C:\BASEMENT\AI_SKILLS.DIR
Lower Level 2.0 continues to serve as a proving ground for my ongoing transition from traditional development to structured, AI-assisted workflows. Each new feature informs the next iteration of my methodology, improving how humans and AI collaborate on complex, interactive VR/XR environments.
Below you will see AI tools and techniques I've learned and applied throughout this project. Each skill includes how I used it and links to relevant devlog entries.
[ Claude Code ] AI coding assistant with direct file access. Used for: automated documentation, code generation, git workflow, and project management. Reduced manual coding time by ~60%.
Autonomous coding agent that reads/writes files, executes terminal commands, and manages git workflow.
Used for: XML docstring generation, UdonSharp validation, automated commits, issue tracking
[ Unity MCP (Model Context Protocol) ] Protocol that lets AI agents communicate with Unity Editor directly. Used to: compile scripts, attach components, enter play mode, and read console errors - all without human intervention.
Bridge between AI agents and Unity Editor for full automation loop.
Used for: Automated UdonSharp compilation, component attachment, play mode testing, console error detection
[ Agent Swarms ] Multiple AI agents working in parallel on different tasks. Used to: simultaneously update documentation, validate code, run tests, and commit changes - reducing iteration time from hours to minutes.
Parallel AI agents handling different aspects of development simultaneously.
Used for: Parallel code review, multi-file refactoring, concurrent testing, distributed documentation
[ Prompt Engineering ] Crafting effective AI prompts to get consistent, accurate results. Key learnings: context windows, few-shot examples, constraint specification, and iterative refinement. Created custom CLAUDE.md with 1000+ lines of project-specific instructions.
Designing prompts and system instructions for reliable AI-assisted development.
Used for: UdonSharp constraint documentation, code review checklists, automated validation rules
[ API Integration (VRChat PlayerData) ] Working within UdonSharp constraints (no generics, no LINQ) to implement cloud persistence. Learned: parallel array patterns, edge case handling, async data caching, and Quest optimization (5-second cache vs per-frame reads).
Cloud persistence API with severe language constraints.
Used for: Achievement tracking, visit persistence, leaderboard data, cross-session state
→ Jul 13: 11-Hour Debug Marathon
→ Oct 20: Full System Working
[ GitHub Actions & Automation ] Automated workflows triggered by git events. Set up: pre-commit UdonSharp validation, automatic issue labeling, changelog generation, and deployment pipelines. Prevents broken commits from reaching repo.
CI/CD pipeline for VRChat world development.
Used for: Pre-commit validation, automated testing, issue management, deployment
[ Real-Time API Integration ] GitHub Pages as free API endpoint for weather data. Learned: CORS handling, JSON parsing in UdonSharp, platform-specific caching (Quest: 10min, PC: 2min), and graceful degradation when API fails.
Live weather data fetched from custom JSON endpoint.
Used for: Dynamic weather display, rain shader triggers, weather-based achievements
C:\BASEMENT\DOCUMENTATION.DIR
Custom AI agent prompts and reference documentation created during development. These "skills" can be loaded into AI coding assistants to provide domain-specific expertise.
[ protv-integration.md ] Custom AI agent prompt for ProTV 3.x integration. Covers event-driven vs polling patterns, IN_/OUT_ variable conventions, sortView shuffle mapping, and critical pitfalls learned from real implementation.
Custom coding agent for future ProTV integrations in VRChat UdonSharp projects.
Used for: ProTV plugin development, event-driven registration, playlist control, shuffle/sortView handling
[ devlog.md ] Interactive devlog creation workflow. Guides AI through asking the right questions ([Milestone], [TIL], [Meta]), then synthesizes responses into structured markdown posts.
AI agent workflow for creating devlog entries through guided reflection interviews.
Used for: Structured devlog creation, capturing learning moments, consistent post formatting
[ bbp-issue-spec.md ] Basement Build Priority system for AI-driven issue prioritization. Calculates BBP = (Agentic_Feasibility × Nostalgia_Score) / Story_Points to identify high-automation, high-impact wins.
SCRUM master agent for analyzing GitHub issues and creating implementation-ready specs.
Used for: Backlog prioritization, BBP scoring, spec generation, agentic feasibility analysis
[ unity-editor-skills.md ] Unity Editor automation agent for solving MCP limitations. Covers SerializedObject patterns for VRCUrl fields, UdonSharp component wiring, and visual verification with ScreenshotHelper when MCP searches fail.
Custom coding agent for Unity Editor scripts that bridge Unity MCP limitations.
Used for: VRCUrl field setting, UdonSharp reference wiring, visual verification, SerializedProperty patterns