vault

Claude Tooling

Stop vibe-coding. Ship with structure.

Plugins that make Claude Code follow a real development process.

_{No frameworks. No dependencies. No bloat. Just markdown files that work.}

---

The Problem

You: "Build me an auth system"

Claude: *writes 500 lines*
Claude: "Done! Here's your auth system."

You: *tests it*
You: "This doesn't actually validate tokens."

Claude: "Oh right, let me fix that..."
Claude: *writes 200 more lines that break something else*

LLMs skip steps. They write code that looks right but doesn't work. They forget what they were doing between sessions.

Every AI coding tool adds more features, more complexity, more noise. This goes the other way: simple markdown files that enforce discipline.

---

The Key Insight: Questions

Most AI tools just generate. This one asks.

┌─────────────────────────────────────────────────────────────────────────┐
│                                                                         │
│   TYPICAL AI                         THIS WORKFLOW                      │
│                                                                         │
│   You: "Build auth"                  You: "Build auth"                  │
│        │                                  │                             │
│        ▼                                  ▼                             │
│   AI: *generates 500 lines*          AI: "JWT or session-based?        │
│        │                                   What's the token lifetime?   │
│        ▼                                   Refresh token strategy?"     │
│   You: "This is wrong"                    │                             │
│        │                                  ▼                             │
│        ▼                             You: *answers*                     │
│   AI: *generates 300 more*                │                             │
│        │                                  ▼                             │
│        ▼                             AI: *builds exactly that*          │
│   You: "Still wrong"                      │                             │
│        │                                  ▼                             │
│        ▼                             AI: "I built X, Y, Z.              │
│   *repeat until frustrated*               Is that what you meant?"      │
│                                           │                             │
│                                           ▼                             │
│                                      You: "Yes" or "No, actually..."   │
│                                           │                             │
│                                           ▼                             │
│                                      *catches drift before it compounds*│
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘

Questions appear at every phase:

• Planning: "Which approach? What are the tradeoffs you care about?"
• Alignment: "This function does X—is that right?"
• Testing: "What SHOULD happen in this edge case?"
• Review: "Does this architecture match your vision?"

LLMs are confident. Confident and wrong. Questions force verification before the wrong code compounds into a mess.

---

The Solution

┌────────────────────────────────────────────────────────────────────┐
│                                                                    │
│    ┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐      │
│    │  PLAN   │───▶│  BUILD  │───▶│  VERIFY │───▶│  AUDIT  │      │
│    └─────────┘    └─────────┘    └─────────┘    └─────────┘      │
│         │                                             │           │
│         │              ┌─────────┐                    │           │
│         └─────────────▶│   FIX   │◀───────────────────┘           │
│                        └─────────┘                                │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

Plan before you code. Track deviations. Audit for fake code. Fix systematically.

---

Install

git clone https://github.com/jsai/claude-tooling.git
cd claude-tooling && ./install.sh

---