universal-skill-tree-naneg 0.1.0

Plug & Play skill framework — connect any LLM to your PC in 3 lines

🌳 Universal Skill Tree (UST)

Plug & Play skill framework — connect any LLM to your PC in 3 lines.

from ust import enable_branch, USTAdapter

enable_branch("system")
agent = USTAdapter(api_key="sk-or-...", model="openai/gpt-4o-mini")
reply = agent.chat_sync("Open Chrome and go to github.com")

---

The Problem UST Solves

Building a "Jarvis"-style AI agent today means:

• 80+ packages in requirements.txt
• Version conflicts, broken installs, dependency hell
• Hours of setup before writing a single line of agent code

UST separates the Brain (your LLM) from the Hands (the tools).
Install only what you need. Connect in seconds. Works with any LLM.

---

Installation

# Core only (ultra-light: just httpx + pydantic)
pip install universal-skill-tree

# With PC control skills
pip install 'universal-skill-tree[system]'

# With web search
pip install 'universal-skill-tree[system,web]'

# Everything
pip install 'universal-skill-tree[all]'

Or with uv (faster):

uv add 'universal-skill-tree[system]'

---

Quick Start

from ust import enable_branch, USTAdapter

enable_branch("system")   # Load PC control skills

agent = USTAdapter(
    api_key="sk-or-...",          # OpenRouter / OpenAI / local
    model="openai/gpt-4o-mini",   # Any OpenAI-compatible model
)

# Sync
reply = agent.chat_sync("What's my CPU usage?")

# Async
import asyncio
reply = asyncio.run(agent.chat("Open Notepad and type Hello World"))

---

Skill Branches

Branch	Skills	Install
system	Shell commands, CPU/RAM info, mouse/keyboard, screenshots	[system]
files	Read/write files, PDF, Excel, Word docs	[files]
web	DuckDuckGo search, webpage fetch, scraping	[web]
media	System volume, audio control	[media]
apps	App launcher, process manager	[apps]
vision	Screenshot + AI analysis	[vision]
osint	Sherlock, Osintgram, network scans (whois, dns, geoip)	[osint]
cyber	Sniffnet, Trippy, network packet sniffing	[cyber]
crypto	Cryptocurrency prices via CCXT	[crypto]
selfhosted	Awesome-selfhosted apps search and Docker control	[selfhosted]
aitools	Awesome-ai-tools search	[aitools]
agent_skills	Awesome-agent-skills search	[agent_skills]
mcp	Awesome-mcp-servers search, Chrome DevTools MCP	[mcp]
ai	CrewAI tasks, LiteLLM arbitrary LLM calls	[ai]
browser	Playwright browser automation	[browser]

---

System Branch — Available Skills

Skill	What it does
execute_command	Run any shell command (cmd/bash)
get_system_info	CPU%, RAM, disk, battery, top processes
manage_processes	List / kill / start processes
control_window	Minimize, maximize, close, focus windows
control_mouse	Move, click, scroll
type_text	Type text, keyboard shortcuts
manage_clipboard	Read / write clipboard
take_screenshot	Screenshot full screen or a region
set_volume	Set volume 0–100, mute/unmute (Windows)
open_application	Open any app by name or path

---

Model Context Protocol (MCP) Integration — 100% Plug & Play

UST allows you to easily connect thousands of MCP servers with zero configuration using npx. You can generate configuration blocks or directly launch these directly from your agent.

The "Meta" Tool: Search 4,800+ MCP Servers

{ "command": "npx", "args": ["-y", "a2asearch-mcp"] }

Allows the agent to search and find its own tools.

The "Mega-Pack" (62 tools in one)

{ "command": "npx", "args": ["skills", "add", "blink-new/blink-plugin"] }

Massive infrastructure pack including database, hosting, storage, and authentication tools.

