What they are, what's inside them, and where they work.
A skill is a reusable set of instructions that teaches Claude how to do a specific task — your way, every time. Think of it as writing an SOP for a team member who follows it perfectly.
A skill is literally a text file with instructions. That's it. You write what you want Claude to do, how you want it done, and save it. Every time you trigger that skill, Claude follows those instructions. No code. No programming. Just clear directions.
Every skill is a folder with one essential file — SKILL.md — and optionally some supporting files.
The SKILL.md file has two parts:
You can control exactly which tools Claude can access when running a skill. Here's what's available:
Read files, search through folders, find content across your documents.
Create new files, edit existing ones. Build documents, spreadsheets, HTML pages.
Access Gmail, Google Drive, Canva, Slack, Notion — any connected tool.
Search the web, fetch pages, interact with websites via Claude in Chrome.
A skill can trigger other skills. Chain them together for complex workflows.
The skill can pause and ask you a question if it needs more info before continuing.
You can lock a skill down to read-only (it can look but not change anything), or give it full access. This is useful for skills you share with others — you control the permissions.
Learn exactly how inside the Wright Mode Membership — or go deep with the Claude Code Workshop.
This is the important bit. Skills don't work the same way everywhere.
Skills work in the browser via /commands. You can trigger skills from claude.ai directly. May have fewer tool integrations than Desktop.
The power home for skills. /commands, auto-triggering, full tool access, file system, connectors, scheduled tasks.
Full skill support with all features. Same SKILL.md format. Most powerful tool access and agent capabilities.
Skills don't automatically sync between the three Claudes. If you build a skill in Cowork and want it in Claude Code (or vice versa), you'd download/copy the skill folder and put it in the right place. Same file format — just different locations on your machine.
Skills are stored in specific folders. Claude looks in these locations (in order of priority):
What the SKILL.md says: "You are a social media caption writer. Use the brand voice guidelines in the reference file. Every caption must include a hook (first line), body (2-3 lines), and CTA. Use Australian spelling. Never use the word 'leverage'. Keep captions under 150 words."
Supporting files: brand-voice-guidelines.md (reference), good-captions.md (examples)
How you trigger it: Type /caption-writer or just say "write me a caption for…" and Claude recognises it should use this skill.
What the SKILL.md says: "Read the data file provided, analyse key metrics, create an interactive HTML dashboard with charts. Use the client's brand colours (provided in the reference file). Include executive summary, key findings, and recommendations."
Supporting files: report-template.md (template), client-colours.md (reference)
Tools it uses: Read (to access data), Write (to create the HTML file), connectors (to pull data from Google Sheets if needed)
What the SKILL.md says: "Take the meeting transcript from Google Drive. Extract: key decisions, action items (with who and deadline), follow-ups needed, and notable quotes. Format as a clean document. Save to the client's folder."
Supporting files: meeting-template.md (consistent structure every time)
Tools it uses: Google Drive connector (to find transcript), Read, Write (to create summary), possibly Notion connector (to log action items)
A skill is just a text file with instructions. You don't need to code. You don't need to be technical. If you can write a clear brief for a contractor or VA, you can write a skill.
If a skill is a single recipe, a plugin is the whole cookbook — bundling skills, connectors, agents, and tools into one installable package.
A plugin is a bundle. Instead of setting up 5 skills, 3 connectors, and 2 agents one by one, you install a plugin and it gives you everything at once. It's like hiring a specialist who brings their own tools, playbooks, and workflows on day one.
A plugin is a folder that can contain up to six different types of components. You don't need all of them — use what you need.
| Skill | Plugin | |
|---|---|---|
| What it is | A single set of task instructions | A bundle of skills + connectors + agents + hooks |
| Analogy | One recipe you wrote | A whole cookbook from a specialist |
| Files | One SKILL.md + optional reference files | A folder with multiple components |
| Installation | Create it yourself or paste instructions | One-click install from a marketplace |
| Distribution | Copy the file manually | Published to a marketplace, installed by anyone |
| Naming | /skill-name | /plugin-name:skill-name |
| Best for | Your own repeatable tasks | Sharing workflows, team toolkits, packaged solutions |
Wright Mode members get access to real plugin examples, templates, and step-by-step build guides.
The browser has its own plugin marketplace at claude.com/plugins. These are different from Desktop/Code plugins — they're not interchangeable.
Install from marketplace or custom sources. Full access to all plugin components: skills, connectors, agents, hooks.
Same plugin format as Desktop. Install via CLI. Test locally before publishing. Full access to everything.
This is important: the plugin system in claude.ai (browser) is completely separate from the plugin system in Claude Desktop and Claude Code. A plugin built for Desktop/Code won't work in the browser, and vice versa. They're different architectures designed for different environments.
Plugins can be installed at different levels, depending on who needs them:
What's bundled:
→ ⚡ Skills: IG hook creator, DM responder, video repurposer, trial reels creator, YouTube scripter
→ 📋 Reference files: Brand voice guidelines, content pillars, audience profiles
→ ⚙️ Settings: Default model, Australian spelling rules
What it does: Install it and you immediately have Brooke's entire content workflow — brand voice, skills, and templates — ready to go. One install instead of setting up 10+ individual skills.
What's bundled:
→ 🔌 MCP Server: GitHub API integration (issues, PRs, repos)
→ ⚡ Skills: PR review, commit formatting, issue triage
→ 🪝 Hooks: Auto-lint before commits, auto-review on PR creation
What it does: Claude can create issues, review pull requests, manage repos — all through one plugin install. The hooks automatically run quality checks.
What's bundled:
→ 🔌 MCP Server: Apify (web scraping), Google Analytics connector
→ ⚡ Skills: Metrics dashboard builder, competitor report generator, social audit skill
→ 🤖 Agents: "Data Analyst" agent focused on interpreting metrics
What it does: Scrape data, analyse it, and build beautiful reports — all from one plugin. The Data Analyst agent specialises in finding patterns humans miss.
Anthropic runs an official plugin marketplace. You can browse and install plugins with one click. Plugins cover everything from code quality tools to social media workflows to data analysis. You can also build your own plugin and publish it for others to use — or keep it private for your team.
Skills are the recipes. Plugins are the cookbooks. MCP Servers (connectors) are the kitchen appliances. A plugin bundles all three so you can go from zero to working workflow with one install.