What is the difference between a Claude Skill and a Claude Plugin?

A skill is one packaged capability, scoped to one task. A plugin is a container that can hold several skills alongside hooks, agents, and tool connections, published as one signed unit through a marketplace. Put simply, a skill is an ingredient and a plugin is the dish: a team writes a skill to teach Claude one job, then a plugin packages that skill with whatever hooks and connectors the job needs so the whole capability installs and updates as one piece.

How is a Claude Skill different from a hook?

A skill teaches Claude how to do something; a hook enforces that something happens, or does not happen, regardless of what the model decides. A skill is instructions the model reads and reasons over. A hook is a shell command that fires automatically on a lifecycle event, such as before or after a tool call, and its exit code, not the model's judgment, controls whether the action proceeds. Teams pair the two: a skill for the task, a hook for the guardrail around it.

How is a Claude Plugin different from an MCP connector?

A plugin is a bundle of capabilities, skills, hooks, agents, and connections, packaged for one team to install together. An MCP connector is one of the things a plugin can declare: an application-tier service that exposes an external system to Claude through the Model Context Protocol. A plugin can ship an MCP server declaration alongside a skill and a hook, but the connector itself is a separate piece of software with its own authentication, scopes, and audit behavior. Lightbridge Automation covers that architecture in its guide to what MCP is.

Where can Claude Skills and Plugins be used?

Skills are designed to be portable across the Claude product family: the same skill can apply in claude.ai, Claude Code, the Claude Developer Platform, and Claude Cowork, so a capability written once does not need to be rewritten per surface. Plugins are the extension and marketplace mechanism most documented for Claude Code, where a team's plugin marketplace can be pushed to every developer's workstation through existing device management. Verify current availability at docs.anthropic.com before planning a rollout, since surface support moves quickly.

Who decides which skills and plugins a team can use?

In a governed deployment, the organization does, not each individual user. A team typically holds a private plugin marketplace, often a Git repository with a manifest that names every approved plugin and pins each one to a specific version. A capability reaches every authorized person by being added to that manifest; a capability is revoked by removing the entry. That central control point is what lets a security or compliance team review an AI capability before it reaches a user, rather than discovering it after the fact.

Why do Claude Skills and Plugins matter for a business, not just a developer?

Because they replace an unauditable habit, prompt engineering copied from person to person, with a reviewable software artifact. A skill or plugin can be read before it is approved, version-pinned so everyone runs the same one, and logged so its use leaves a record. For a regulated buyer, that is the difference between an AI capability a security team can sign off on and one it can only hope is being used safely. Building that substrate is production software engineering, which is the work Lightbridge Automation delivers through custom AI development.

RL Written by Robert LabardeeFounder and CEO

Last updated August 2026.

What is a Claude Skill? What is a Claude Plugin?

Lightbridge Automation defines a Claude Skill as a packaged, reusable capability: a folder of instructions and resources that teaches Claude how to perform one specific task on demand. A Claude Plugin bundles skills, hooks, agents, and tool connections into one distributable, installable unit that a team shares through a marketplace, rather than configuring each piece by hand.

Build custom skills and plugins See what MCP is

A skill packages one task; a plugin packages a whole capability set.

Claude Skills and Claude Plugins are two of the mechanisms Anthropic offers for extending what Claude can do beyond a single conversation. Both replace instructions that would otherwise be retyped, copy-pasted, or explained from memory with something a team writes once and reuses. They sit at different scopes: a skill is one packaged capability, while a plugin is a container that can hold several skills alongside other extension mechanisms.

The vocabulary runs through this and the surrounding guides. The explainer on Claude Code covers where skills, hooks, and plugins first became a developer-facing surface, and the AI and Claude glossary is a quick-reference companion. Because this vocabulary moves quickly, verify current specifics at docs.anthropic.com before planning around a particular detail.

Claude Skill

A packaged, reusable capability: a folder holding a set of instructions, and optionally supporting resources such as templates, reference material, or scripts, that teaches Claude how to perform one specific, repeatable task. A skill for drafting a weekly report differs from a skill for reconciling two ledgers; each is scoped to one job.

Claude Plugin

A distributable unit that bundles skills, hooks, agents, and tool connections, including Model Context Protocol server declarations, into one package a marketplace publishes. Installing a plugin gives a user or a team the whole bundle at once, instead of assembling each piece by hand.

Skills, plugins, hooks, and MCP connectors are four distinct pieces.

Skills and plugins are frequently mentioned alongside two other extension mechanisms: hooks and MCP connectors. They solve different problems and are easiest to tell apart side by side.

Skill

A packaged capability bundle: read this data set, draft this report, run this reconciliation. A folder of instructions Claude loads when a task calls for it, rather than content that sits in context at all times.

Plugin

A signed, distributable unit that bundles skills, hooks, agents, and MCP server declarations into one package a marketplace publishes, so a team runs the same reviewed capability set.

Hook

A shell command that fires on a Claude lifecycle event, such as before or after a tool call, and enforces a deterministic check the model cannot talk past: a secret-leak scan, a path allow-list, an audit-log emission.

MCP connector

An application-tier service that exposes an external system, a database, an API, a business platform, to Claude through the Model Context Protocol. See the Lightbridge Automation explainer on what MCP is for the full architecture.

A plugin is the unit that ties the other three together: it can declare a skill, a hook, and an MCP server connection as one signed package. For the connector piece specifically, the Lightbridge Automation guide to what MCP is covers the client-server architecture and the tools, resources, and prompts primitives an MCP connector can expose.

Skills and plugins matter because they turn habits into governed artifacts.

