Claude Code

Getting Started

1. Install globally: npm install -g @anthropic-ai/claude-code (requires Node.js 18+).
2. Navigate to your project and run claude to start an interactive session — it automatically maps your codebase.
3. Run /init to generate a CLAUDE.md file with project-specific instructions, conventions, and context that persists across sessions.
4. Start coding: ask it to fix bugs, add features, refactor code, write tests, create commits, or open PRs.

Slash Commands

• /init — Set up a CLAUDE.md with project conventions and context
• /compact — Compress conversation history to free up context window
• /cost — Show token usage and cost for the current session
• /clear — Reset conversation while staying in the same project
• /help — View all available commands and keyboard shortcuts
• /review — Review a PR or diff with detailed feedback
• /commit — Stage changes and create a well-formatted git commit

Key Features

• Powered by Opus 4 — Uses the most capable Claude model for deep codebase understanding, multi-step reasoning, and complex autonomous coding tasks.
• Agentic tool use — Reads and edits files, runs terminal commands (builds, tests, linters), searches with glob/grep, and fetches web content — all autonomously chained together.
• Git-native workflow — Creates branches, stages files, writes commit messages following your repo’s conventions, opens PRs via gh, and handles merge conflicts.
• CLAUDE.md project memory — Persistent project instructions that teach Claude your stack, conventions, testing patterns, and preferences across sessions.
• Hooks system — Configure shell commands that run automatically in response to events (pre/post tool calls, notifications). Perfect for enforcing formatting, running linters, or custom validation.
• MCP server integration — Connect to external tools and services via the Model Context Protocol — databases, APIs, Slack, Jira, GitHub, and any custom MCP server.
• IDE extensions — First-class VS Code extension with inline diff previews, status bar integration, and the ability to run Claude Code directly from your editor.
• Claude Code Agent SDK — Build custom AI agents on top of Claude Code’s infrastructure. Create specialized agents with their own tools, system prompts, and orchestration logic.
• Background agents — Kick off long-running tasks (test suites, large refactors) that run asynchronously while you continue working.
• Multi-file refactoring — Performs complex changes across dozens of files simultaneously with awareness of imports, types, cross-file dependencies, and test coverage.
• Subagent architecture — Spawns specialized sub-agents for parallel research, exploration, and task execution within a single session.
• Auto-memory — Automatically saves patterns, debugging insights, and project-specific knowledge to a persistent memory directory for use across sessions.