Overview
The official Claude Code plugin directory maps skills, commands, agents, hooks, MCP services, and staged workflows into a governed extension ecosystem. This series traces how catalog metadata becomes runtime authority, crosses project or service boundaries, and is checked through verification and governance. It helps users, builders, and maintainers match capability to need, narrow permissions, inspect evidence, and keep uncertainty visible: curation improves discovery and control but is not a universal safety guarantee.
Learning outcomes
- An official listing improves discovery and adds controls, but it is not a universal safety guarantee.
- A plugin’s real behavior depends on its package, trigger, timing, authority, data boundary, and external services.
- Skills, commands, agents, hooks, and MCP integrations create different control and trust models, even in one package.
- Verification and governance support bounded claims; users still need scoped inspection, trials, and ongoing monitoring.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
What does the official Claude Code plugin directory guarantee?
It guarantees neither universal payload safety nor trustworthy remote services. It provides discovery, source information, conventions, and layered checks, while users retain the final trust decision.
How should I evaluate a Claude Code plugin before installing it?
Evaluate its purpose, source, components, triggers, authority, data flow, credentials, failure behavior, and update path. Increase inspection as access, irreversibility, or external dependence grows, then use a controlled trial.
What is the difference between skills, commands, agents, hooks, and MCP?
Skills add contextual guidance, commands expose explicit workflows, agents perform delegated roles, hooks react to events, and MCP exposes tools backed by local or remote services. Their triggers, timing, authority, and data boundaries differ.
Why do verification and governance not prove a plugin is safe?
They support narrower claims about structure, policy, reachability, review, or compatibility under stated conditions. They cannot establish every vulnerability is found, every endpoint is trustworthy, or every future update is suitable.