Claudeを10倍強力にする隠しフォルダ：開発者必見

Most developers who use Claude Code interact with it the same way they interact with any other AI tool. They type a prompt. They get a response. They move on. The developers building serious systems with Claude know something different. They know that the real power of Claude Code is not in the prompts you type. It is in the configuration that shapes every single interaction before you say a word. That configuration lives in the .claude/ folder. This folder is the brain behind your Claude Code setup. It holds your custom commands, your project memory, your conversation history, and the instructions that turn a generic AI into a development partner that knows your codebase, your standards, and your preferences. Most developers never touch it. The ones who do build things that feel impossible to everyone else. Here is the complete anatomy. The .claude/ Folder Structure When you initialize Claude Code in a project the .claude/ folder is created automatically. Here is what it looks like: your-project/ ├── .claude/ │ ├── commands/ │ │ ├── deploy.md │ │ ├── review.md │ │ └── test.md │ ├── memory/ │ │ └── project.md │ ├── conversations/ │ │ └── [session-id].json │ └── settings.json ├── CLAUDE.md └── src/ Quick Reference: What Each File Does commands/ → Custom slash commands you define and reuse memory/ → Project context that persists across sessions conversations/→ Session history Claude can reference settings.json → Claude Code configuration and preferences CLAUDE.md → The master instructions file Claude reads first The CLAUDE.md File: The Most Important File in Your Project CLAUDE.md sits at the root of your project and is the first thing Claude Code reads when you start a session. Think of it as the onboarding document you would give to a new senior developer joining your team. Whatever you put in CLAUDE.md becomes standing context for every conversation in your project. Claude will follow these instructions without you having to repeat them. Here is a production ready CLAUDE.md template: # CLAUDE.md — Project Intelligence for Claude Code ## Project Overview This is a Next.js 14 application for [YOUR APP NAME]. Stack: TypeScript, Tailwind, Prisma, PostgreSQL, Vercel. ## Architecture Rules - All API routes live in /app/api/ - Database queries go through /lib/db.ts only - Never write raw SQL — use Prisma client - All components must be in /components/ with TypeScript ## Code Standards - Functional components only — no class components - Always use async/await, never .then() - All functions need JSDoc comments - Error handling required in every API route ## Testing Requirements - Write tests for every new function - Use Vitest for unit tests - Test files live next to the files they test ## What To Avoid - Never use inline styles - Never use any as a TypeScript type - Never commit secrets or API keys - Never modify migration files after they are applied The more specific your CLAUDE.md the more Claude behaves like a team member who has been on your project for months rather than a blank slate AI. The commands/ Folder: Your Custom Slash Commands The commands/ folder is where you define custom slash commands that Claude executes on demand. Each file in this folder becomes a command you can call with /command-name. This is one of the most powerful and most underused features of Claude Code. Instead of typing out complex instructions every time, you define them once and call them with a single word. Example 1: The /review Command Create .claude/commands/review.md to add a /review command: # /review — Comprehensive Code Review When this command is run, perform a complete code review of the files I have open or that I specify. Check for: ## Security - SQL injection vulnerabilities - XSS attack vectors - Exposed secrets or API keys - Authentication bypass risks - Input validation gaps ## Performance - N+1 database query patterns - Missing indexes on queried columns - Unnecessary re-renders in React components - Memory leaks in event listeners ## Code Quality - Functions longer than 40 lines - Missing error handling - Duplicate code that could be abstracted - TypeScript any types ## Output Format For each issue found: 1. File name and line number 2. Severity: Critical / Warning / Suggestion 3. What the problem is 4. The fix with code example Example 2: The /deploy Command Create .claude/commands/deploy.md for pre-deployment checks: # /deploy — Pre-Deployment Checklist Before any deployment, run through this checklist automatically. ## Step 1: Environment Check Verify all required environment variables are documented in .env.example. Flag any variables in the code that are not in the example file. ## Step 2: Migration Check Check if there are any pending database migrations. Warn if migrations will cause downtime on a live database. ## Step 3: Test Coverage Check Identify any new functions added since last commit with no test coverage. Flag them and ask if I want tests generated before deploying. ## Step 4: Dependency Audit Check package.json for known vulnerable packages. Suggest updates for any critical vulnerabilities. ## Step 5: Final Output Generate a deployment summary with all findings. Give a GO or NO-GO recommendation with reasons. Example 3: The /feature Command Create .claude/commands/feature.md to scaffold new features consistently: # /feature [FEATURE NAME] — Feature Scaffolding When given a feature name, scaffold all the files needed following our project conventions. ## Files to Create 1. /app/api/[feature]/route.ts — API endpoint 2. /components/[Feature]/index.tsx — Main component 3. /components/[Feature]/[Feature].test.tsx — Tests 4. /lib/[feature].ts — Business logic 5. /types/[feature].ts — TypeScript types ## Each File Should Include - Proper TypeScript types - JSDoc comments on all exports - Error handling - Basic test structure ## After Scaffolding Show me a summary of all files created. Ask me what the feature should actually do before writing logic. The memory/ Folder: Persistent Project Intelligence The memory/ folder stores information that persists between