Claude Code Plugins vs Skills: Marketplace, JetBrains & Cross-Harness Guide

For developers and engineering leads standardizing coding agents across repositories and IDEs: the phrase "Claude Code plugin" now refers to two separate layers—IDE extensions (VS Code and JetBrains Marketplace) and in-harness workflow packages (/plugin marketplace add). Verified as of November 2026 across Claude Code CLI, IntelliJ/WebStorm, VS Code, and OpenAI Codex CLI, this guide untangles both layers, compares plugins against standalone SKILL.md folders, and shows how to share skills across Claude Code and Codex.

Disambiguating the two meanings of "Claude Code plugin"

Before choosing an installation path, separate the IDE bridge from the workflow package:

Once the JetBrains or VS Code extension connects to your local claude binary, every /plugin package and .claude/skills/ folder in your project works identically inside the IDE and the standalone terminal.

Claude Code plugins vs standalone SKILL.md folders

A skill teaches the agent how to execute a domain workflow. A plugin packages skills alongside commands, hooks, subagents, and MCP servers for multi-repo distribution.

Practical rule of thumb

1. Author inside .claude/skills/ first. Prove the SKILL.md triggers accurately and passes real pull requests in one codebase.
2. Extract into a plugin repository once a third repo copies the folder. Centralizing too early slows iteration; extracting on the third copy eliminates copy-paste drift.

Anatomy of a Claude Code plugin marketplace repository

Any public or private GitHub repository becomes a self-hosted plugin marketplace by adding .claude-plugin/marketplace.json. Here is the minimal production layout modeled on frankxai/claude-skills-library:

team-agent-marketplace/
├── .claude-plugin/
│   └── marketplace.json
└── plugins/
    └── web-release-pack/
        ├── .claude-plugin/
        │   └── plugin.json
        ├── skills/
        │   └── accessibility-check/
        │       └── SKILL.md
        ├── commands/
        │   └── preflight.md
        └── agents/
            └── release-verifier.md

Sample .claude-plugin/plugin.json:

{
  "name": "web-release-pack",
  "version": "1.2.0",
  "description": "WCAG 2.2 AA accessibility skill, preflight slash command, and independent release verifier.",
  "author": {
    "name": "Platform Engineering"
  }
}

Team installation takes two commands inside Claude Code:

/plugin marketplace add your-org/team-agent-marketplace
/plugin install web-release-pack@team-agent-marketplace

Setting up the Claude Code JetBrains plugin (IntelliJ, WebStorm, PyCharm)

Engineers on IntelliJ IDEA, WebStorm, PyCharm, GoLand, or Rider get three concrete workflow upgrades from the JetBrains integration:

1. Install the CLI first — Verify claude --version runs in your OS terminal.
2. Install the JetBrains plugin — Open Settings -> Plugins -> Marketplace, search Claude Code [Beta], install, and restart the IDE.
3. Launch from the integrated terminal — Running claude inside WebStorm or IntelliJ's built-in terminal automatically sets up the local IDE bridge socket.
4. Use selection + diagnostics handoff — Highlighting lines 40–75 in your active editor tab immediately exposes that exact code range and current IDE inspection warnings to Claude without copy-pasting file paths.
5. Keymap tip — Bind Send Selection to Claude to a dedicated shortcut so code review loops stay keyboard-only.

How "Codex skill for Claude Code" works (two-way interoperability)

Developers searching for a Codex skill for Claude Code are solving one of two problems:

1. Sharing one SKILL.md library across both Claude Code and Codex CLI

Because Anthropic and OpenAI converged on the open Agent Skills specification (SKILL.md with YAML name and description), you can maintain one canonical skills/ directory and expose it to both harnesses through directory symlinks or Windows junctions:

# macOS / Linux symlink from canonical repo into both harnesses
ln -s ~/repos/team-skills/skills ~/.claude/skills/team
ln -s ~/repos/team-skills/skills ~/.codex/skills/team

2. Calling OpenAI Codex CLI from Claude Code as an independent verifier

Multi-model verification catches blind spots that single-model self-review misses. A lightweight .claude/skills/codex-second-opinion/SKILL.md lets Claude Code dispatch a read-only audit to Codex CLI (codex exec --sandbox read-only):

---
name: codex-second-opinion
description: Run an independent read-only code review using OpenAI Codex CLI on the current git diff before merging high-impact changes.
---

# Codex second-opinion review

1. Verify `codex --version` is installed locally.
2. Run a read-only review pass against the current branch diff:
   `codex exec --sandbox read-only "Review git diff origin/main...HEAD for regressions, security flaws, and missing edge cases. Cite exact file:line numbers."`
3. Address every verified defect and re-run local unit tests before opening the pull request.

Next steps

• Copy starter CLAUDE.md, .claude/settings.json, and hook configs from the Claude Code templates & starter kit.
• Pair your skills with live data tools in Best MCP servers for Claude Code & business workflows.
• Read the rapid comparison in Q&A: Claude Code plugins vs skills or build a custom skill in the Skills Hub.