Claude Code: handle a codebase and daily tasks from the terminal in natural language

Who it's for

Use Claude Code when you want an agent coding tool that lives in the terminal, understands the current codebase in natural language, runs routine tasks, explains complex code, or handles git workflows. It is not a guide to an editor plugin marketplace; the home ground is the terminal command claude.

Precondition: install via the official script, not deprecated npm. The script will run on this machine:

https://claude.ai/install.sh

Open that address and review the contents before you decide whether to pipe it to bash.

Another precondition: during use, the vendor receives usage data (for example accepted or rejected code), related conversation data, and feedback you submit with /bug. If you cannot accept that, do not install.

The notes also say you can use it in an IDE, or tag @claude on Github. Those are other entries; the first time, do not treat them as an install path with written steps.

First try in 3 minutes

Official page:

https://github.com/anthropics/claude-code

Recommended install is only this one (it runs a remote script on this machine; confirm, then run):

curl -fsSL https://claude.ai/install.sh | bash

Do not use npm. The notes write Installation via npm is deprecated.

After install, enter a project directory of your own and run claude. The first time, only check: ask it to explain an existing file or function. Do not let it change files, and do not let it run git write operations.

Success signal: a Claude Code session appears in the terminal, and it can explain that code from the current directory. If claude is missing or the script fails, stop; do not switch to an installer that never appeared.

Use cases

Case 1: Handle code in a project terminal. Follow the notes: enter the project directory and run claude. Do one low-risk thing in natural language: explain complex code, or describe the next routine task. Git workflows are named, but the notes do not list allowed git subcommands; when commit, push, or history changes are involved, have it state the commands first, confirm, then continue, and immediately check with git status.

Case 2: Switch to the IDE or Github entry. The original sentence is “Use it in your terminal, IDE, or tag @claude on Github.” Outside the terminal, evidence only shows that an IDE usage and an @claude tag exist; no plugin name, no config steps. If your environment can already open it in an IDE or tag @claude in a Github comment, you may ask a read-only question once; if the entry does not exist, do not invent an install path.

Boundaries

The main install is curl -fsSL https://claude.ai/install.sh | bash. npm is deprecated. Do not add other installers.

During use the vendor receives usage data, conversation data, and /bug feedback. The notes do not write a way to turn that off.

Terminal, IDE, and @claude on Github are all named, but the last two have no steps. Do not infer that only the terminal works, and do not invent installs for unwritten entries.

The notes give no language coverage, pricing, or Windows-only commands. Keep undocumented items unknown.