A practical guide and rules template for building, debugging, securing, and shipping production software using AI coding agents (Cursor, Antigravity, Claude Code, Windsurf, Copilot) without losing control of your codebase.
- The Problem: The Infinite Debugging Loop
- Supported AI Coding Agents & IDEs
- Who Is This For?
- The 10 Core Principles of Vibe Coding
- Recommended Workflow
- Repository Structure & Guides
- Quick Start
- AI Search Index (
llms.txt) - Contribution & License
We've all been there. You have a great idea and spin up an AI coding agent:
Idea
↓
Ask AI to build everything
↓
Looks impressive
↓
Add more features
↓
Architecture starts drifting
↓
Something breaks
↓
AI patches it (symptom masking)
↓
Another thing breaks
↓
Infinite debugging loop
↓
Unmaintainable codebase
This repository teaches a better workflow. Vibe coding isn't about letting AI build everything for you. It's about learning how to direct, constrain, inspect, test, and improve AI-generated software.
This playbook and its .agents / system prompt templates are compatible with:
- Cursor IDE (via
.cursorrulesor.clinerules) - Google Antigravity / Gemini CLI (via
AGENTS.mdand.agents/rules/) - Claude Code (via
CLAUDE.mdand project context) - Windsurf Cascade (via
.windsurfrules) - GitHub Copilot Workspace & Chat
- ChatGPT / Claude Web Interfaces
Primary Audience: Students, junior developers, indie hackers, and self-taught developers building their first serious SaaS or web application with AI. If you understand basic programming but struggle with architecture, debugging, or production-readiness, this playbook is for you.
Who it's NOT for: This is not a repository of "Make me a SaaS" prompts for non-technical founders, nor is it a comprehensive manual for senior staff engineers.
- AI is the implementation engine, not the owner of the architecture.
- The developer owns the final result.
- Never blindly trust generated code.
- Build in small, verifiable increments.
- Understand the problem before asking AI to modify code.
- Prefer root-cause fixes over patches.
- Context is part of the prompt.
- Every significant AI-generated change should be verified.
- A working prototype is not automatically production-ready software.
- The goal is not to write less code. The goal is to build better software with AI.
Plan → Context → Implement → Inspect → Test → Verify → Commit
For debugging: Reproduce → Diagnose → Isolate → Fix → Test → Verify
VIBECODING.md- The core conceptual playbook and methodology.AGENTS.md- Instructions to copy into your project to align your AI agent.PROMPTS.md- A prompt library organized by development phase (Planning, Implementation, Review).DEBUGGING.md- The guide to escaping the infinite debugging loop.PRODUCTION-CHECKLIST.md- How to take your local prototype to the real world.skills/- Reusable instructions and personas that AI coding agents can follow.examples/- Good vs. bad workflows and prompts.case-studies/- Real-world examples of this methodology in practice (e.g., PPT Maker).
- Read
VIBECODING.mdto understand the fundamental shift in how you should interact with AI. - Copy
AGENTS.mdinto your project's.agents/or.cursorrulesfolder. - Keep
DEBUGGING.mdopen when you hit your first major error.
This repository includes standardized LLM context files for AI search engines (ChatGPT, Perplexity, Claude, AI Overviews):
llms.txt- Fast API/AI index of all rules, skills, and case studies.llms-full.txt- Complete context compilation for full prompt ingestion.
Read our Contribution Guidelines to submit new skills, debugging patterns, or case studies.
Licensed under the MIT License.
Disclaimer: This is an educational resource. The author is not responsible for any software bugs, security breaches, financial losses, or data loss caused by AI coding agents. Read the full Legal Disclaimer before using these methodologies in production.