Claude Code Starter Kit

23 slash commands. 9 hooks. 10 rules. 2 agents. 2 skills. 12+ project profiles. One template.

The Problem

You've read the guides. You know you need a CLAUDE.md, hooks for secret prevention, database connection pooling, quality gates, testing structure. But every new project starts the same way: 45 minutes of setup before you write a single line of business logic. And if you skip the setup, Claude creates new MongoClient() in six different files, commits your .env, and pushes untested Docker images to production.

What This Is

The Claude Code Starter Kit is a ready-to-use project template based on Claude Code Mastery Guides V3-V5. Clone it, run /setup, and start building. Every best practice is already wired in.

It's not a runnable application — it's the scaffold that makes every application you build with Claude better from the first prompt.

Three Ways to Use It

1. Scaffold — "/new-project my-app clean" creates a new project with full infrastructure
2. Convert — "/convert-project-to-starter-kit ~/existing-project" merges the kit into an existing project without overwriting your code
3. Customize — Clone the repo, modify commands/hooks/skills, use as your own template source

What's Inside

23 Slash Commands

On-demand tools you invoke directly in Claude Code:

Setup & Scaffolding:

Code Quality:

Development:

Infrastructure:

Project Management:

Monitoring:

10 Critical Rules (CLAUDE.md)

Battle-tested rules that prevent the most common Claude Code failures:

• Never publish sensitive data — No secrets in git, Docker, or anywhere online
• TypeScript always — Strict mode, no any. When editing .js files, convert to TypeScript first
• API versioning — Every endpoint uses /api/v1/ prefix. No exceptions
• Database wrapper only — All database access through src/core/db/index.ts. No rogue new MongoClient() anywhere
• Explicit test success criteria — Every E2E test must have URL, element visibility, and content assertions. "Page loads" is not a test
• Never hardcode credentials — All secrets in .env, all configuration via environment variables
• Quality gates — No file > 300 lines. No function > 50 lines. TypeScript strict. All tests pass
• Auto-branching on main — Claude never touches main. Every code-modifying command auto-creates a feature branch
• Parallelize independent awaits — Claude must use Promise.all when multiple await calls don't depend on each other
• Docker push gate — When enabled, blocks docker push until the image is built, run locally, and health-checked

9 Deterministic Hooks

CLAUDE.md rules are suggestions. Hooks are guarantees — they run automatically, every time:

PreToolUse (blocks actions before they happen):

PostToolUse (validates after actions):

Stop (runs when Claude's session ends):

2 Skills (Automatic Template Loading)

Skills trigger automatically based on keywords in your prompts:

• Code Review Skill — Triggered by "review" or "audit". Loads a systematic 7-point checklist covering security, types, error handling, performance, testing, database patterns, and API versioning
• Create Service Skill — Triggered by "create service". Scaffolds microservices with server/handlers/adapters pattern

2 Custom Agents

Specialists that Claude delegates to automatically:

• Code Reviewer — Read-only agent for security audits, type safety, and quality checks. No write permissions — can't accidentally "fix" things during review
• Test Writer — Writes tests with explicit assertions and proper structure. Not just "passes" — real success criteria

Database Wrapper (src/core/db/index.ts)

A production-grade MongoDB wrapper that enforces every database best practice:

import { queryOne, queryMany, insertOne, updateOne, bulkOps } from '@/core/db/index.js';

// Singleton pool — one connection per URI, prevents connection exhaustion
// NoSQL injection sanitization — automatic on ALL inputs
// $limit BEFORE $lookup — enforced automatically
// Graceful shutdown — idempotent, wired to SIGTERM/SIGINT/uncaught errors

• 3 pool presets: high (20 connections), standard (10), low (5)
• Automatic sanitization: Strips $ operators and . path traversal from user input
• Aggregation-only reads, BulkWrite-only writes: Consistent patterns, no mixed approaches
• E11000 concurrent upsert auto-retry: Handles race conditions automatically
• Transaction and change stream support: Built in when you need them
• Next.js hot-reload safe: Connection pool persists via globalThis

12+ Project Profiles

Pre-configured stacks via /new-project:

/new-project my-app go             # Go API with standard layout
/new-project my-app clean          # Just the Claude infrastructure
/new-project my-app sveltekit      # SvelteKit full-stack app

Test Infrastructure

• CHECKLIST.md — Master test status tracker with visual pass/fail indicators
• ISSUES_FOUND.md — User-guided testing log for manual QA
• Playwright E2E Config — Auto-spawns test servers on fixed ports (4000, 4010, 4020)
• Test Query Master (scripts/db-query.ts) — Centralized dev/test database queries, prevents script scatter

Content Builder

Markdown-to-HTML article builder (scripts/build-content.ts) with:

• Config-driven article registry
• SEO optimization (Open Graph, Twitter Cards, Schema.org)
• Syntax highlighting and table of contents generation
• Parent/child article relationships

Documentation Templates

Pre-structured docs that Claude actually follows:

• ARCHITECTURE.md — System overview with authority boundaries and "STOP" scope-creep preventers
• INFRASTRUCTURE.md — Deployment topology, environments, prerequisites
• DECISIONS.md — Architectural Decision Records with rationale

MCP Servers

Pre-configured integrations:

• Context7 — Live documentation lookup for any library
• GitHub — PR management and issue tracking from Claude
• Playwright — E2E test debugging and browser automation
• ClassMCP — Semantic CSS patterns (auto-included in CSS-enabled profiles)

Featured npm Packages

Three open-source packages integrated into the kit:

• ClassMCP — MCP server providing semantic CSS patterns
• Classpresso — Post-build CSS consolidation
• TerseJSON — Memory-efficient JSON with ~70% size reduction

Quick Start

# 1. Clone
git clone https://github.com/TheDecipherist/claude-code-mastery-project-starter-kit my-project
cd my-project && rm -rf .git && git init

# 2. Install global config (one-time, never overwrites existing)
claude /install-global

# 3. Configure your project
claude /setup

# 4. Start building
claude

Already have a project? Convert it without losing anything:

claude "/convert-project-to-starter-kit ~/my-existing-project"

Live AI Monitor

Run /what-is-my-ai-doing in a separate terminal and watch everything Claude does in real-time:

• Every tool call — what files it reads, writes, and executes
• Token usage — input, output, cache hits, running total
• Cost tracking — per-call and cumulative session cost

Zero token overhead. Runs completely outside Claude's context window — it doesn't know it's being watched, and it doesn't cost you anything extra.

The command runs the RuleCatch AI pooler in free monitor mode (--no-api-key) — no account, no subscription, no setup. For the full RuleCatch platform (rule engine, dashboards, violation alerts, MCP server), there's a 7-day free trial with no credit card required.

RuleCatch Integration

The starter kit includes RuleCatch.AI integration because it's part of TheDecipherist's own development workflow. RuleCatch is not required to use the starter kit — everything else works without it. But if you want 200+ pre-built rules, session analytics, violation alerts, and an MCP server so you can ask Claude directly "RuleCatch, what was violated today?" — it's there and ready to go.

The AI pooler's monitor mode is free — no API key, no account required:

npx @rulecatch/ai-pooler monitor --no-api-key

Real-time monitoring of tokens, costs, and tool usage right in your terminal. For the full platform (rule engine, dashboards, MCP server), RuleCatch offers a 7-day free trial with no credit card required.

The check-rulecatch.sh stop hook automatically reports violations at the end of every session.

Links

• GitHub Repository
• Interactive Guide (GitHub Pages)
• Claude Code Mastery Guides V3-V5
• RuleCatch.AI
• TheDecipherist