Build Your Own Claude Code

This repo contains solutions to CodeCrafters' "Build Your own Claude Code" challenge, implemented in two languages:

• codecrafters-claude-code-go/ — Go solution
• codecrafters-claude-code-python/ — Python solution

The challenge is to build a minimal, working clone of Claude Code — an LLM-powered coding assistant that can read files, write files, and run shell commands on request. Each solution sends prompts to an LLM (via the OpenRouter API, using an OpenAI-compatible chat.completions interface) and drives an agent loop:

1. Send the conversation history + available tool specs to the model.
2. If the model responds with tool_calls, execute the requested tool(s) locally (Read, Write, Bash) and feed the results back as tool messages.
3. Repeat until the model replies with plain text instead of a tool call, then print that as the final answer.

Structure

Each solution directory is a self-contained CodeCrafters project with its own app/ entry point, codecrafters.yml config, and your_program.sh runner:

Directory	Entry point	Status
codecrafters-claude-code-go	app/main.go	Early stage — first-stage response handling only
codecrafters-claude-code-python	app/main.py	Further along — full agent loop with Read, Write, and Bash tools

See codecrafters-claude-code-python/NOTES.md for detailed notes on how tool calling and the agent loop are implemented.

Running

Each subdirectory has its own your_program.sh. For example, from codecrafters-claude-code-python:

export OPENROUTER_API_KEY=...
./your_program.sh -p "Summarize the README for me."

Both solutions require an OPENROUTER_API_KEY environment variable (and optionally OPENROUTER_BASE_URL, which defaults to OpenRouter's API).

Note: If you're viewing this on GitHub, head over to codecrafters.io to try the challenge yourself.