Agent Skills

Agent Skills are modular capabilities that extend Claude's functionality for domain-specific tasks. They operate through a progressive, filesystem-based loading mechanism that optimizes token usage and performance.

Architecture - Three Content Levels:

1. Metadata (Always Loaded ~100 tokens)
2. Lightweight system prompt inclusion
3. Describes the skill's purpose and when to use it

4. Minimal token consumption

5. Instructions (Loaded When Triggered <5,000 tokens)

6. Procedural knowledge and workflows
7. Loaded dynamically when relevant

8. Step-by-step task guidance

9. Resources and Code (Loaded as Needed)

10. Optional executable scripts
11. Reference materials
12. Executed via bash without consuming context window

Key Benefits:

• Specialize Claude for domain-specific tasks
• Reduce repetitive instructions
• Enable complex workflow composition
• Efficient token usage through progressive loading

Structure:

• Requires a SKILL.md file
• Must include name (64 character max)
• Must include description (1024 character max)

Availability:

• Claude API
• Claude Code
• Claude.ai (with limitations)

Pre-built Skills:

• PowerPoint processing
• Excel data manipulation
• Word document editing
• PDF document handling

Security:

• No network access by default
• Pre-configured dependencies
• Sandboxed execution environment

Links

• Agent Skills

https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview

Equipping agents for the real world with Agent Skills

https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills