skilllite 0.1.3

A lightweight Skills execution engine with LLM integration for LLM agents

SkillLite Python SDK

A Python SDK for the SkillLite execution engine, using OpenAI-compatible API format as the unified interface.

Supported Providers

Works with any OpenAI-compatible LLM provider:

• OpenAI (GPT-4, GPT-3.5, etc.)
• Azure OpenAI
• Anthropic Claude (via OpenAI-compatible endpoint or native)
• Google Gemini (via OpenAI-compatible endpoint)
• Local models (Ollama, vLLM, LMStudio, etc.)
• DeepSeek, Qwen, Moonshot, Zhipu, and other providers

Installation

pip install skilllite

# With OpenAI SDK (recommended, works with all compatible providers)
pip install skilllite[openai]

# With Anthropic SDK (for Claude's native API)
pip install skilllite[anthropic]

Prerequisites

You need to have the skillbox binary installed:

# From the skillbox directory
cargo install --path .

Quick Start

Basic Usage (Universal - Works with Any Provider)

from openai import OpenAI
from skilllite import SkillManager

# Works with ANY OpenAI-compatible provider
# Just change base_url and api_key for different providers:

# OpenAI
client = OpenAI()

# Ollama (local)
# client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")

# DeepSeek
# client = OpenAI(base_url="https://api.deepseek.com/v1", api_key="...")

# Qwen (通义千问)
# client = OpenAI(base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", api_key="...")

# Moonshot (月之暗面)
# client = OpenAI(base_url="https://api.moonshot.cn/v1", api_key="...")

# Initialize SkillManager
manager = SkillManager(skills_dir="./my_skills")

# Get tools (OpenAI-compatible format - works with all providers)
tools = manager.get_tools()

# Call any OpenAI-compatible API
response = client.chat.completions.create(
    model="gpt-4",  # or "llama2", "deepseek-chat", "qwen-turbo", etc.
    tools=tools,
    messages=[{"role": "user", "content": "Please help me with..."}]
)

# Handle tool calls (same code works for all providers!)
if response.choices[0].message.tool_calls:
    tool_results = manager.handle_tool_calls(response)
    
    # Continue conversation with results
    messages = [
        {"role": "user", "content": "Please help me with..."},
        response.choices[0].message,
        *[r.to_openai_format() for r in tool_results]
    ]
    
    follow_up = client.chat.completions.create(
        model="gpt-4",
        tools=tools,
        messages=messages
    )

Agentic Loop (Automatic Multi-turn Tool Execution)

from openai import OpenAI
from skilllite import SkillManager

# Works with any provider
client = OpenAI()  # or OpenAI(base_url="...", api_key="...")
manager = SkillManager(skills_dir="./my_skills")

# Create an agentic loop
loop = manager.create_agentic_loop(
    client=client,
    model="gpt-4",
    system_prompt="You are a helpful assistant with access to various skills.",
    max_iterations=10,
    temperature=0.7  # Additional kwargs passed to chat.completions.create()
)

# Run until completion - handles multiple tool calls automatically
final_response = loop.run("Please analyze this data and generate a report.")
print(final_response.choices[0].message.content)

Claude Native API (Optional)

If you prefer using Claude's native API directly:

import anthropic
from skilllite import SkillManager

client = anthropic.Anthropic()
manager = SkillManager(skills_dir="./my_skills")

# Use Claude-specific methods
tools = manager.get_tools_for_claude_native()
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=4096,
    tools=tools,
    messages=[{"role": "user", "content": "..."}]
)

if response.stop_reason == "tool_use":
    results = manager.handle_tool_calls_claude_native(response)

Creating Skills

Skills are defined in directories with a SKILL.md file:

my_skills/
├── web-search/
│   ├── SKILL.md           # Metadata and docs (includes dependency declaration)
│   └── scripts/
│       └── main.py
└── calculator/
    ├── SKILL.md
    └── scripts/
        └── main.py

Note: Python dependencies are declared in the compatibility field of SKILL.md, not in a separate requirements.txt file.

SKILL.md Format

---
name: web-search
description: Search the web for information
compatibility: Requires Python 3.x with requests library, network access
license: MIT
metadata:
  author: example-org
  version: "1.0"
---

# Web Search Skill

This skill searches the web for information.

## Input Parameters

- `query`: The search query (required)

The compatibility field is used to:

• Detect language (Python/Node/Bash)
• Enable network access (keywords: network, internet, http, api, web)
• Auto-install dependencies (known packages like requests, pandas, axios, etc.)

Skill Entry Point

# main.py
import json
import sys

def main():
    # Read input from stdin
    input_data = json.loads(sys.stdin.read())
    
    # Process the input
    query = input_data.get("query", "")
    
    # Do something...
    result = {"results": [f"Result for: {query}"]}
    
    # Output JSON to stdout
    print(json.dumps(result))

if __name__ == "__main__":
    main()

API Reference

SkillManager

The main class for managing and executing skills.

Constructor

SkillManager(
    skills_dir: Optional[str] = None,  # Directory containing skills
    binary_path: Optional[str] = None,  # Path to skillbox binary
    cache_dir: Optional[str] = None,    # Cache directory for venvs
    allow_network: bool = False          # Default network access
)

Methods

• scan_directory(directory) - Scan for skills
• register_skill(skill_dir) - Register a single skill
• get_skill(name) - Get skill by name
• list_skills() - List all skills
• get_tools_for_claude() - Get Claude-format tools
• get_tools_for_openai() - Get OpenAI-format tools
• execute(skill_name, input_data) - Execute a skill
• handle_tool_calls(response, format) - Handle LLM tool calls
• create_agentic_loop(...) - Create an agentic loop

ToolFormat

Enum for LLM provider formats:

• ToolFormat.CLAUDE
• ToolFormat.OPENAI

Framework Adapters

SkillLite provides adapters for popular AI frameworks with security confirmation support.

LangChain Integration

For LangChain/LangGraph integration, we recommend using the dedicated langchain-skilllite package:

pip install langchain-skilllite

from langchain_skilllite import SkillLiteToolkit
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

# Load all skills from a directory as LangChain tools
tools = SkillLiteToolkit.from_directory("./skills")

# Create a LangGraph agent
agent = create_react_agent(ChatOpenAI(model="gpt-4"), tools)

# Run the agent
result = agent.invoke({
    "messages": [("user", "Convert 'hello world' to uppercase")]
})

With security confirmation (sandbox_level=3):

def confirm_execution(report: str, scan_id: str) -> bool:
    print(report)
    return input("Continue? [y/N]: ").lower() == 'y'

tools = SkillLiteToolkit.from_directory(
    "./skills",
    sandbox_level=3,  # 1=no sandbox, 2=sandbox only, 3=sandbox+scan
    confirmation_callback=confirm_execution
)

For more details, see the langchain-skilllite documentation.

Alternative: You can also use the built-in adapter:

from skilllite import SkillManager
from skilllite.core.adapters.langchain import SkillLiteToolkit

manager = SkillManager(skills_dir="./skills")
tools = SkillLiteToolkit.from_manager(manager).get_tools()

LlamaIndex Integration

from skilllite import SkillManager
from skilllite.core.adapters.llamaindex import SkillLiteToolSpec

manager = SkillManager(skills_dir="./skills")

# Basic usage
tool_spec = SkillLiteToolSpec.from_manager(manager)
tools = tool_spec.to_tool_list()

# With security confirmation
def confirm(report: str, scan_id: str) -> bool:
    print(report)
    return input("Continue? [y/N]: ").lower() == 'y'

tool_spec = SkillLiteToolSpec.from_manager(
    manager,
    sandbox_level=3,
    confirmation_callback=confirm
)

# Use with LlamaIndex agent
from llama_index.core.agent import ReActAgent
agent = ReActAgent.from_tools(tools, llm=llm)

Security Levels

Level	Description
1	No sandbox - direct execution
2	Sandbox isolation only
3	Sandbox + static security scan (requires confirmation for high-severity issues)

OpenCode Integration

SkillLite can be integrated with OpenCode as an MCP (Model Context Protocol) server, providing secure sandbox execution capabilities.

Quick Setup

# Install with MCP support
pip install skilllite[mcp]

# One-command setup for OpenCode
skilllite init-opencode

# Start OpenCode
opencode

The init-opencode command automatically:

• Detects the best way to start the MCP server (uvx, pipx, skilllite, or python)
• Creates opencode.json with optimal configuration
• Generates .opencode/skills/skilllite/SKILL.md with usage instructions
• Discovers your pre-defined skills

Available MCP Tools

Tool	Description
skilllite_list_skills	List all available skills
skilllite_get_skill_info	Get skill details and input schema
skilllite_run_skill	Execute a pre-defined skill
skilllite_scan_code	Scan code for security issues
skilllite_execute_code	Execute code in secure sandbox

Security Features

• System-level Sandbox: macOS Seatbelt / Linux Namespace isolation
• Security Scanning: Static analysis before execution
• User Confirmation: Dangerous code requires explicit approval
• Scan ID Verification: Prevents code modification between scan and execution

For detailed documentation, see OpenCode Integration Tutorial.

CLI Commands

skilllite install        # Install skillbox sandbox binary
skilllite uninstall      # Remove skillbox binary
skilllite status         # Show installation status
skilllite version        # Show version information
skilllite mcp            # Start MCP server
skilllite init-opencode  # Initialize OpenCode integration

License

MIT License