The business case for skills and plugins is not novelty. It is that a packaged, versioned capability can be reviewed, distributed consistently, and audited, in a way an individual's prompt habits cannot. That is the shift a regulated buyer's security and compliance review is actually asking for.

A skill turns a prompt into a reviewed artifact

Prompt engineering alone is a per-user habit: it produces inconsistent results and leaves no record of what was actually asked. A skill is a versioned file a team can read, review, and reuse the same way every time, closer to how software is built than how a single chat is typed.

A plugin is the unit of fleet-wide distribution

A plugin ships an entire capability set, skills, hooks, agents, and connections, to every authorized user at once. Updating the plugin updates the whole fleet; there is no per-person copy-paste to keep in sync.

A marketplace is the governance chokepoint

A private plugin marketplace names every approved plugin. Deciding what enters that list, and at what version, is how an organization decides which AI capabilities exist for its people, before any capability reaches a user.

Skills and plugins are software, not prompts

Treating a skill or plugin like any other production artifact, source review, version pinning, and an audit trail, is what lets a security team sign off on an AI capability rather than trusting an individual's prompt habits.

Skills and plugins are one piece of a larger build. Deciding which tasks deserve a skill, which standards belong in a hook, and how a plugin marketplace is governed is the work around the artifacts, not inside them. Teams that want a partner to build and centrally control that capability set can work with the Lightbridge Automation custom AI development practice, which pairs every build with the AI governance practice.

What is a Claude Skill and Plugin: frequently asked questions

What is a Claude Skill?: A Claude Skill is a packaged, reusable capability: a folder of instructions, and optionally supporting resources such as templates or scripts, that teaches Claude how to perform one specific, repeatable task. Rather than re-explaining a task in every conversation, a team writes the instructions once as a skill, and Claude draws on it whenever the task recurs. Skills are portable across the Claude product family, so the same skill can apply in claude.ai, Claude Code, the Claude Developer Platform, or Claude Cowork.
What is a Claude Plugin?: A Claude Plugin is a distributable unit that bundles skills, hooks, agents, and tool connections, including Model Context Protocol server declarations, into one package a marketplace publishes. Plugins are the extension and distribution mechanism most associated with Claude Code: instead of a developer configuring skills, hooks, and connectors one by one, installing a plugin gives them the whole reviewed bundle at once, at a version everyone on the team shares.
What is the difference between a Claude Skill and a Claude Plugin?: A skill is one packaged capability, scoped to one task. A plugin is a container that can hold several skills alongside hooks, agents, and tool connections, published as one signed unit through a marketplace. Put simply, a skill is an ingredient and a plugin is the dish: a team writes a skill to teach Claude one job, then a plugin packages that skill with whatever hooks and connectors the job needs so the whole capability installs and updates as one piece.
How is a Claude Skill different from a hook?: A skill teaches Claude how to do something; a hook enforces that something happens, or does not happen, regardless of what the model decides. A skill is instructions the model reads and reasons over. A hook is a shell command that fires automatically on a lifecycle event, such as before or after a tool call, and its exit code, not the model's judgment, controls whether the action proceeds. Teams pair the two: a skill for the task, a hook for the guardrail around it.
How is a Claude Plugin different from an MCP connector?: A plugin is a bundle of capabilities, skills, hooks, agents, and connections, packaged for one team to install together. An MCP connector is one of the things a plugin can declare: an application-tier service that exposes an external system to Claude through the Model Context Protocol. A plugin can ship an MCP server declaration alongside a skill and a hook, but the connector itself is a separate piece of software with its own authentication, scopes, and audit behavior. Lightbridge Automation covers that architecture in its guide to what MCP is.
Where can Claude Skills and Plugins be used?: Skills are designed to be portable across the Claude product family: the same skill can apply in claude.ai, Claude Code, the Claude Developer Platform, and Claude Cowork, so a capability written once does not need to be rewritten per surface. Plugins are the extension and marketplace mechanism most documented for Claude Code, where a team's plugin marketplace can be pushed to every developer's workstation through existing device management. Verify current availability at docs.anthropic.com before planning a rollout, since surface support moves quickly.
Who decides which skills and plugins a team can use?: In a governed deployment, the organization does, not each individual user. A team typically holds a private plugin marketplace, often a Git repository with a manifest that names every approved plugin and pins each one to a specific version. A capability reaches every authorized person by being added to that manifest; a capability is revoked by removing the entry. That central control point is what lets a security or compliance team review an AI capability before it reaches a user, rather than discovering it after the fact.
Why do Claude Skills and Plugins matter for a business, not just a developer?: Because they replace an unauditable habit, prompt engineering copied from person to person, with a reviewable software artifact. A skill or plugin can be read before it is approved, version-pinned so everyone runs the same one, and logged so its use leaves a record. For a regulated buyer, that is the difference between an AI capability a security team can sign off on and one it can only hope is being used safely. Building that substrate is production software engineering, which is the work Lightbridge Automation delivers through custom AI development.

This guide is independent, general educational information published by Lightbridge Automation. Claude Skills, Claude Plugins, and their supporting mechanisms change over time, so verify current details at docs.anthropic.com. Claude and Anthropic are trademarks of Anthropic, PBC. Lightbridge Automation is not affiliated with, endorsed by, or a partner of Anthropic, PBC.

From understanding skills and plugins to a governed deployment.

When the question shifts from what a Claude Skill or Plugin is to how to put one into production under real access control, Lightbridge Automation builds the skill, the plugin, and the guardrails around it as reviewable software.

Talk to Lightbridge Automation See custom AI development