Skip to content

Aetheris Skills System

Aetheris v3.0 introduces a Skills system that enables AI agents to interact with files, execute commands, and perform actions.

Overview

Skills are executable capabilities that AI models can invoke through their respective tool/function calling interfaces:

  • Claude: Tool Use format
  • Gemini: Function Calling format
  • OpenAI: Function Calling format (planned)

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    AI Provider                              │
│                (Claude/Gemini/OpenAI)                       │
└─────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                  Provider Adapter                           │
│         (Converts between formats)                          │
└─────────────────────────────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────┐
│                   Skill Registry                            │
│   ┌─────────────┬─────────────┬─────────────────────────┐  │
│   │  Built-in   │   Custom    │   MCP Tools             │  │
│   │  Skills     │   Skills    │   (via Bridge)          │  │
│   └─────────────┴─────────────┴─────────────────────────┘  │
└─────────────────────────────────────────────────────────────┘

Skill Types

1. Built-in Skills

Pre-defined skills that come with Aetheris:

Skill Description Requires Confirmation
read_file Read file contents No
write_file Write/modify files Yes
search_files Search for files No
execute_command Run shell commands Yes

See builtin-skills.md for details.

2. Custom Skills

User-defined skills via YAML configuration:

skills:
  - name: run_tests
    description: Run project tests
    command: pytest {path}
    parameters:
      - name: path
        type: string
        description: Test path
        default: tests/

See custom-skills.md for details.

3. MCP Tools

Skills imported from MCP (Model Context Protocol) servers:

{
  "mcp": {
    "servers": [
      {
        "name": "filesystem",
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-filesystem"]
      }
    ]
  }
}

See ../mcp-integration.md for details.

Safety Features

Sandbox Validation

All file operations are sandboxed to the project directory by default:

# Allowed
skill.execute({"path": "src/main.py"})

# Blocked (outside project)
skill.execute({"path": "/etc/passwd"})

Confirmation Prompts

Dangerous operations require user confirmation:

Skill 'write_file' wants to modify: src/config.py
Allow? [y/N]:

Blocked Commands

Certain shell commands are blocked:

  • rm -rf /
  • sudo rm
  • Format/partition commands
  • Network attacks

Provider Adapters

Adapters convert skills to provider-specific formats:

Claude (Tool Use)

{
  "name": "read_file",
  "description": "Read a file's contents",
  "input_schema": {
    "type": "object",
    "properties": {
      "path": {"type": "string"}
    },
    "required": ["path"]
  }
}

Gemini (Function Calling)

{
  "name": "read_file",
  "description": "Read a file's contents",
  "parameters": {
    "type": "object",
    "properties": {
      "path": {"type": "string"}
    },
    "required": ["path"]
  }
}

Configuration

Configure skills in config.json:

{
  "skills": {
    "enabled": true,
    "custom_skills_path": "./aetheris-skills/",
    "sandbox_enabled": true,
    "confirmation_required": ["write_file", "execute_command"],
    "disabled_skills": []
  }
}

Quick Start

  1. Enable skills in your config
  2. Start interactive mode: aetheris --cli
  3. Ask the AI to perform actions: 
    aetheris[gemini]> Read the README.md file
  4. Confirm any write operations when prompted

Documentation Index