Claude Bootstrap

An opinionated project initialization system for Claude Code. TDD-first, iterative loops, security-first, AI-native.

The bottleneck has moved from code generation to code comprehension. AI can generate infinite code, but humans still need to review, understand, and maintain it. Claude Bootstrap provides guardrails that keep AI-generated code simple, secure, and verifiable.

Core Philosophy

┌────────────────────────────────────────────────────────────────┐
│  ITERATIVE LOOPS BY DEFAULT                                    │
│  ─────────────────────────────────────────────────────────────│
│  Every task runs in a self-referential loop until tests pass.  │
│  Claude iterates autonomously. You describe what, not how.     │
│  Powered by Ralph Wiggum - iteration > perfection.             │
├────────────────────────────────────────────────────────────────┤
│  TESTS FIRST, ALWAYS                                           │
│  ─────────────────────────────────────────────────────────────│
│  Features: Write tests → Watch them fail → Implement → Pass    │
│  Bugs: Find test gap → Write failing test → Fix → Pass         │
│  No code ships without a test that failed first.               │
├────────────────────────────────────────────────────────────────┤
│  SIMPLICITY IS NON-NEGOTIABLE                                  │
│  ─────────────────────────────────────────────────────────────│
│  20 lines per function │ 200 lines per file │ 3 params max     │
│  If you can't understand the whole system in one session,      │
│  it's too complex.                                             │
├────────────────────────────────────────────────────────────────┤
│  SECURITY BY DEFAULT                                           │
│  ─────────────────────────────────────────────────────────────│
│  No secrets in code │ No secrets in client env vars            │
│  Dependency scanning │ Pre-commit hooks │ CI enforcement       │
├────────────────────────────────────────────────────────────────┤
│  CODE REVIEWS ARE MANDATORY                                    │
│  ─────────────────────────────────────────────────────────────│
│  Every commit requires /code-review before push.               │
│  🔴 Critical + 🟠 High = blocked │ 🟡 Medium + 🟢 Low = can ship │
│  AI catches what humans miss. Humans catch what AI misses.     │
└────────────────────────────────────────────────────────────────┘

Why This Exists

After hundreds of AI-assisted projects across Node, React, Python, and React Native, patterns emerged:

1. Engineers struggle with Claude Code not because of the tool, but because of how they instruct it - The delta is in the guardrails
2. Complexity has a ceiling - There's a point where AI loses coherent understanding of the system. That's a signal, not a failure
3. Restart is a feature, not failure - When fixing something increases complexity, restart with learnings. Each iteration is faster

This toolkit encodes those learnings into reusable skills.

Quick Start

# Clone and install
git clone https://github.com/alinaqi/claude-bootstrap.git ~/.claude-bootstrap
cd ~/.claude-bootstrap && ./install.sh

# In any project directory
claude
> /initialize-project

Claude will:

1. Validate tools - Check gh, vercel, supabase CLIs
2. Ask questions - Language, framework, AI-first?, database
3. Set up repository - Create or connect GitHub repo
4. Create structure - Skills, security, CI/CD, specs, todos
5. Prompt for specs - Transition to defining first feature

Automatic Iterative Loops (Ralph Wiggum)

You talk naturally. Claude automatically runs iterative TDD loops.

┌─────────────────────────────────────────────────────────────┐
│  You say: "Add email validation to signup"                  │
├─────────────────────────────────────────────────────────────┤
│  Claude automatically:                                       │
│  1. Extracts requirements from your request                 │
│  2. Structures as TDD loop with completion criteria         │
│  3. Runs /ralph-loop with tests as exit condition           │
│  4. Iterates until all tests pass + lint clean              │
└─────────────────────────────────────────────────────────────┘

No need to manually invoke /ralph-loop. Just describe what you want:

Opt-out phrases (for when you don't want loops):

• "Just explain..." → explanation only
• "Quick fix..." → one-liner, no loop
• "Don't loop..." → explicit opt-out

Setup Ralph Wiggum Plugin

# Install from official marketplace (in Claude Code)
/plugin install ralph-loop@claude-plugins-official

**Trouble

...