Ryan Williams ryanwi

Inside the Claude Code source

Anthropic's Claude Code CLI source code leaked onto GitHub recently. All of it. About 1,900 files and a lot of TypeScript.

I read through the key modules. What follows is a breakdown of the surprising parts: how the system actually works, where Anthropic made clever engineering choices, and where their approach diverges from OpenAI's Codex in ways you wouldn't guess from using either tool.

Lifecycle of a request

Here's what happens when you type a message into Claude Code:

OpenCode + Kimi For Coding (K2.5) Setup

1. Add credentials

opencode auth login
# Select "Kimi For Coding"
# Enter your API key (sk-kimi-...)

Purpose

Bootstrap knowledge of LLMs ASAP. With a bias/focus to GPT.

Avoid being a link dump. Try to provide only valuable well tuned information.

Prelude

Neural network links before starting with transformers.

I have an updated version of this on my blog here: https://chrisamico.com/blog/2023-01-14/python-setup/.

Python

This is my recommended Python setup, as of Fall 2022. The Python landscape can be a confusing mess of overlapping tools that sometimes don't work well together. This is an effort to standardize our approach and environments.

Tools and helpful links:

Python docs: https://docs.python.org/3/
Python Standard Library: - Start here when you're trying to solve a specific problem

	require "json"

	struct = { "a" => 1, "b" => 2, "c" => [1, 2, 3], "d" => [{ "e" => 3 }, nil, false, true, [], {}] }
	source = JSON.dump(struct)

	tokens = []
	index = 0

	until source.empty?
	tokens <<

	This is free and unencumbered software released into the public domain.

	Anyone is free to copy, modify, publish, use, compile, sell, or
	distribute this software, either in source code form or as a compiled
	binary, for any purpose, commercial or non-commercial, and by any
	means.

	In jurisdictions that recognize copyright laws, the author or authors
	of this software dedicate any and all copyright interest in the
	software to the public domain. We make this dedication for the benefit

	Card = Struct.new(:suite, :rank) do
	include Comparable

	def precedence() = [SUITES_SCORES[self.suite], RANKS_SCORES[self.rank]]
	def rank_precedence() = RANKS_SCORES[self.rank]
	def suite_precedence() = SUITES_SCORES[self.rank]

	def <=>(other) = self.precedence <=> other.precedence

	def to_s() = "#{self.suite}#{self.rank}"

	ruby '2.7.1'

	gem 'rails', github: 'rails/rails'
	gem 'tzinfo-data', '>= 1.2016.7' # Don't rely on OSX/Linux timezone data

	# Action Text
	gem 'actiontext', github: 'basecamp/actiontext', ref: 'okra'
	gem 'okra', github: 'basecamp/okra'

	# Drivers

	# LVDB - LLOOGG Memory DB
	# Copyriht (C) 2009 Salvatore Sanfilippo <antirez@gmail.com>
	# All Rights Reserved

	# TODO
	# - cron with cleanup of timedout clients, automatic dump
	# - the dump should use array startsearch to write it line by line
	# and may just use gets to read element by element and load the whole state.
	# - 'help','stopserver','saveandstopserver','save','load','reset','keys' commands.
	# - ttl with milliseconds resolution 'ttl a 1000'. Check ttl in dump!

	{
	repository(name: "sickvim", owner: "jonathansick") {
	ref(qualifiedName: "master") {
	target {
	... on Commit {
	id
	history(first: 5) {
	pageInfo {
	hasNextPage
	}