Claude Code

Claude Code is Anthropic's coding agent that works directly in your terminal and editor. It can read your codebase, run commands, edit files, and carry out multi-step engineering tasks, not just answer questions about code.

This page is a primer. If you don't write software, you can skip it, but the ideas (agents that take action) are worth understanding.

What makes it different from chat

In claude.ai, Claude tells you how to write code. Claude Code does it, in your actual project:

• Reads and navigates your real files.
• Runs commands and tests.
• Makes edits across multiple files.
• Iterates toward a working result, checking its own work.

It's a true agent: given a goal, it takes a sequence of actions to achieve it.

What people use it for

• Implementing features from a description.
• Large refactors across many files.
• Writing and fixing tests.
• Debugging from an error message.
• Automating repetitive dev chores.

How it connects to the rest of this handbook

• Skills work in Claude Code; package your team's conventions and Claude follows them.
• MCP connects it to external tools and data.
• The same prompting principles apply: clear goals, context, constraints, and iteration.

Good habits

• State the goal and constraints clearly, like any prompt.
• Work in small steps and review changes as they happen.
• Use version control. Let it work on a branch so you can review and roll back.
• Keep a human in the loop for anything important; you're the reviewer.

Beyond engineers

"Agentic" tools like Claude Code point to where work is heading: you describe an outcome, and the agent executes multi-step tasks. Even non-developers benefit from learning to delegate goals, not just ask questions.

For the bigger picture, see agents & computer use.