AI glossary
Claude Skills: What They Are and How to Build Them
What Claude Skills (Agent Skills) are, how Claude decides when to use one and how to build your first.
By Chad Westleigh · Updated
What are Claude Skills?
Claude Skills package a repeatable way of doing something, such as your report format, brand style or a QA checklist, so Claude applies it whenever the task comes up. Anthropic describes Skills as "folders that include instructions, scripts, and resources that Claude can load when needed."
Anthropic ships its own Skills for Excel, PowerPoint, Word and fillable PDFs, and lets users and organizations write custom ones. On December 18, 2025, it added organization-wide Skill management for admins and published Agent Skills as an open standard.
How Claude Skills work: progressive disclosure
Claude Skills rely on what Anthropic calls progressive disclosure, which keeps many Skills available without filling Claude's context window:
- Metadata first. At startup, Claude sees only each installed Skill's name and description.
- Instructions when relevant. When a task matches a description, Claude reads that Skill's full SKILL.md.
- Files only as needed. Reference documents, templates and scripts bundled in the folder are opened or run only when the instructions call for them.
That is why the description matters: it's the only part Claude reads before deciding whether to use a Skill.
How to build Claude Skills, step by step
A first Skill takes an afternoon. Start from a task your team already does the same way every time.
1. Pick one repeatable procedure
A weekly report format, a proposal structure or a brand voice check: steps that fit on a page and recur often.
2. Write SKILL.md and add files
Begin the file with YAML frontmatter containing a name and a description, then write plain-Markdown instructions: steps, rules and examples of good output. Put templates or scripts in the same folder and reference them.
3. Make the description specific
Say what the Skill does and when to use it, in the words people actually type. A vague description means Claude won't pick the Skill up.
4. Test on real tasks
Run the Skill on past examples, check that Claude uses it when it should and ignores it when it shouldn't, and refine the wording.
5. Install, share and own it
Upload it in the Claude apps, add it to ~/.claude/skills or a plugin in Claude Code, or manage it through the API. Then name an owner who updates it when the process changes.
Claude Skills vs Projects vs MCP vs saved prompts
| Option | What it holds | When it's used | Use it for |
|---|---|---|---|
| Claude Skill | Procedures, rules, templates and scripts | Loaded when a task matches its description | Doing a recurring task the same way every time |
| Claude Project | Instructions plus knowledge files for one body of work | Whenever someone works inside that Project | Context for an ongoing workstream, client or team |
| MCP connector | Access to an external system's data and actions | When Claude needs to read or change something in another tool | Live data: CRM records, orders, files, tickets |
| Saved prompt | A reusable instruction a person pastes in | When a person chooses it | Quick, varied tasks that don't need packaging |
How Operator Intelligence uses Claude Skills
Operator Intelligence builds Claude Skills as part of its Claude training and implementation work. In Claude training, each team writes and tests one or two Skills on its own work; Claude Code training covers Skills for operators who aren't developers. For broader rollouts, our Claude for business implementation pairs Skills with Projects, connectors and the Model Context Protocol, so Claude knows both how to do the work and where the data lives. Teams still choosing an assistant can start with our Claude vs ChatGPT for business comparison.
Each Skill we build has an owner and gets reviewed when the process changes; third-party Skills are audited before install. Many start as a well-used entry in a team's prompt library.
Claude is a trademark of Anthropic. Operator Intelligence is an independent consultancy and is not affiliated with or endorsed by Anthropic.
Frequently asked questions
What is the difference between Claude Skills and MCP?
MCP connects Claude to external tools and data; Skills teach Claude how to carry out a task. They work together: an MCP connector can give Claude access to your CRM, and a Skill can define how to turn those records into your weekly pipeline report.
Do I need to code to build a Claude Skill?
No. A Skill can be a single SKILL.md file written in plain language. Scripts are optional and only needed when a Skill should run code, such as transforming a spreadsheet.
Where do Claude Skills work?
Anthropic's October 2025 launch made Skills available in the Claude apps on Pro, Max, Team and Enterprise plans, in Claude Code, and through the API and Agent SDK. As an open standard, the format can also travel to other tools, which is why we rebuild custom GPTs and Gems as SKILL.md-style assets; see our custom GPT migration guide.
Are Claude Skills safe to install?
For third-party Skills, Anthropic advises installing only from trusted sources and auditing the files first, especially scripts and instructions to connect to outside network sources, since a malicious Skill could try to exfiltrate data.
Build your team's first Claude Skills
We'll pick the procedures worth packaging, build and test the Skills in your Claude account and train the people who maintain them. Or start with the AI readiness assessment.
Book a scoping callSources
- Anthropic (claude.com blog) (2025): Skills announced Oct 16, 2025: 'folders that include instructions, scripts, and resources that Claude can load when needed'; available to Pro, Max, Team and Enterprise users in Claude apps (Team/Enterprise admins enable org-wide), Claude Code (plugins or ~/.claude/skills), API (/v1/skills; requires Code Execution Tool beta) and Agent SDK; Anthropic-built skills for Excel, PowerPoint, Word and fillable PDFs. Dec 18, 2025 update: organization-wide skill management, partner skills directory, Agent Skills published as an open standard (agentskills.io).
- Anthropic Engineering, 'Equipping agents for the real world with Agent Skills' (2025): Progressive disclosure: name and description preloaded at startup; full SKILL.md read when relevant; bundled files read only as needed. SKILL.md must start with YAML frontmatter containing name and description. Install skills only from trusted sources and audit less-trusted ones; malicious skills could direct Claude to exfiltrate data. Published Oct 16, 2025.
Get the Operator Intelligence briefing
A short, practical note every two weeks on what is working with AI in real businesses. No spam; unsubscribe anytime.
Related
Claude training for business teams
Projects, Skills and connectors configured in your account.
Claude Code training
Claude Code and Skills for operators and non-developers.
Model Context Protocol (MCP)
The open standard for connecting AI to tools and data.
Prompt library
A team's shared, tested AI prompts.
ChatGPT and Gemini training
Plugins and Gemini skills built from the same portable instructions.
AI glossary
Sixteen AI terms defined for business leaders.