Claude Code isn’t just another chatbot—it acts directly in your terminal (or in the Claude for Desktop app) to read, edit and run code, manage files and install software without shifting you between windows. Tell it your goal in plain English—say, “Build a portfolio site with dark mode, two columns, contact form at bottom”—and Claude figures out the commands. It pauses for your approval on big actions like installing packages or deleting files, but you can pre-approve low-risk tasks in settings.local.json or block sensitive files entirely with a deny list. Under the hood, Claude Code combines three built-in tools—file I/O, code editing and terminal commands—with a context window that holds your conversation history and any files it’s read. When that window reaches about 85–95% capacity, you can run /compact to summarize and free up space or /clear to start fresh. Sessions save automatically, and you can jump back in with claude — resume. You’re billed in tokens—roughly 0.75 words each—and you track usage with /cost. You can switch models midstream using /model: Haiku for speed and low cost, Sonnet as a balanced default, and Opus when you need maximum horsepower. Project-specific notes live in a CLAUDE.md file created by /init; automatic memory across projects picks up your preferred style and workflows. Slash commands let you run built-in actions or define your own, skills load specialist instruction sets on demand, and hooks trigger background scripts for tasks like auto-formatting. Flags at startup—like —verbose or —model—adjust detail levels, tool access and autonomy. With these controls you decide how much Claude can do on its own. The tool’s design means you spend less time swapping between apps, documenting context or micromanaging commands, and more time focusing on the outcome.