Claude Code sessions. Unlike a regular conversation that resets when you close the terminal, memory files give Claude ongoing context about your project. This is how you stop repeating yourself. You write it once and Claude remembers it forever. project.md — The Project Memory File # Project Memory — [YOUR PROJECT NAME] ## Current Sprint Goals - Implement user authentication with JWT - Build the dashboard analytics component - Migrate legacy endpoints to new API structure ## Known Issues - The /api/users endpoint is slow above 1000 records — needs pagination - Dark mode toggle broken on Safari — see issue #142 - Email sending fails silently in staging environment ## Architecture Decisions - We chose Prisma over Drizzle for team familiarity - Auth is JWT not sessions because we need mobile support - All dates stored as UTC in the database ## Team Conventions - PR must have at least one reviewer before merging - Commit messages follow conventional commits format - Feature branches named: feat/[ticket-number]-[description] ## What We Have Tried That Did Not Work - Redis caching for user sessions — caused issues with multi-region - GraphQL — team velocity dropped, switched back to REST Update this file regularly as your project evolves. The more current it is the more useful every Claude session becomes. The settings.json File: Configuration and Preferences The settings.json file controls how Claude Code behaves in your project. Here is a production configuration: { "model": "claude-opus-4-5", "context": { "include": [ "src/**/*.ts", "src/**/*.tsx", "CLAUDE.md", "package.json", "prisma/schema.prisma" ], "exclude": [ "node_modules/**", ".next/**", "dist/**", "**/*.test.ts" ] }, "commands": { "directory": ".claude/commands" }, "memory": { "enabled": true, "directory": ".claude/memory" }, "output": { "format": "markdown", "verbosity": "normal" } } Real World Use Cases Here is how professional developers are using the .claude/ folder to build serious systems. 🔐 Use Case 1: Security-First Development Team A security team creates a /security command that runs OWASP checks on every new endpoint. The CLAUDE.md file lists all known vulnerabilities in their stack. Memory tracks every security decision made across sprints. Every developer on the team gets the same security standards automatically. 🚀 Use Case 2: Automated Code Review Before PRs A solo developer creates a /review command that runs before every PR. It checks for their personal code standards, generates test suggestions for uncovered code paths, and produces a review summary they can attach to the PR description. Cuts review time in half. 📊 Use Case 3: Data Engineering Pipeline Team A data team stores their entire schema documentation in memory/ so Claude always knows the shape of the data. Custom commands handle common transformations. CLAUDE.md documents which tables are append-only versus mutable. Claude never suggests a query that would break their data contracts. 🎨 Use Case 4: Design System Enforcement A frontend team puts their entire component library in CLAUDE.md. Claude knows every component that exists, every prop, and every variant. The /component command scaffolds new components that automatically match the design system. No more inconsistent implementations across the codebase. 🤖 Use Case 5: AI Agent Development An AI engineer building Claude-powered agents stores their agent prompts, tool definitions, and conversation templates in .claude/memory/. The /agent command scaffolds new agents with their standard structure. CLAUDE.md documents the multi-agent architecture so Claude never suggests patterns that break the system. The 10-Minute Setup: Get Started Right Now Here is everything you need to set up a professional .claude/ folder from scratch in 10 minutes. Step 1: Create the folder structure mkdir -p .claude/commands .claude/memory Step 2: Create your CLAUDE.md touch CLAUDE.md Then add your project overview, architecture rules, code standards, and what to avoid. Use the template from earlier in this article as your starting point. Step 3: Add your first custom command touch .claude/commands/review.md Add the /review command from this article. Run it on your first file to see what it catches. Step 4: Start your memory file touch .claude/memory/project.md Add your current sprint goals, known issues, and the most important architectural decisions. Update it every week. Step 5: Add .claude/ to .gitignore selectively # Add to .gitignore # Keep commands and CLAUDE.md in git — share with team # Exclude conversations — those are personal .claude/conversations/ Your commands and CLAUDE.md should be version controlled and shared with your team. Your conversation history is personal. Advanced: Dynamic Memory Updates The most powerful pattern is having Claude update its own memory files as your project evolves. Add this to your CLAUDE.md: ## Memory Update Instructions After every session where a significant decision is made: 1. Add the decision to .claude/memory/project.md 2. Include why the decision was made 3. Note any alternatives that were considered 4. Flag if the decision should be revisited If a known issue is resolved, remove it from the issues list. If a new pattern emerges, document it under Architecture Decisions. With this in place Claude maintains its own project intelligence over time. Each session builds on the last. The longer you use it the smarter it gets about your specific project. The Bottom Line The .claude/ folder is the difference between using Claude Code as a fancy autocomplete and using it as a genuine development partner. The setup takes 10 minutes. The payoff compounds indefinitely. Every custom command you write saves you from typing the same instructions again. Every memory entry you add means one less thing you have to re-explain. Every CLAUDE.md improvement means Claude works more like a developer who has been on your team for months. Most developers will read this and never create the folder. The ones who set it up this week will never go back to working without it. follow @cyrilxbt for more ai /crypto content