Plugin = Skills + Agents + Hooks in a Box

What's Inside a Plugin

• **plugin.json** — The manifest declaring name, version, and description
  • › The entry point that Claude Code and Copilot CLI both read
  • › Defines what the plugin contains and how to discover it
• **skills/** — Slash commands that automate workflows
  • › Each skill is a directory with a SKILL.md file
  • › Triggered by typing /skill-name in the terminal
• **agents/** — Specialized AI personas with model tiers
  • › Each agent has its own system prompt and tool access
  • › Can run in isolation with worktree support
• **hooks/** — Event-driven guardrails and automation
  • › Fire on specific events (pre-commit, post-save, etc.)
  • › Prevent mistakes before they happen
• **Supporting files** — MCP configs + rules + templates
  • › .mcp.json registers external tool connections
  • › rules/ loads conventions based on file paths

Why Plugins Beat Loose Files

1. 1 **Install once, use everywhere** — Add to any project with one command. Versioned, shared across teams, and updateable by bumping a single version number.
2. 2 **Encapsulation** — MCP servers, hooks, and rules are bundled together. No scattered config files across your project — everything lives in one installable package.
3. 3 **Shared distribution** — The whole team uses the same skills without copy-pasting. Roll back if a new version breaks something. Everyone stays in sync.
4. 4 **Our setup** — Four plugins (dx-core, dx-hub, dx-aem, dx-automation), distributed to 4 consumer repos from a single source.
5. 5 **Start exploring** — Open an installed plugin's plugin.json to see the manifest, then browse skills/ and agents/ to understand what's available.