Understanding Agentic Tools

Introduction

Welcome to the fourth unit of Foundation — Getting Started with Claude Code! We have covered a lot of ground already: we understand what Claude Code is, how to start and manage sessions, and how to control our environment using essential commands like /status, /model, and /context . Now, we are ready to explore what makes Claude truly powerful: its agentic tool use . In this lesson, we will discover what it means for Claude to be agentic and examine the tools it uses to accomplish tasks autonomously. Rather than simply suggesting code for us to implement manually, Claude Code can create files, modify existing code, execute commands, and even search the web. We will see these tools in action and understand how the permission system keeps us in control.

What Does "Agentic" Mean?

The term agentic refers to Claude Code's ability to act autonomously on our behalf rather than merely providing suggestions. When we describe a goal, Claude does not just tell us what to do; it actually does it by selecting and using the appropriate tools. Think of the difference between a traditional coding assistant that tells us "write this function here" and Claude Code that actually writes the function. We provide the destination, and Claude takes the wheel, making decisions and performing actions to get us there. This autonomy extends across the entire development workflow. Claude can create new files, read and analyze existing code, modify files with surgical precision, execute shell commands, and fetch information from the web. Each action uses a specific tool, and Claude decides which tools to employ based on our goals.

Claude's Toolbox

Claude Code has access to several specialized tools organized into categories: File Operations: Read - Examine files (code, images, PDFs, notebooks) Write - Create new files Edit - Modify existing files with precise string replacements Glob - Find files using patterns (e.g., **/*.ts) Grep - Search file contents with regex Code Execution: Bash - Execute shell commands BashOutput - Monitor background processes KillShell - Terminate processes Web & Search: WebFetch - Fetch and analyze web content WebSearch - Search for current information Additional capabilities include: Task Management - Launch specialized agents and manage task lists Jupyter Notebooks - Edit notebook cells User Interaction - Ask clarifying questions and execute custom commands For this lesson, we will focus on the core file operations and code execution tools. You will encounter the others naturally as your projects grow more sophisticated.

Join the 1M+ learners on CodeSignal

Be a part of our community of 1M+ users who develop and demonstrate their skills on CodeSignal

Creating Files with Write

Let's see tools in action. Watch what happens when we ask Claude to create a simple Python file:

text

> Create a Python file called hello.py that prints "Hello, World!"

● Write(hello.py)

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Create file
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ hello.py                                                                                                   │
│                                                                                                            │
│ print("Hello, World!")                                                                                     │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
 Do you want to create hello.py?
 ❯ 1. Yes
   2. Yes, allow all edits during this session (shift+tab)
   3. No, and tell Claude what to do differently (esc)

> Create a Python file called hello.py that prints "Hello, World!"

● Write(hello.py)

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Create file
╭────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ hello.py                                                                                                   │
│                                                                                                            │
│ print("Hello, World!")                                                                                     │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
 Do you want to create hello.py?
 ❯ 1. Yes
   2. Yes, allow all edits during this session (shift+tab)
   3. No, and tell Claude what to do differently (esc)

Several important details emerge here. First, Claude automatically chose the Write tool—we did not specify this. Second, before making any changes, Claude shows us a preview of what will be created. This is the permission system in action.

Notice the three options:

Yes - Approve this specific operation
Yes, allow all edits during this session - Trust Claude to make changes without asking again
No, and tell Claude what to do differently - Reject and provide feedback

Once we select "Yes," Claude proceeds:

text

● Write(hello.py)
  ⎿  Wrote 2 lines to hello.py
     print("Hello, World!")

● Done! I've created hello.py with a simple print statement.

● Write(hello.py)
  ⎿  Wrote 2 lines to hello.py
     print("Hello, World!")

● Done! I've created hello.py with a simple print statement.

The indented output confirms exactly what was written and provides a friendly confirmation. The file now exists in our working directory, ready to run.

Modifying Files with Edit

Creating new files is essential, but what about changing existing ones? This is where Edit shines. Unlike Write, which creates files from scratch, Edit makes targeted changes to existing files without rewriting the entire content.

Let's modify our hello.py file:

text

> Update hello.py to also print "Welcome to Claude Code!"

● I'll update the file to include both print statements.

● Read(hello.py)
  ⎿  Read 2 lines

● Edit(hello.py)

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Edit file hello.py
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
   1    print("Hello, World!")
   2 +  print("Welcome to Claude Code!")
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
 Do you want to make this edit to hello.py?
 ❯ 1. Yes
   2. Yes, allow all edits during this session (shift+tab)
   3. No, and tell Claude what to do differently (esc)

> Update hello.py to also print "Welcome to Claude Code!"

● I'll update the file to include both print statements.

● Read(hello.py)
  ⎿  Read 2 lines

● Edit(hello.py)

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Edit file hello.py
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
   1    print("Hello, World!")
   2 +  print("Welcome to Claude Code!")
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
 Do you want to make this edit to hello.py?
 ❯ 1. Yes
   2. Yes, allow all edits during this session (shift+tab)
   3. No, and tell Claude what to do differently (esc)

Notice several important details here. First, Claude automatically reads the file before editing it to understand the current content. Then it uses the Edit tool to make the change. The preview shows us a diff format with line numbers: line 1 remains unchanged, and line 2 (marked with +) is being added. This precision is crucial: Claude preserves existing content while adding exactly what we requested.

After approving:

text

● Edit(hello.py)
  ⎿  Updated hello.py with 1 addition       
       1    print("Hello, World!")
       2 +  print("Welcome to Claude Code!")

● Done! The file now prints both messages. When you run python hello.py, it will output:
  Hello, World!
  Welcome to Claude Code!

● Edit(hello.py)
  ⎿  Updated hello.py with 1 addition       
       1    print("Hello, World!")
       2 +  print("Welcome to Claude Code!")

● Done! The file now prints both messages. When you run python hello.py, it will output:
  Hello, World!
  Welcome to Claude Code!

The confirmation shows exactly what was added and provides helpful context about what the file will do. This distinction between Write and Edit is important. When we ask Claude to "add error handling to the login function," it uses Edit to insert code in the right place. When we ask it to "create a new test file," it uses Write to build the entire file from scratch.

Executing Commands with Bash

Creating and modifying files is only half the story. Let's execute our Python script to see Bash in action:

text

> Run it

● Bash(python hello.py)
  ⎿  Running…

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Bash command

   python hello.py
   Run hello.py script

 Do you want to proceed?
 ❯ 1. Yes
   2. Yes, and don't ask again for python commands in /usercode/FILESYSTEM
   3. Tell Claude what to do differently

 Esc to exit

> Run it

● Bash(python hello.py)
  ⎿  Running…

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
 Bash command

   python hello.py
   Run hello.py script

 Do you want to proceed?
 ❯ 1. Yes
   2. Yes, and don't ask again for python commands in /usercode/FILESYSTEM
   3. Tell Claude what to do differently

 Esc to exit

Claude selected the Bash tool and determined the correct command from context. We simply said "Run it," and Claude understood what to execute. Notice that Bash also uses the permission system—before running commands, Claude shows us what will be executed and asks for approval.

After approving:

text

● Bash(python hello.py)
  ⎿  Hello, World!          
     Welcome to Claude Code!

● Perfect! The script ran successfully and printed both messages.

● Bash(python hello.py)
  ⎿  Hello, World!          
     Welcome to Claude Code!

● Perfect! The script ran successfully and printed both messages.

The output appears indented under the tool call, confirming the script executed successfully and showing both print statements from our file.

This demonstrates the power of agentic behavior: we gave high-level instructions, and Claude translated them into specific tool calls, executed them in the right order, and verified the results.

Introduction

What Does "Agentic" Mean?

Claude's Toolbox

Join the 1M+ learners on CodeSignal

Discovering Available Tools

Creating Files with Write

Modifying Files with Edit

Executing Commands with Bash

Tool Chaining in Action

The Permission System

How Claude Chooses Tools

Conclusion and Next Steps