AgentSkills.NET

A .NET implementation of the Agent Skills open standard by Anthropic.

Overview

AgentSkills.NET enables .NET-based AI agents to discover, load, and execute skills defined in a standardized format. It provides:

• Skill Discovery: Scan directories for skill definitions
• YAML + Markdown Parsing: Parse SKILL.md files with frontmatter and instructions
• Validation: Comprehensive validation with diagnostics (no exceptions for normal errors)
• Progressive Disclosure: List skills first, load full content only when activated
• Host-Agnostic Design: Core libraries work with any agent framework
• Safety-First: Scripts are treated as data; no execution by default

Quick Start

Installation

# Coming soon - NuGet packages will be available after initial release dotnet add package AgentSkills

Basic Usage

// Coming soon - walking skeleton sample // scan → metadata → validate → render list → activate skill → render instructions

Project Status

🚧 Early Development - This project is in active development. APIs may change.

Documentation

• Project Brief - Architecture and design principles
• Contributing Guide - How to contribute
• Agent Guidelines - Guidelines for AI coding agents
• Architecture Decisions - ADRs documenting key decisions

Repository Structure

├── src/ # Core library source code │ ├── AgentSkills/ # Core domain models │ ├── AgentSkills.Loader/ # Skill loading │ ├── AgentSkills.Validation/ # Validation rules │ └── AgentSkills.Prompts/ # Prompt rendering ├── tests/ # Test projects ├── samples/ # Sample applications ├── docs/ # Documentation ├── fixtures/ # Test fixtures and samples └── .github/workflows/ # CI/CD automation

Building

# Restore dependencies dotnet restore # Build all projects dotnet build # Run tests dotnet test

Contributing

Contributions are welcome! Please read our Contributing Guide before submitting PRs.

Key points:

• Follow the conventions in AGENTS.md
• Keep changes small and focused
• Add tests for new functionality
• Update documentation as needed

Design Principles

1. Safety First: No script execution by default
2. Progressive Disclosure: Load metadata fast, full content on demand
3. Host Agnostic: Core libraries don't depend on specific frameworks
4. Diagnostics over Exceptions: Return diagnostics for validation errors
5. Explicit APIs: Boring, clear, discoverable interfaces

See docs/project_brief.md for details.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Agent Skills Standard by Anthropic
• The .NET and AI communities