin marketplace remove pm-skills-marketplace
/plugin marketplace add product-on-purpose/agent-plugins
/plugin install pm-skills@product-on-purpose

The skill is reading the right SKILL.md, but the example may not be loading. Confirm references/EXAMPLE.md exists at skills/{phase-skill}/references/EXAMPLE.md. If your install path skipped the references folder, re-install or clone the repo for full coverage.

See pm-skills-mcp for client-specific troubleshooting. The MCP server is in maintenance mode; if you need a feature that ships in newer file-based versions but not in v2.9.2, use one of the file-based install paths above.