Agent plugins (marketplaces)
Installable packages that add commands, skills, subagents, hooks and MCP servers to a coding agent, shared through public or company marketplaces.
A plugin bundles what a team has learned into one install: slash commands, skills, hooks and MCP servers. Claude Code plugins are installed from marketplaces, which are just git repositories or URLs with a catalog; the PRP workflow, for example, is distributed this way. Companies run an internal marketplace to give every developer the same setup.
A plugin runs with the agent’s permissions, and hooks run as code on the developer’s machine, so installing one is like adding a dependency with shell access. Review it, pin the version and prefer a company marketplace for anything that touches secrets. The risks are the usual supply chain attack plus tool poisoning through instructions hidden in skills or tool descriptions.
Sources
- Claude Code Docs: Plugins overview code.claude.com
Related terms
- Slash commands (custom commands): Saved prompts stored as files in a repository and run in a coding agent by typing a short command such as /review or /plan.
- Agent Skills: Folders with instructions, scripts and references that an agent loads only when the task needs them.
- Hooks: Scripts the harness runs automatically at fixed points, such as before a tool call or after an edit.
- MCP (Model Context Protocol): An open protocol for plugging tools and data into AI apps: write a server once, use it from any client.
- Software supply chain attack: Compromising software through what it is built from: a package, a build tool, a CI action or a maintainer’s account.
- Tool poisoning (MCP): An attack that hides instructions in a tool’s name, description or output, so an agent using that tool follows the attacker instead of the user.
