vibe-workflow von KhazP/vibe-coding-prompt-template

A practical AI workflow for shipping MVPs

Turn an idea into an MVP with structured prompts, agent docs, and AI-assisted coding workflows.

Used on projects like vibeworkflow.app, moneyvisualiser.com, caglacabaoglu.com, and RealDex App.

Built with this workflow
Workflow overview
Quick start and the 5 steps
Prerequisites and tools
Advanced agent practices
Project structure and deployment
Common pitfalls and troubleshooting
Further reading

Built with this workflow

This repo documents the workflow behind a handful of shipped projects. The goal is simple: do the thinking upfront, hand clean context to your tools, and keep the build phase moving.

Project	What it is
vibeworkflow.app	An interactive web app built around the same structured vibe-coding workflow documented here.
moneyvisualiser.com	A money visualization website that visualized money in a 3D environment.
caglacabaoglu.com	A production portfolio and gallery site built with the same PRD-to-agent execution approach.
alpyalay.org/realdex	A mobile app built on React Native that lets you catch animals, and put them in a Pokemon-like collection.

_{Maintained by Alp Yalay.}

Workflow overview

The process moves through five stages, from idea validation to working code:

flowchart LR
    subgraph Phase1[" "]
        A[💡 Idea]
    end

    subgraph Phase2["Research"]
        B[📊 Market Analysis]
    end

    subgraph Phase3["Define"]
        C[📋 PRD]
    end

    subgraph Phase4["Design"]
        D[🏗️ Tech Design]
    end

    subgraph Phase5["Generate"]
        E[🤖 AGENTS.md]
    end

    subgraph Phase6["Build"]
        F[🚀 MVP]
    end

    A --> B --> C --> D --> E --> F

    style A fill:#667eea,stroke:#667eea,color:#fff
    style B fill:#764ba2,stroke:#764ba2,color:#fff
    style C fill:#f093fb,stroke:#f093fb,color:#fff
    style D fill:#4facfe,stroke:#4facfe,color:#fff
    style E fill:#00f2fe,stroke:#00f2fe,color:#000
    style F fill:#43e97b,stroke:#43e97b,color:#000

Quick start and the 5 steps

TL;DR: run research, turn it into a PRD, choose the stack, generate your agent files, then build in small passes.

Phase 1: thinking through the product

Do the first three steps in ChatGPT, Claude.ai, Gemini, or any other chat tool. You do not need a repo yet.

Deep Research

Check whether the idea is worth building - 20-30 min

This step gives you a quick read on demand, competitors, and whether the scope looks realistic.

Open part1-deepresearch.md and copy all of its contents.
Paste it into your preferred AI platform Chat (like Claude.ai, ChatGPT, or Gemini) and press Enter.
The AI will ask you a few questions about your idea. Answer them truthfully in the chat.
The AI will generate a comprehensive research document based on your answers.
Save the output into a local file named research-[YourAppName].md (or .txt) or simply keep this chat open for Step 2.

Tip: if your chat tool supports web search, turn it on so the stats and competitor references are current.

Product Requirements (PRD)

Write down what the MVP actually needs to do - 15-20 min

This turns the rough idea into a scope you can build against.

Copy the contents of part2-prd-mvp.md.
Option A (Same Chat): If you kept your chat open, paste the prompt right below the Deep Research output.
Option B (New Chat): Start a fresh chat, paste your saved research-[YourAppName].md content, and then paste the Part 2 prompt below it.
Press Enter, answer any clarifying questions the AI asks, and let it generate your requirements.
Save the final output as PRD-[YourAppName]-MVP.md.

Technical Design

Pick a stack you can actually ship with - 15-20 min

This step helps you choose the stack and decide where to keep things simple.

Copy the contents of part3-tech-design-mvp.md.
Paste it into your ongoing conversation (or into a new one, making sure to attach the PRD-[YourAppName]-MVP.md from Step 2 as context).
The AI will ask questions regarding your budget, timeline, and complexity tolerance.
Discuss the trade-offs it presents (e.g., full-code vs. no-code builder).
Once a stack is decided, save the output as TechDesign-[YourAppName]-MVP.md.

Phase 2: execution in your IDE

Move into Cursor, VS Code with Copilot, Claude Code, or your preferred coding setup. This is where the project becomes code.

Set up the agent files

Create the docs and instructions your coding agent will rely on - 1-2 min

This step fills out AGENTS.md and the supporting docs from your PRD and tech design.

Click "Use this template" in GitHub (or clone this repository locally).
Open this cloned repository folder in your AI IDE (like Cursor or VS Code).
Create a docs/ folder in your project root if it does not already exist.
Move your saved documents into docs/ using these names:
- docs/PRD-[YourAppName]-MVP.md
- docs/TechDesign-[YourAppName]-MVP.md
- optional: docs/research-[YourAppName].md (or .txt for backward compatibility)
Open the AI Chat inside your IDE, type: "Read part4-notes-for-agent.md, follow its instructions, and set up my workspace."
The agent should copy the boilerplates from /templates/ into your project root and fill in the placeholders using the files in docs/.

Build with AI Agent

Build the MVP in small, reviewable chunks - 1-3 hrs

Choose your development environment and start iterating:

Ensure your newly generated AGENTS.md and configuration files are physically in the project folder.
Give your agent its first command:

"Read AGENTS.md, propose a Phase 1 plan, wait for my approval, and then build it step by step."
Treat the agent like a junior developer. Ask it to stop after each major feature, explain the diff, and run tests where possible.
Repeat the loop until your MVP is complete:

Recommended Loop:

╭──────────────╮      ╭──────────────╮      ╭──────────────╮
│   📝 Plan    │ ───>│  ⚡ Execute │ ───>│  🔍 Verify  │
│  (Approve)   │      │  (One Feat)  │      │    (Test)    │
╰──────────────╯      ╰──────────────╯      ╰──────────────╯
       ▲                                           │
       └───────────────────────────────────────────┘

Prerequisites and tools

You need a modern browser, a few hours, and enough comfort with files and copy-paste to move between tools. You do not need to be an experienced developer.

