assafelovic/skyll

Skyll - Agent Skill Discovery

Skyll enables agents to dynamically discover and retrieve skills at runtime. Instead of having all skills pre-loaded, agents can search for relevant skills on-demand and inject them into context.

When to Use This Skill

Use Skyll when:

• The current task requires specialized knowledge or workflows not in context
• Looking for best practices for a specific framework, library, or domain
• Needing procedural knowledge for complex multi-step tasks
• Wanting to enhance capabilities with community-contributed skills

Quick Usage

Option 1: Python Client (Recommended)

from skyll import Skyll

async with Skyll() as client:
    # Search for relevant skills
    skills = await client.search("react performance optimization", limit=3)
    
    for skill in skills:
        print(f"Skill: {skill.title}")
        print(f"Score: {skill.relevance_score}")
        print(skill.content)  # Full SKILL.md content

Option 2: REST API

Search for skills:

curl "https://api.skyll.app/search?q=react+performance&limit=3"

Get a specific skill:

curl "https://api.skyll.app/skills/anthropics/skills/skill-creator"

Response Format

The API returns ranked skills with relevance scores (0-100):

{
  "query": "react performance",
  "count": 3,
  "skills": [
    {
      "id": "react-best-practices",
      "title": "React Best Practices",
      "description": "Performance optimization for React and Next.js",
      "source": "vercel-labs/agent-skills",
      "relevance_score": 85.5,
      "install_count": 82800,
      "content": "# React Best Practices\n\n## Performance\n..."
    }
  ]
}

Key Fields

| Field | Description | |-------|-------------| | id | Skill identifier | | title | Human-readable title | | description | What the skill does | | source | GitHub repository (owner/repo) | | relevance_score | 0-100 match score | | install_count | Popularity from skills.sh | | content | Full SKILL.md markdown content | | references | Additional reference files (if requested) |

Search Parameters

| Parameter | Default | Description | |-----------|---------|-------------| | q / query | required | Search query | | limit | 10 | Max results (1-50) | | include_content | true | Fetch full SKILL.md content | | include_references | false | Include reference files |

Example Workflow

When an agent needs to build a React Native app but lacks mobile expertise:

1. Search for relevant skills:

   skills = await client.search("react native best practices", limit=2)
   

2. Inject top skill into context:

   context = skills[0].content  # Full SKILL.md content
   

3. Use the skill's knowledge to complete the task

API Endpoints

| Endpoint | Description | |----------|-------------| | GET /search?q={query} | Search skills | | POST /search | Search with JSON body | | GET /skills/{source}/{skill_id} | Get specific skill | | GET /health | Health check | | GET /docs | Interactive API documentation |

Installation

For Python agents:

pip install skyll

For other languages, use the REST API directly at https://api.skyll.app.

Links

• API: https://api.skyll.app
• Docs: https://api.skyll.app/docs
• Demo: https://skyll.app
• GitHub: https://github.com/assafelovic/skyll