Claude Code

Anthropic's AI-Powered CLI for Developers

Code Faster, Debug Smarter, Ship More

March 2026

1 / 16
 What is Claude Code?

Claude Code is an agentic AI coding tool that lives in your terminal.

It understands your codebase, edits files, runs commands, and
helps you ship software faster -- all through natural language.

It's not a chatbot. It's a coding partner that can:

- Read and understand your entire codebase

- Edit multiple files in a single operation

- Run shell commands and interpret results

- Search code with grep, glob, and semantic understanding

- Create commits, PRs, and manage git workflows

- Debug errors by reading logs and tracing issues

- Work autonomously on complex multi-step tasks

2 / 16
 Installation & Setup

Install via npm ([Link] 18+ required):

npm install -g @anthropic-ai/claude-code

Launch it in any project directory:

cd ~/my-project

claude

First run will prompt you to authenticate with your Anthropic API key.

Requirements:

- [Link] 18+ and npm

- Anthropic API key (or Claude Max/Pro subscription)

- Works on Linux, macOS, and WSL

3 / 16
 The Interactive Interface

Claude Code runs as an interactive REPL in your terminal:

$ claude

Claude Code v1.x

Working directory: ~/my-project

> fix the bug in the login handler where

passwords aren't being hashed

Claude reads src/auth/[Link], identifies the issue,

proposes an edit, and asks for your approval.

> /help # Show all commands

> /compact # Compress conversation context

> /clear # Clear conversation history

4 / 16
 Core Capabilities

Code Intelligence System Integration

- Read & understand files - Run shell commands

- Search with grep & glob - Git operations (commit, PR)
- Multi-file editing - Run & interpret tests
- Refactoring at scale - Manage dependencies
- Code generation - Docker & infrastructure
- Bug diagnosis & fixing - Web fetching & APIs

Claude sees tool outputs, interprets errors, adjusts its approach,

and keeps working until the task is done -- like a real collaborator.

5 / 16
 Permission Modes

You control how much autonomy Claude has:

Ask mode Claude asks before every action (safest)

Auto-edit File edits are automatic, commands need approval

Full auto Everything runs automatically (most productive)

You can also configure granular permissions:

# Allow specific tools without prompting

# In settings or via flags:

claude --allowedTools 'Read,Glob,Grep,Edit'

# Or trust specific bash commands

claude --allowedTools 'Bash(npm test)'

Every action is shown in the terminal. You always see what's happening.
claude --allowedTools 'Bash(git status)'

6 / 16
 Everyday Workflows

Just describe what you want in plain English:

> add input validation to the user registration endpoint.

email must be valid, password at least 8 chars with

one uppercase and one number.

> the tests in test_auth.py are failing. figure out why

and fix them.

> refactor the database module to use connection pooling

instead of creating a new connection per request.

> explain how the payment processing pipeline works.

trace the flow from checkout to confirmation.

> create a PR for the changes on this branch with a

good description.
7 / 16
 Slash Commands

/help Show all available commands

/compact Compress context to free up token space

/clear Reset conversation history

/commit Create a git commit with a good message

/pr Create a pull request

/review Review code changes

/fast Toggle fast mode (same model, faster output)

/bug Report a bug to the Claude Code team

/memory View or edit project memory

You can also create custom slash commands by placing

markdown files in .claude/commands/ in your project.

8 / 16
 [Link]: Project Memory

[Link] is a file at your project root that gives Claude

persistent instructions, conventions, and context.

# [Link]

## Project overview

This is a FastAPI backend for an e-commerce platform.

## Tech stack

- Python 3.12, FastAPI, SQLAlchemy, PostgreSQL

- Tests: pytest with async fixtures

## Conventions

- Use snake_case for all Python identifiers

- All endpoints must have type hints and docstrings

- Run `make lint` before committing

9 / 16
## Important paths
 Custom Slash Commands

Create reusable prompts your whole team can share:

# .claude/commands/[Link]

Review the current changes for security vulnerabilities.

Check for:

- SQL injection

- XSS

- Authentication bypass

- Secrets in code
Use-it:Insecure deserialization

Report findings with severity levels.

> /review-security

Commit the .claude/commands/ folder so the whole team benefits.

10 / 16
 Headless Mode & CI Integration

Run Claude non-interactively for automation:

# One-shot command (no interactive session)

claude -p "add type hints to all functions in src/[Link]"

# Pipe input

cat [Link] | claude -p "explain this error and suggest a fix"

# In a CI pipeline

claude -p "review the diff and comment on any issues" \

--allowedTools "Read,Glob,Grep"

# Output as JSON for scripting

Useclaude
cases: -p "list all TODO comments" --output-format json
automated code review, PR descriptions, changelog generation.

11 / 16
 Multi-File Operations

Claude excels at changes that span multiple files:

> rename the User model to Account across the entire

codebase. Update the model, all imports, API endpoints,

tests, and documentation.

> add error logging to every API endpoint. Use the

existing logger from src/utils/[Link].

> migrate from unittest to pytest across all test files.

convert assertions, fixtures, and parametrize where

it makes sense.

> add a new "notifications" feature. Create the model,

Claudemigration,
reads the existing code first, then makes coordinated changes.
API routes, service layer, and tests.

12 / 16
 Debugging with Claude

Give Claude an error and let it investigate:

> npm test is failing with "Cannot read property 'id'

of undefined" in [Link]. Find and fix it.

Claude will:

1. Read the test file

2. Read the service being tested

3. Run the test to see the full stack trace

4. Identify the root cause

5. Propose and apply a fix

6. Re-run the test to verify

> the app crashes when a user uploads a file > 10MB.
Claudefind
iterates: fix, test, read error, fix again -- until it works.
where the limit is set and increase it to 50MB.

13 / 16
 Git & GitHub Workflows

# Smart commits

> /commit

Claude analyzes staged changes and writes a descriptive

commit message following your project conventions.

# Pull request creation

> /pr

Claude reviews all commits on the branch, writes a

summary, test plan, and creates the PR via gh CLI.

# Code review

> review the PR at [Link]/org/repo/pull/42

Claude reads the diff and provides detailed feedback.

# Resolving merge conflicts

> resolve the merge conflicts in src/api/[Link]

keeping our changes for the auth logic but theirs

for the pagination updates. 14 / 16

 Tips & Best Practices

- Write a good [Link] -- it's the single biggest productivity boost

- Be specific in your prompts: name files, functions, expected behavior

- Use /compact when context gets long to free up token space

- Let Claude run tests after changes to verify correctness

- Create custom slash commands for repeated workflows

- Use headless mode (-p) for scripting and CI pipelines

- Review diffs before approving -- Claude is good but not perfect

- Start with ask mode, graduate to auto as you build trust

- Use subagents (Agent tool) to parallelize research tasks

- Keep [Link] in version control so the team shares context

15 / 16
 Claude Code at a Glance

Install npm install -g @anthropic-ai/claude-code

Launch cd project && claude

Ask anything Describe what you want in plain English

[Link] Project conventions + persistent context

Slash commands /commit, /pr, /compact, /review

Custom commands .claude/commands/[Link]

Headless claude -p "your task here"

Permissions Ask mode -> Auto-edit -> Full auto

Think less about tooling. Think more about building.

[Link]/claude-code

16 / 16