Claude Code is Anthropic's terminal-native AI coding agent — not an IDE plugin, not a completion engine, but an autonomous agent that reads your codebase, reasons about what needs to change, executes the changes, runs tests, and iterates until the task is complete. Getting started requires understanding Claude Code's agentic interaction model, which is fundamentally different from Copilot or Cursor. This guide walks through installation, the CLAUDE.md configuration system, effective task patterns, and enterprise deployment best practices.
What Claude Code Is (and Isn't)
Read, Glob, Grep), create an implementation plan, make file changes using Write and Edit tools, run commands via Bash, read output (including test failures and compiler errors), and iterate until the task is complete. Unlike completion tools that suggest the next line, Claude Code reasons about multi-file tasks holistically. It uses Anthropic's Claude claude-sonnet-4-6 model by default via your API key — you pay per token, not a flat subscription.Installation and Setup
Prerequisites: Node.js 18+. Install globally: npm install -g @anthropic-ai/claude-code. Authenticate with your Anthropic API key: export ANTHROPIC_API_KEY=your-key-here (add to .zshrc/.bashrc). Launch from any project directory: claude. First run opens an interactive session. Verify: claude --version. For enterprise deployment, configure the key via your secrets manager and set ANTHROPIC_API_KEY in your development environment. Rate limits apply per your Anthropic API tier — ensure your tier supports the token volume for your team size.
Create a CLAUDE.md file in your repository root — this is the most important configuration step. Claude Code reads it at the start of every session. Include: project architecture overview, file structure conventions, key commands (npm test, make build), coding standards, important patterns to follow, known gotchas, and what not to change. The more context CLAUDE.md provides, the higher quality and more architecturally consistent Claude Code's outputs will be. Commit it to the repo — it is part of your engineering onboarding infrastructure.
Claude Code asks for permission before running bash commands that modify files or execute code — you can approve per-command or trust specific commands for the session. For production environments, use --allowedTools flag to restrict which tools Claude Code can use. Never run Claude Code with bash access on production infrastructure. For CI/CD pipeline use (automated code generation, test writing), use claude --non-interactive mode with precisely scoped allowed tools and a dedicated API key with appropriate rate limits.
Effective Task Patterns
| Task Type | Example Prompt | Tips |
|---|---|---|
| Feature implementation | "Add rate limiting middleware to all API routes using the existing Redis client — max 100 req/min per API key" | Reference existing patterns; Claude finds similar code automatically |
| Bug fixing | "The /users endpoint returns 500 when the email is null. Fix it and add a test for null email input." | Give exact reproduction details; Claude will read error logs |
| Refactoring | "Refactor the auth module to use the new JWT library replacing the old one — maintain all existing tests passing" | Specify tests as the success criterion — Claude iterates until they pass |
| Test generation | "Add unit tests for every exported function in src/utils/pricing.ts — target 90% line coverage" | Give coverage target; Claude reads function signatures and generates matching tests |
| Documentation | "Generate JSDoc comments for all public API methods in src/api/ — follow the pattern in src/api/users.ts" | Reference an example for consistent style |
- Multi-file feature implementations with consistent patterns
- Test generation — reads code, generates comprehensive tests, iterates until passing
- Large refactors across many files where context is critical
- CI/CD pipeline tasks — automated code generation, migration scripts, documentation
- Security-critical changes — authentication, authorisation, crypto
- Database migrations — always review before running on production data
- Any change that modifies existing behaviour rather than adding new capability
Our software development and DevOps teams help enterprises deploy Claude Code — API key management, CLAUDE.md configuration, CI/CD integration, and team enablement. Book a free advisory session.