はじめてのClaude Codeプロジェクト

Claude Code is installed as a global npm package. After installation, you authenticate with your Anthropic account. The CLI stores your credentials securely and manages sessions automatically. Once authenticated, navigate to your project directory and run 'claude' to start an interactive session. Claude Code automatically detects your project structure, reads package.json, and understands your tech stack without any configuration. For team environments, consider using environment variables for authentication so that CI/CD pipelines and shared development machines can use Claude Code without manual login steps.

CLAUDE.md is the single most important file for Claude Code productivity. It acts as persistent instructions that Claude reads at the start of every session. Think of it as onboarding documentation for your AI pair programmer. Start minimal. A good CLAUDE.md answers three questions: What is this project? What tech stack does it use? What conventions should Claude follow? Resist the urge to write a comprehensive document on day one. Instead, add rules organically as you discover patterns. Place CLAUDE.md at the project root. Claude Code searches for it automatically. You can also create a ~/.claude/CLAUDE.md for global instructions that apply to all projects.

Claude Code uses a permission model to control which actions it can take autonomously. By default, it asks for confirmation before running shell commands, editing files, or making network requests. This is safe but can slow you down. For trusted projects, you can pre-approve specific actions. The three permission modes are: ask (default), auto-approve for read-only operations, and full auto-approve. Most developers start with the default and gradually allow more as they build trust. You can also configure per-tool permissions. For example, you might allow file edits automatically but require confirmation for git operations. This granular control lets you balance speed with safety.

With setup complete, it is time to do real work. Start with a small, well-defined task to build confidence. Good first tasks include: adding a utility function, writing tests for existing code, or fixing a known bug. Be specific in your prompts. Instead of 'make the code better,' say 'add input validation to the createUser endpoint in src/routes/users.ts.' Specific prompts produce better results and consume fewer tokens. After each session, review what Claude did. Check the git diff, run your tests, and note any patterns you want to encode into CLAUDE.md for future sessions. This feedback loop is how you progressively make Claude Code more effective for your specific project.