What Are Claude Skills and Plugins? A Plain-English Guide
Reusable instructions, packaged workflows and tool access explained.
Part of the guide Rhize Plugins for Claude Code: Which Ones Should You Install?

On this page
A Claude skill is reusable instructions for a particular job. A plugin packages related skills, commands, hooks or tool connections so you can install and maintain them together. For a business, the point is to stop rewriting the same brief while keeping a person responsible for checking important outputs.
If you've ever explained the same task to an AI chat window for the fourth time, this is the fix. Here's what each piece is, what it does for a business, and what to watch out for.
What is a Claude skill?
A Claude skill describes the steps, constraints and output expected for a task. Think of an SOP: which records to read, what tone to use and when to escalate. Claude can load relevant skills or you can invoke one explicitly. The instructions remain available across tasks, but their presence is not proof that each output followed them correctly.
For example, an invoice-follow-up skill might say: read the invoice status, draft a polite reminder, exclude disputed invoices and stop for review before sending. Test it against an unpaid invoice, an already-paid invoice and a disputed one. A useful skill should handle those differences instead of blindly repeating the same message.
A skill saves you from re-explaining the procedure. Scheduling, access to customer data and permission to send messages remain separate parts of the system.
What is a plugin?
A plugin is the binder that holds it all: a package of related skills, ready-made commands, and connections to outside tools, installed in one step. If a skill is one SOP, a plugin is the full operations manual for one area of your business : plus the keys to the filing cabinets it needs.
Open the binder and you'll find three compartments:
- Skills: reusable instructions that Claude can load for relevant work or that you can invoke directly.
- Commands : shortcuts you invoke on purpose, like
/seo-audit, that kick off a defined job. - Tool connections : the wiring (called MCP servers) that lets Claude pull live data from real tools instead of guessing.
For a concrete example, a reporting plugin might bundle a report-writing skill, a command to start the report and a read-only connector to the data source. Installing the bundle makes those pieces discoverable; you still configure the connector and review the report before sharing it.
Skills vs. plugins vs. MCP servers : what's the difference?
Skills are the instructions, MCP servers are the tool access, and plugins are the package that bundles both. Here's the clean version:
| Skill | Plugin | MCP server | |
|---|---|---|---|
| What it is | Written instructions for one job | A bundle: skills + commands + connections | A connector to one real tool (CRM, calendar, data source) |
| Plain-English version | An SOP | The binder of SOPs, plus the keys | The standard plug into a filing cabinet |
| What it gives Claude | Know-how | Everything for one area, in one install | Live data and the ability to act in that tool |
| How you get it | Written, or installed inside a plugin | Installed once from a marketplace | Usually comes wired into a plugin |
If you want the full story on that third column, it gets its own explainer: What is an MCP server?
What can a business actually do with them?
A plugin can give Claude a repeatable procedure and the tools needed for a domain. The following examples describe available workflows, not a guarantee of output quality or saved hours:
- SEO audits on live data. The seo-aeo-geo plugin runs site crawls, keyword research, and AI-visibility checks using current search data : not a chatbot's memory of the internet. One command, a real report.
- A second brain that answers back. The obsidian-second-brain plugin connects Claude to your notes, so "what did we decide about that client in March?" gets answered from your own records.
- Projects that start planned. The project-launcher plugin takes a rough idea and turns it into a researched, stress-tested spec before anything gets built.
- Consistent client work. Skills that carry your brand voice, your onboarding steps, your delivery checklists : so the output sounds like you on the intern's laptop too.
The test is whether the workflow produces a useful, checked result with less repeated explanation. Compare a representative task before and after adoption, including the time spent correcting the output.
Go deeper: Rhize Plugins for Claude Code: Which Ones Should You Install? The marketplace guide matches capabilities to tasks and lists setup requirements.
Are they safe to install?
Read a skill and its supporting scripts before trusting it with your work. Instructions can be wrong or malicious, and a plugin can request tool access beyond what a task needs. Review the publisher, permissions and update history, then start with a bounded example and inspect the result.
The fix isn't fear : it's a vetting step. Install from sources you trust, read what a skill actually does before it runs, and quarantine anything third-party until it's been checked. We built tooling for exactly this: Skill Forge scans and quarantines outside skills before they're promoted into use.
Go deeper: Skill Forge: a security gate for agent skills : how the quarantine-scan-promote pipeline works.
The former rhize-meta plugin has been retired. Its skill-vetting and refinement responsibilities moved to the separate Skill Forge tooling; use the current marketplace guide rather than installing the old package name.
Where do you get them?
Claude Code supports plugin marketplaces: catalogs you register and then install selected plugins from. The Rhize marketplace currently lists 12 plugins, covering setup, search, knowledge work, planning, development, operations and research. Read each package's requirements before choosing it.
In Claude Code, register the marketplace with /plugin marketplace add https://github.com/Rhize-Media/rhize-plugins, then install the chosen package with /plugin install <plugin-name>@rhize-plugins. Restart the host, confirm the skill or command appears, and test a read-only task. Support for another host must be checked in that plugin's documentation; a common package format does not prove identical behavior everywhere.
Frequently Asked Questions
Are Claude skills free?
Do skills work in the Claude app, or just Claude Code?
Can Claude write its own skills?
What happens if two skills conflict?
Start with one workflow you repeat often and choose the smallest plugin that supports it. Browse the Rhize plugin marketplace, inspect the instructions and test an example before expanding its access.
Sources and implementation references
Claude skills: https://code.claude.com/docs/en/skills
Current marketplace: https://github.com/Rhize-Media/rhize-plugins
Want this working in your business — not just on paper?
Book a 30-min call to talk through your workflow, ask your questions, and discuss a useful next step. No pitch, no obligation.
Book a 30-min call