Skills Guide

Skills are loadable knowledge modules that provide domain expertise on demand. Unlike always-present context, skills are loaded when needed, keeping the base context lean. This section covers how to use existing skills, create custom ones, and integrate them into your workflows.

Section Contents

Page Description
Curl HTTP requests and API interactions
Image Vision Image understanding and analysis
Playwright Browser automation and testing

Quick Tips

  • Load on demand - Only load skills when the domain knowledge is needed
  • Check before loading - Use info operation to preview skill metadata
  • Skills complement docs - They provide workflows, not just reference
  • Workspace priority - Project skills override user skills
  • Companion files - Skills can reference additional resources

Skills vs Context

Aspect Skills Context
Loading On-demand Always present
Scope Domain-specific General behavior
Size Can be large Should be concise
Location Discoverable directories Bundle-defined

Where to Start

Want HTTP capabilities? Start with Curl for HTTP requests and API interactions.

Working with images? Jump to Image Vision for image understanding.

Need browser automation? See Playwright for browser testing.

Skill Operations

# List all available skills
load_skill(list=True)

# Search for skills by topic
load_skill(search="python")

# Check skill metadata
load_skill(info="skill-name")

# Load full skill content
load_skill(skill_name="skill-name")

Skill Discovery Paths

Skills are discovered from configured directories in priority order:

  1. Workspace skills - .amplifier/skills/ (project-specific)
  2. User skills - ~/.amplifier/skills/ (personal)
  3. Bundle skills - From active bundles
  4. Collection skills - From installed collections

First match wins—workspace skills override user skills.

Next Steps

After understanding skills, explore Advanced topics for complex patterns like recipes and multi-agent workflows.