ChocoVine

Stop prompting. Start building.

A workflow wrapper for Claude Code that forces AI to think before it types.

Getting Started • The Workflow • Commands

⚡️ The Problem

We've all been there: You ask an AI to build a feature. It blindly writes 200 lines of code. You run it. It crashes. You spend the next hour prompt-battling to fix it.

🌲 The Solution

ChocoVine turns "Vibes Coding" into Engineering. It stops hallucinations by enforcing a strict Plan → Test → Build loop.

• 🛡️ TDD Guardrails: The AI writes a failing test first. It cannot mark a task complete until that test passes.
• 🧠 Infinite Context: Uses sub-agents to read your entire codebase without clogging up the chat memory.
• 🔎 You are the Architect: You approve the Plan before a single line of code is written.
• 📈 Scales with You: Optimized for mid-to-large codebases where "context limits" usually break other tools.

🚀 Getting Started

You need Claude Code installed.

1. Installation

Option A: npx (Recommended)

# Navigate to your project
cd your-project

# Install ChocoVine directly from GitHub
npx github:vneseyoungster/ChocoVine init

Option B: Clone the Repo

# Clone and copy manually
git clone https://github.com/vneseyoungster/ChocoVine.git
cp -r ChocoVine/.claude your-project/

2. Configuration

Edit the generated CLAUDE.md with your project details:

# Tell ChocoVine your stack (React, Python, etc.)
nano CLAUDE.md

3. Initialization

Tell ChocoVine to scan your project so it understands your architecture patterns.

# /cv:init auto-detects: new project wizard OR existing project scan
/cv:init

🎮 How to Use

The Core Workflow

For most tasks, follow this flow. You control each phase with explicit approval gates.

/cv:research Add a login page with Google OAuth
/cv:plan
/cv:build
npm test
/cv:review

What happens at each step?

1. Research: Claude maps your codebase, identifies patterns, asks clarifying questions.
2. Plan: Presents architecture and tasks. You approve before any code is written.
3. Build: Implements the approved plan with inline validation.
4. Test: You run tests manually to verify.
5. Review: Code quality, security audit, and coverage check.

🧩 The Workflow

How do we guarantee code quality? We strictly follow the Research → Plan → Build → Review loop. This prevents the "spaghetti code" effect common with other AI tools.

graph LR
    A[/cv:research] --> B[/cv:plan]
    B --> C[/cv:build]
    C --> D[test]
    D --> E[/cv:review]

    subgraph Research
    A1[Scan] --> A2[Clarify]
    A2 --> A3[Requirements]
    end

    subgraph Plan
    B1[Architecture] --> B2[Tasks]
    B2 --> B3[Test Specs]
    end

    subgraph Build
    C1[Implement] --> C2[Validate]
    end

    style A3 fill:#f96,stroke:#333,stroke-width:2px
    style C1 fill:#9f6,stroke:#333,stroke-width:2px

📚 Command Reference

Core Commands

| Command | Description | | :--- | :--- | | /cv:research [topic] | Research. Gather context, requirements, resolve conflicts. | | /cv:plan [session] | Plan. Architecture, tasks, test specs (from research). | | /cv:build [session] | Build. Implement tasks, validate, generate test instructions. | | /cv:review [session] | Review. Code quality, security audit, coverage check. | | /cv:fix [error] | Fix. Systematic debugging or quick fixes. | | /cv:refactor [type] | Refactor. Dead code cleanup, rename/move, extract, or general. | | /cv:init | Setup. New project wizard OR scan existing codebase. |

Research Modes

| Command | Description | | :--- | :--- | | /cv:research [topic] | Auto-routes. Detects Figma URLs, docs, or general topics. | | /cv:research --ui [figma-url] | Figma to Code. Extracts tokens, CSS, and layout. | | /cv:research --docs [library] | External Knowledge. Reads library documentation. |

| Old Command | New Equivalent | | :--- | :--- | | /start [task] | /cv:research → /cv:plan → /cv:build → /cv:review | | /init | /cv:init | | /plan | /cv:plan | | /build | /cv:build | | /fix | /cv:fix | | /refactor | /cv:refactor | | /research | /cv:research |

Why the change?

• Namespace isolation: cv: prefix prevents conflicts with other tools
• Explicit workflow: each command has one job with approval gates
• Better visibility: clear separation of concerns

⚙️ Configuration

To get the best results, your CLAUDE.md needs to be accurate. This is the "System Prompt" for your project.

# CLAUDE.md Example
- **Stack:** Next.js 14, Tailwind, Supabase, TypeScript
- **Build Command:** npm run build
- **Test Command:** npm test
- **Conventions:** Use arrow functions, no 'any' types, prefer functional components.

📄 License

MIT ©