Platform selection guide

Focus Area	Recommended Tools
Fast Prototype (Full-stack)	(Includes Agent/Plan mode, DB, Auth)
Production-Ready Frontend	(Vercel-native, exact Next.js/React components)
Learning / Sandbox Coding	(Dynamic Context) or VS Code with Copilot
Complex Logic / Multi-Agent	(Agent Teams) or GitHub Copilot CLI
Budget-Limited	(Free) + VS Code

Note: I would not use this workflow as-is for native hardware work, heavily regulated products, or safety-critical systems.

Advanced agent practices

1. Artifact-first memory and compaction

To avoid context overload, let the agent write things down instead of trying to keep everything in one giant chat:

Compaction and handoffs: Use native compaction (/compact in Copilot CLI, Claude Code logic) instead of hard resets. When you switch sessions, have the agent write a 001-spec.md or recap.md and load only that file into the new chat.
Dynamic context (Cursor): Let the agent save findings into real files instead of burying them in chat history.
If you must restart, attach AGENTS.md, docs/PRD-[YourAppName]-MVP.md, and your latest handoff artifact.

2. Multi-agent orchestration and plugins

Agent teams: Tools like Claude Code support multiple agents working in parallel. Treat them like assigned roles, not magic.
Plan before edit: Ask for an approved plan from the lead agent before the execute agent starts changing files. It cuts down on silent regressions.
Keep AGENTS.md as the source of truth, then add tool-specific plugins or .cursor/rules/ to seamlessly extend capabilities.

3. Model strategy matrix

Use model families instead of pinned version names. It ages better as models get swapped underneath you.

Strategy	Primary Families	Best For	Speed
Speed-first	Gemini Flash, Claude Sonnet	Fast prototyping, broad iteration	High
Balanced	Claude Sonnet, Gemini Pro	Daily coding, debugging, planning	Med-High
Depth-first	Claude Opus, Gemini Pro	Deep reasoning, complex refactors	Medium

4. Agent observability

When an agent ignores instructions or behaves inconsistently:

Check which instructions/rules/hooks were loaded.
Confirm tool permissions and blocked actions.
Verify the active session context was not reset.
Re-run with explicit instruction order: "Read AGENTS.md, then agent_docs/, then execute."

Project structure and deployment

Recommended project skeleton

your-app/
├── 📁 docs/
│   ├── research-YourApp.md
│   ├── PRD-YourApp-MVP.md
│   └── TechDesign-YourApp-MVP.md
├── 📁 agent_docs/
│   ├── tech_stack.md
│   ├── code_patterns.md
│   ├── project_brief.md
│   ├── product_requirements.md
│   └── testing.md
├── 📄 AGENTS.md                  # Universal AI instructions (The Master Contract)
├── 📄 MEMORY.md                  # Artifact-first memory for session continuity
├── 📁 specs/                     # Agent handoff artifacts (e.g. 001-feature-spec.md)
├── 📁 .cursor/rules/             # Cursor rules (preferred)
└── 📁 src/                       # Your application code

Deployment and security

Once the MVP works, do a final pass on secrets, auth, and basic abuse protections before you deploy:

Security Pass: Check dependencies, secrets, auth paths, and rate limits.
Push & Deploy:
- For Next.js, React, frontend apps.
- For Static sites, edge functions.

Common pitfalls and troubleshooting

Avoid these mistakes

Pitfall	Solution
Skipping discovery work	Run the Part 1 research prompt first
Letting agents ship code alone	Review the diff and run tests before merging
Publishing auto-generated UIs	Test accessibility and security before launch
Forcing one tool to do everything	Mix tools, IDE + terminal + builder usually works better

Agent troubleshooting

Problem	Solution
"AI ignores my docs"	Say: "First read AGENTS.md, PRD, and TechDesign. Summarize key requirements before coding."
"Code doesn't match PRD"	Say: "Re-read the PRD section on [feature], list acceptance criteria, then refactor."
"AI is overcomplicating"	Add to config: "Prioritize MVP scope. Offer the simplest working implementation."
"Deployment failing"	Request: "Walk through deployment checklist, verify env vars, then run health check."

Download and releases

Use this repository directly via Use this template.
Stable snapshots are listed on GitHub Releases.

Communication channels

Discussions: GitHub Discussions
Bug reports: GitHub Issues
Security reports: Private security advisory form

Community and policies

This project is free and open source under MIT. Contributions are welcome in all forms (docs, examples, fixes, and suggestions).

Contribution guide: .github/CONTRIBUTING.md
Code of conduct: .github/CODE_OF_CONDUCT.md
Security policy: .github/SECURITY.md
Governance: .github/GOVERNANCE.md
Support: .github/SUPPORT.md
FAQ: .github/FAQ.md
Checklist: .github/CHECKLIST.md

Monthly update cadence

This template is maintained monthly. Review tool deprecations, refresh model-family references, and update agent capability notes when the ecosystem shifts.

Contributing

PRs and issues are welcome. If you adapt this workflow, add a new tool setup, or ship something interesting with it, that is useful context for everyone else too. For community Q&A and roadmap ideas, use Discussions. See .github/CONTRIBUTING.md for contribution guidelines.

License

Released under the MIT License.

If this workflow helps you ship something real, open an issue or PR and show what changed.

_{Created by @alpyalay and improved through community contributions.}

vibe-workflow