Appearance
Hands-On Tutorial
The best way to understand Responsible Vibe is to experience it. We'll build something simple, enhance it, then fix a bug. Three different workflows, three different approaches.
Prerequisites
Make sure you've got Responsible Vibe configured with your AI agent. If not, check the Quick Setup first.
Part 1: Greenfield Development
Let's build something from scratch. This triggers the Greenfield workflow – the full engineering process for new projects.
The Challenge
Ask your AI: "Build a simple dice game. Two players take turns rolling dice, highest total wins."Hint: In order to not get just another Next.js app, you may instruct it to build for the terminal in a future stage
What You'll Experience
Your AI won't immediately start coding. Instead, it'll:
- Ask clarifying questions: "How many dice per roll? Best of how many rounds? Should we track scores across games?"
- Design the architecture: "I'm thinking a Game class, Player class, and a simple CLI interface. Does that sound right?"
- Plan the implementation: "Let me break this into phases: core game logic, player management, CLI interface, then testing."
- Build systematically: Following the plan it just created, not jumping around randomly.
Part 2: Feature Enhancement
Now let's add something that wasn't in the original scope. This triggers the EPCC workflow (Explore → Plan → Code → Commit).
The Challenge
Ask your AI: "Add a high-score tracking system.
Players should see their win/loss record across games."What You'll Experience
- Explore: "Let me understand the current code structure and see how to integrate scoring..."
- Plan: "I'll need to modify the Player class, add persistent storage, and update the CLI to show stats."
- Code: Focused implementation of just the scoring feature
- Commit: Clean up and finalize the enhancement
Part 3: Bug Fixing
Time to break something and fix it. This triggers the Bugfix workflow.
The Challenge
Ask your AI: "There's a bug in the dice game.
Players are complaining the dice rolls seem unfair."What You'll Experience
- Reproduce: "Let me run the game and see if I can reproduce the issue..."
- Analyze: "I found the problem – the dice rolling logic is hardcoded to return 1."
- Fix: Targeted fix for just the bug, not a rewrite
- Verify: "Let me test this fix to make sure it works correctly..."
Next Steps
- Automatic Workflow Selection – How the AI picks the right methodology
- Advanced Engineering – Project docs, variables, and branch management
- Interactive Workflows – Explore all available methodologies
Try this tutorial with your own project ideas. The workflows adapt to any domain – web apps, CLI tools, games, whatever you're building.