Web Fundamentals

• Web Fetch: { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-fetch"] }
• Live Web Search (Brave): { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-brave-search"] }
• Browser Automation (Puppeteer): { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-puppeteer"] }

Local File & Code Management

• Filesystem: { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/your/path"] }
• GitHub: { "command": "npx", "args": ["-y", "@github/mcp-server"] }

Advanced Reasoning & Logic

• Sequential Thinking: { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"] } (Forces the AI to create a "thinking space" to test hypotheses and self-correct)

Audio & Media

• Whisper Transcription: { "command": "npx", "args": ["-y", "mcp-whisper-server"] }
• Microphone (Direct Input): { "command": "npx", "args": ["-y", "mcp-microphone"] }
• Audio Recorder (Stream): { "command": "npx", "args": ["-y", "mcp-audio-recorder"] }
• FFmpeg (Media Processing): { "command": "npx", "args": ["-y", "mcp-ffmpeg-server"] }

Infrastructure & Data Science

• Docker Control: { "command": "npx", "args": ["-y", "mcp-server-docker"] }
• SQLite Database: { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-sqlite", "--db-path", "database.db"] }

Simply paste any of these block configurations into your mcpServers setting (e.g. claude_desktop_config.json, Cursor, etc) or use the UST MCP skills to interact with them programmatically.

---

Adapters — Supported LLM Providers

UST supports 4 native adapters out of the box so you can use absolutely any AI model:

1. USTAdapter (Default OpenAI format) — Works with OpenRouter, OpenAI, LM Studio, Ollama, etc.
2. GeminiAdapter — Google's modern Gemini SDK (@google/genai equivalent in Python).
3. LiteLLMAdapter — Over 100+ LLMs (Anthropic, Bedrock, vLLM, etc.) via litellm.
4. OllamaAdapter — Native local adapter for ollama python package.

1. Default OpenAI-compatible Adapter

# OpenRouter / OpenAI / Local LM Studio
from ust import enable_branch, USTAdapter
enable_branch("system")

# OpenRouter (default)
agent = USTAdapter(api_key="sk-or-...")

# Local LM Studio
agent = USTAdapter(
    api_key="local",
    base_url="http://localhost:1234/v1",
    model="llama-3.2-3b-instruct"
)

2. Gemini Adapter

from ust import enable_branch
from ust.core.gemini_adapter import GeminiAdapter
enable_branch("system")

agent = GeminiAdapter(model="gemini-2.5-flash", api_key="AIzaSy...")
reply = agent.chat_sync("What is my IP address?")

3. LiteLLM Adapter (Anthropic, Bedrock, etc)

from ust import enable_branch
from ust.core.litellm_adapter import LiteLLMAdapter
enable_branch("system")

# Using Anthropic Claude 3.5 Sonnet
agent = LiteLLMAdapter(model="claude-3-5-sonnet-20240620", api_key="sk-ant-...")

4. Native Ollama Adapter

from ust import enable_branch
from ust.core.ollama_adapter import OllamaAdapter
enable_branch("system")

# Requires tool-capable model
agent = OllamaAdapter(model="llama3.1")
reply = agent.chat_sync("Check my battery percentage")

---

Advanced Usage

from ust import enable_branch, enable_all, status, USTAdapter

# Load only what you need
enable_branch("system")
enable_branch("web")

# Inspect loaded skills
status()

# Multi-turn conversation
history = []
for user_msg in ["Open Chrome", "Go to github.com", "Take a screenshot"]:
    reply = await agent.chat(user_msg, history=history)
    history += [{"role":"user","content":user_msg},
                {"role":"assistant","content":reply}]

# Restrict tools to one branch
reply = await agent.chat("Search for Python news", branch="web")

---

Project Structure

universal-skill-tree/
├── ust/
│   ├── __init__.py          ← enable_branch(), USTAdapter
│   ├── core/
│   │   ├── registry.py      ← Central skill registry
│   │   ├── executor.py      ← Tool call dispatcher
│   │   └── adapter.py       ← OpenAI-compatible LLM adapter
│   └── skills/
│       ├── system/          ← PC control (10 skills)
│       ├── files/           ← File management
│       ├── web/             ← Web search & scraping
│       ├── media/           ← Audio/video control
│       ├── apps/            ← App launcher
│       ├── vision/          ← Screenshot + vision
│       ├── osint/           ← Sherlock, Osintgram
│       ├── cyber/           ← Sniffnet, Trippy, Scapy
│       ├── crypto/          ← Crypto tools
│       ├── selfhosted/      ← Awesome-selfhosted search & Docker tools
│       ├── aitools/         ← Awesome-ai-tools search
│       ├── agent_skills/    ← Awesome-agent-skills search
│       ├── mcp/             ← MCP servers search & Chrome DevTools
│       ├── browser/         ← Playwright browser
│       └── ai/              ← CrewAI and LiteLLM wrappers
├── examples/
│   └── openrouter_agent.py  ← Demo in 3 lines
└── pyproject.toml

---

License

MIT — free to use, modify, and distribute.