Loading...

Build your own Claude Code

Learn about LLM APIs, tool calling, agent loops and more

Start Building
C
C
C#
C#
C++
C++
Go
Go
Java
Java
JavaScript
JavaScript
Python
Python
Rust
Rust
TypeScript
TypeScript
Zig
Zig

This challenge is free to try when it's in beta. We keep challenges in beta for a few weeks to gather feedback.

Claude Code is a terminal-based AI coding assistant that uses Large Language Models (LLMs) to understand code and complete programming tasks. In this challenge, you'll build your own Claude Code that is capable of editing files, running commands, and iterating until the task is done.

Along the way, you'll learn about LLM APIs, tool calling, agent loops, and how to integrate multiple tools into an AI assistant.

Stages

Communicate with the LLM
Login via GitHub to view this stage
Advertise the read tool
Login via GitHub to view this stage
Execute the read tool
Login via GitHub to view this stage
Implement the agent loop
Login via GitHub to view this stage
Implement the write tool
Login via GitHub to view this stage
Implement the bash tool
Login via GitHub to view this stage

Skills

Advertise skills to the LLM
Login via GitHub to view this stage
Invoke a skill by name
Login via GitHub to view this stage
Pass arguments to a skill
Login via GitHub to view this stage
Stack multiple skills
Login via GitHub to view this stage
Run a script bundled with a skill
Login via GitHub to view this stage
Let the model choose a skill
Login via GitHub to view this stage
Run a skill in a subagent
Login via GitHub to view this stage

Language Server Protocol

Add the LSP tool
Login via GitHub to view this stage
Find the implementation
Login via GitHub to view this stage
Find the affected callers
Login via GitHub to view this stage
Keep the analysis current
Login via GitHub to view this stage
Repair edits with diagnostics
Login via GitHub to view this stage