niaka3dayo/agent-skills-vrc-udon

Skills, rules, and validation hooks that teach AI coding agents to generate correct UdonSharp code

What it solves

AI coding agents often struggle with VRChat's UdonSharp (C# to Udon Assembly) because it has strict compile constraints that differ from standard C#. This project provides a knowledge base of skills, rules, and validation hooks to prevent AI agents from using incompatible features like List<T>, async/await, or incorrect networking patterns, ensuring the generated code is actually compilable and functional within VRChat worlds.

How it works

The project functions as a set of "skills" that can be integrated into AI agents (such as Claude Code) via npm or CLI tools. It consists of:

  • Rules: Markdown files that guide the AI before it writes code, detailing blocked C# features and correct networking patterns (e.g., using DataList instead of List<T>).
  • Validation Hooks: Bash and PowerShell scripts that automatically run after the AI edits a .cs file to detect errors like blocked runtime features or sync bloat.
  • Templates: A library of 17 code templates for common interactions and sync patterns.
  • References: Detailed documentation mapping SDK version features to AI-friendly instructions.

Who it’s for

VRChat world creators who use AI coding agents to help write UdonSharp scripts and want to avoid the tedious process of manually correcting AI-generated compile errors.

Highlights

  • Automated Validation: Includes hooks to catch async/await, try/catch, and LINQ in runtime code.
  • Networking Guidance: Provides a sync decision tree to minimize data budget and correct ownership models.
  • SDK Versioning: Maps features across SDK versions up to 3.10.5.
  • Agent Integration: Distributed as an npm package for easy installation into AI agent environments.

Related

  • Project
  • Project
  • Project
  • Project