⚠️ SkillsManager is Deprecated

Deprecation Notice (July 2025):

• The SkillsManager project is now officially deprecated and will receive no further updates.
• Please use HoloLink for all new development. HoloLink is a direct evolution of SkillsManager, offering the same powerful features plus new improvements and ongoing support.
• For file or skill syncing, use HoloSync.

Package	Purpose
HoloLink	Advanced LLM skill/action management
HoloSync	Modern skill/file syncing utility
SkillsManager	(Deprecated) – Use HoloLink instead

HoloLink is fully compatible with SkillsManager. Upgrade today for the latest features and improvements.

SkillsManager – Natural Language Skills for LLMs (No Vendor Lock-In, No Naming Restrictions)

Overview

SkillsManager is a next-generation framework for managing and executing skills/actions with large language models (LLMs) using natural language—not vendor-bound JSON or TYPED schemas definitions. Unlike traditional function-calling, SkillsManager lets your LLM output actions in freeform(Natural Language) text. The SkillsManager framework interprets, maps, and executes those actions using your own Python code—no matter what you call your skills/actions or where you store them.

Highlights:

• Zero vendor lock-in: Not tied to any model provider’s format or schema.
• No naming restrictions: Name your skills, folders, or groups anything you like—SkillsManager doesn’t care.
• No JSON schemas: No decorators, no rigid typing, no endless schema maintenance.
• Natural language as the interface: The LLM outputs actions in text, not function_call JSON or TYPED.
• Add/organize skills freely: Just point SkillsManager at your code—group, reload, or restrict however you want.
• Centralized execution: All skill mapping, argument parsing, and execution happens in one place.

Backward compatible:

• Want to stick with traditional function calling? No problem. SkillsManager can still work with JSON and TYPED schemas definitions if you prefer that style.
• Want to use SkillsManager with existing function calls? You can still use it alongside traditional function calling, so you can migrate gradually or keep both styles.

NOTE: If you want to use JSON schemas or TYPED definitions (like OpenAI, Gemini, Anthropic, etc.), SkillsManager automatically generates and manages all required schemas for you based on your code and docstrings— You NEVER have to write, maintain, or register schemas manually.

Why SkillsManager?

Traditional LLM “function calling” means:

• Rigid function and argument definitions.
• Tedious JSON and TYPED schema management.
• Forced adherence to provider conventions.
• Locked into one vendor’s API.

SkillsManager:

• Uses only natural language—models describe actions however they want.
• Lets you migrate skills and switch model providers instantly.
• Removes all JSON and TYPED schema headaches.
• Lets you organize, group, and name skills however you want.

Key Features

• Flexible Skill Loading: Add, reload, or restrict any set of skills, with any folder or group name.

• Natural Language Action Parsing: Models output plain English (or any language), and SkillsManager interprets it to execute your code.

• Centralized Execution: No scattered registries—skill mapping, argument parsing, and execution are all in one place.

Organizing and Naming Your Skills

There are no reserved names, no required folder layout, and no fixed naming conventions in SkillsManager. You control all naming, grouping, and structure—use whatever makes sense for your team, project, or workflow.

Examples:

• Refer to SMExamples Directory or the Examples on GitHub:

Just point SkillsManager at any directory or skill group you want—no restrictions, ever.

Example Layouts (ALL are valid)

project_root/
├── Skills/ # This is just an example, you can name it whatever you want, you can even nest the directories it does not have to be at the root
│   ├── Foo/ # Can do nested folders, any names
│   │   ├── summarize.py
│   │   └── summarize_pdf.py
│   ├── Bar/
│   │   ├── sendEmail.py
│   │   └── joke_skill.py
│   ├── EmailTool.py # Or directly in the directory
│   └── ping_tool.py
├── Tools/ # This is just an example, you can name it whatever you want, you can even nest the directories it does not have to be at the root
│   ├── Foo/ # Can do nested folders, any names
│   │   ├── summarize.py
│   │   └── summarize_pdf.py
│   ├── Bar/
│   │   ├── sendEmail.py
│   │   └── joke_skill.py
│   ├── EmailTool.py # Or directly in the directory
│   └── ping_tool.py
├── SkillLoader.py # Whatever you want to call it
├── .env
└── ...

Or any structure you prefer.

How It Works

1. Model outputs natural language: "Summarize the latest report and update our dashboard."
2. SkillsManager parses the text and finds the right skills (whatever you’ve named them).
3. Skills are executed and results returned—no schemas, no JSON or TYPED schemas, just Python code.

Example: Comparing Approaches

1. With SkillsManager (No Naming Restrictions)

Behind the scenes:

• Model outputs: “Summarize PDF, then send summary via email.”
• SkillsManager parses and executes your skills, regardless of how you’ve named or grouped them.
• No decorators, no JSON no TYPED schemas.

2. Traditional Function/Tool Calling (Provider Schema Required)

• Must define every function and argument as a JSON or TYPED schema.
• Register all with the provider.
• Model outputs strict function_call blobs.
• You parse, validate, and execute under the provider’s constraints.

SkillsManager does this part for you automatically if you ever need it—no manual schemas, no boilerplate.

Adding and Organizing Skills

• Drop Python modules into any folder, with any name or grouping.
• Point SkillsManager at those folders in your class (loadComponents).
• No JSON or TYPED, no schemas, no required naming—total freedom.

Why Use SkillsManager?

• Never locked into a provider.
• No JSON or TYPED schema maintenance.
• Flexible skill organization: Name and group skills however you want.
• Production ready: Clean, scalable, and easy to extend.

FAQ

Q: Do I need to follow a specific naming or folder structure? A: No. You can use any names and any folder hierarchy.

Q: Can I use this with any LLM PROVIDER? A: Yes.

Q: How does SkillsManager know what to execute? A: It parses the model’s plain language output and matches actions to your Python skills, regardless of naming.

Q: Do I have to write JSON or TYPED schemas or tool definitions? A: Never. SkillsManager auto-generates and manages them for you when required.

Code Examples

You can find code examples on my GitHub repository.

License

This project is licensed under the Apache License, Version 2.0. Copyright 2025 Tristan McBride Sr.

Acknowledgements

Project by:

• Tristan McBride Sr.
• Sybil