Claude Code's source code leaked, and Riley Brown used OpenAI Codex to analyze it, then forked it into a custom CLI called RileyCode. He swapped the personality, replaced Anthropic's action verbs with his own, and enabled 'dangerously skip permissions' by default, exposing how thin the abstraction layer between Claude Code and a fully custom agent actually is.
The more technically interesting move is what came next. