The AI Agent Skill
Navigator

What is Skills Tree?

Skills Tree is an open-source engine that answers one question: “What do I need to learn to build a [goal]?”

It combines a hand-curated goal taxonomy, a calibrated skill graph, and a recommendation engine to produce ranked, confidence-scored learning paths for AI agent builders.

Quick Start

From zero to your first recommendation in under 60 seconds.

# 1. Install
pip install skills-tree

# 2. Validate stack
skills-tree validate

# 3. Get recommendations
skills-tree recommend --goal "Coding Agent"

# 4. Generate a blueprint
skills-tree blueprint --goal "Coding Agent"

Or launch a ready-to-use environment instantly:

Install CLI

The skills-tree command exposes all engine capabilities from your terminal.

skills-tree recommend --goal "Coding Agent"
skills-tree recommend --goal "RAG Assistant" --experience intermediate --time-budget 80
skills-tree recommend --goal "Coding Agent" --format table

skills-tree blueprint --goal "Coding Agent"
skills-tree blueprint --goal "RAG Assistant" --format pretty

skills-tree goals
skills-tree goals --format table

skills-tree skills
skills-tree skills --format table

skills-tree validate
skills-tree validate --goal "Coding Agent"

MCP Integration

Skills Tree ships an MCP server exposing 4 tools so Claude Desktop, Cursor, and VS Code can call live Architect recommendations.

{
  "mcpServers": {
    "skills-tree-architect": {
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/skills-tree"
    }
  }
}

{
  "mcpServers": {
    "skills-tree-architect": {
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/skills-tree"
    }
  }
}

{
  "servers": {
    "skills-tree-architect": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "mcp.server"],
      "cwd": "/path/to/skills-tree"
    }
  }
}

See MCP Quickstart for full setup instructions.

API Documentation

The Architect REST API (FastAPI) serves all engine capabilities over HTTP. Swagger UI and ReDoc are built-in.

# Start the API server
uvicorn api.main:app --reload --port 8000

# Swagger UI:  http://localhost:8000/docs
# ReDoc:       http://localhost:8000/redoc
# OpenAPI:     http://localhost:8000/openapi.json

Example Recommendations

Recommendations for Coding Agent (intermediate, no budget):

{
  "goal": "Coding Agent",
  "goal_id": "G01",
  "confidence_score": 0.86,
  "required_skills": [
    { "id": "skill:prompt-engineering",  "name": "Prompt Engineering",  "rank": 1 },
    { "id": "skill:code-generation",      "name": "Code Generation",      "rank": 2 },
    { "id": "skill:function-calling",     "name": "Function Calling",     "rank": 3 },
    { "id": "skill:error-recovery",       "name": "Error Recovery",       "rank": 4 },
    { "id": "skill:context-management",   "name": "Context Management",   "rank": 5 }
  ],
  "calibration_applied": true
}

Example Blueprint

{
  "id": "blueprint-20260615143000",
  "title": "Coding Agent",
  "goal_id": "G01",
  "architecture_type": "Single-Agent",
  "confidence_score": 0.86,
  "required_skills": [...],
  "learning_path": ["Prompt Engineering", "Code Generation", "Function Calling"],
  "risks": ["LLM hallucination in code output", "Tool call error handling"]
}

Architecture Diagram

All four layers share a single execution path — no duplicated logic.

GitHub Repository & Community