Claude
How Agent Skills work in Claude. Progressive disclosure, SKILL.md format, the VM architecture, pre-built skills, custom skills, and where skills run across Claude products.
For the official Claude documentation, see Agent Skills Overview on Anthropic's platform.
How Claude Uses Skills
Claude operates in a virtual machine with filesystem access. Skills exist as directories on this VM, and Claude interacts with them using standard bash commands: reading files, executing scripts, and navigating directories just like you would on your computer.
This setup allows for progressive disclosure. Claude loads skill content in stages instead of taking everything at once.
Three Levels of Loading
| Level | When Loaded | Token Cost | What's Loaded |
|---|---|---|---|
| Metadata | At startup | ~100 tokens | name and description from YAML frontmatter |
| Instructions | When skill is triggered | Under 5,000 tokens | Full SKILL.md body |
| Resources | As needed | Effectively unlimited | Scripts, reference files, templates — executed or read on demand |
This means you can install many skills without a context penalty. Claude recognizes each skill's existence and when to use it until activation is required.
What Happens When a Skill Activates
Startup
Claude's system prompt includes the name and description of all available skills.
User request
When a user asks for something that matches a skill's description, Claude decides to activate it.
Read instructions
Claude reads the SKILL.md file from the filesystem using bash.
Load selectively
If the instructions mention additional files, Claude reads only what is necessary.
Execute
Claude follows the instructions and runs scripts if needed. Script code never enters the context window; only the output does.
SKILL.md in Claude
Claude follows the standard Agent Skills format. Every skill needs a SKILL.md file with YAML frontmatter:
Field Requirements
| Field | Rules |
|---|---|
| name | Max 64 characters. Lowercase letters, numbers, and hyphens only. Cannot contain "anthropic" or "claude." |
| description | Max 1,024 characters. Non-empty. Should explain what the skill does and when Claude should use it. |
The description field is key; it helps Claude decide whether to activate your skill for a particular task.
Skill Directory Structure
On-demand file access
Claude reads only the files needed for each task, so a skill can include many reference files without a context penalty.
Efficient script execution
When Claude runs a script, only the output enters the context window, not the script code itself.
No practical size limit
Skills can bundle extensive documentation, large datasets, or numerous examples without compromising performance.
Where Skills Work
Skills are available across Claude's products, but each has different features:
- Pre-built skills: PowerPoint, Excel, Word, and PDF skills work automatically, requiring no setup.
- Custom skills: Upload as zip files through Settings > Features (Pro, Max, Team, and Enterprise plans).
- Scope: Individual users only — each team member uploads separately.
Pre-Built Skills
Claude comes with pre-built skills for common document tasks:
| Skill | Capabilities |
|---|---|
PowerPoint (pptx) | Create presentations, edit slides, analyze content |
Excel (xlsx) | Create spreadsheets, analyze data, generate charts |
Word (docx) | Create documents, edit content, format text |
PDF (pdf) | Generate formatted PDF documents and reports |
These are available on Claude.ai and the Claude API without any setup.
Tips for Claude-Specific Skills
Selling Claude Skills on Skly
When creating skills for the Skly marketplace that target Claude:
- List "Claude" as a supported AI tool so buyers can find Claude-compatible skills.
- Follow the Agent Skills format — Claude expects
SKILL.mdwith proper YAML frontmatter. - Include clear activation triggers — write descriptions that explain when Claude should use your skill.
- Bundle reference files — take advantage of Claude's progressive disclosure to include detailed documentation without context penalty.