🤖 CLI and Configuration

The three ways to install Claude Code: npm, Homebrew, and winget.

Without the right installation, nothing else works.

Node.js 18+; global npm install; API key authentication.

Run the claude command for the first time and understand what appears.

The first real step. Without it, the rest is theory.

Interactive interface; Claude prompt; inline response.

claude (interactive) vs. claude -p "question" (one-shot, no session).

Knowing when to use each mode saves time and context.

Interactive mode = conversation; -p flag = one-shot; pipes with -p.

The flags you'll use every day: --help, --version, -p, --model, --verbose.

Knowing the flags keeps you from getting lost on Google.

--help as a quick reference; most-used flags.

How to update Claude Code and check the installed version.

New versions bring important features and fixes.

npm update -g; changelog; breaking changes.

The most common installation errors and how to fix each one.

Knowing how to fix errors keeps you from getting stuck right at the start.

EACCES permission; Node version mismatch; auth errors.

Claude Code keyboard shortcuts: Escape, Tab, Ctrl+C, and more.

Shortcuts save clicks and keep you focused on the terminal.

Escape to cancel; Tab to accept; Ctrl+C to interrupt.

Slash commands inside the Claude Code interactive session.

They are the control panel within the conversation.

/help; /clear; /model; /memory; /resume; /compact.

Which model runs behind Claude Code, and how does context work?

Understanding the model helps you calibrate expectations and costs.

Opus/Sonnet; context window; token budget.

Sessions persist between runs. /resume picks up the previous conversation.

Don't lose your work context when you close the terminal.

Session ID; /resume; --resume flag; session picker.

Claude Code reads images from the filesystem and analyzes screenshots.

Visual debugging, reproducing designs, reading diagrams.

Drag-and-drop; paste a screenshot; reference by path.

Use claude -p with Unix pipes: | jq, | grep, > file.

Native integration with the Unix ecosystem multiplies its power.

Stdout; --output-format json; piping with jq; wrapper scripts.

A Markdown file that Claude Code reads automatically in every session.

It’s the only way to give the agent persistent context about your project.

System prompt; persistent instructions; project context.

~/.claude/CLAUDE.md (global) > ./CLAUDE.md (project) > subdir/CLAUDE.md (folder).

Each level solves a different scope problem.

Instruction cascade; inheritance; override by level.

How to write a CLAUDE.md that the model actually follows.

A poorly written CLAUDE.md gets ignored; a good one changes everything.

Clear headings; lists; examples; imperative tone.

Which types of instructions actually change the agent’s behavior.

Fewer, right instructions > many vague instructions.

Stack; conventions; prohibitions; project standards.

Annotated examples of CLAUDE.md files from real projects.

Seeing real examples shortens the learning curve.

Common patterns; anti-patterns; ideal size.

CLAUDE.md should evolve along with the project and be committed to git.

An outdated CLAUDE.md is worse than none at all.

Commit to the repo; review monthly; /memory as a supplement.

The JSON file that controls global and project-level behavior.

Without settings.json, you’re stuck with generic defaults.

~/.claude/settings.json; .claude/settings.json; config merging.

File that tells Claude Code which files/folders to ignore.

Protects secrets, reduces unnecessary context, and speeds up the agent.

Glob syntax; .env; node_modules; build artifacts.

Define which tools Claude Code can use without asking for permission.

Balance security with productivity.

allowedTools; deniedTools; permission prompts; trust levels.

Environment variables recognized by Claude Code: ANTHROPIC_API_KEY, CLAUDE_MODEL, etc.

Configuration by environment without touching files.

ANTHROPIC_API_KEY; CLAUDE_MODEL; local .env.

Different configs for different projects/contexts.

A personal project and a corporate one have different needs.

Local vs. global settings; config inheritance.

How to use --model, /model, and alternative providers (Bedrock, Vertex).

Flexibility to use the right model for each task.

--model flag; CLAUDE_MODEL; Bedrock; Vertex AI.

Claude Code doesn't have magical memory. Everything lives in the session's context window.

Understanding the mental model prevents frustration with “it forgot.”

Context window = working memory; CLAUDE.md = persistent memory.

/memory saves persistent notes that Claude Code reads in every future session.

It’s the quick way to teach preferences without editing CLAUDE.md.

/memory; generated CLAUDE.md; ~/.claude/CLAUDE.md.

The context window has a fixed limit (200k tokens). Everything competes for space.

When the context fills up, the quality of the responses drops.

Tokens; context limit; gradual degradation.

Claude Code automatically summarizes the conversation when the context fills up.

Compaction loses details. Knowing when it will happen helps you prepare.

/compact manual; auto-compaction; what gets lost.

Strategies for transferring context from one session to another.

Long sessions degrade. Knowing when to split them keeps quality high.

/resume; manual summary; CLAUDE.md as handoff.

Techniques for saving the state of your work before the context fills up.

Checkpoints prevent progress loss on long tasks.

Git commits as checkpoints; notes in CLAUDE.md; TODO files.

Shell aliases that shorten frequently used Claude Code commands.

Every second saved adds up across hundreds of uses.

alias cc='claude'; alias ccr='claude --resume'; .bashrc/.zshrc.

Shell scripts that combine claude -p with other tools.

Wrapper scripts automate repetitive tasks with AI.

Wrapper scripts; piped input; output parsing.

Use Claude Code in the VS Code terminal with the official extension.

VS Code is the most widely used editor. Integrating it well maximizes productivity.

Integrated terminal; Claude Code extension; side by side.

Use tmux or screen to run Claude Code in persistent sessions.

Sessions survive disconnections and allow multiple agents.

tmux new-session; side-by-side panes; detach/attach.

Keep Claude Code configs in version-controlled dotfiles.

Reproducible setup on any new machine.

Dotfiles repo; symlinks; bootstrap script.

The complete setup that combines everything: aliases + tmux + VS Code + dotfiles.

A concrete reference for how everything fits together.

Integrated workflow; daily cadence; incremental progress.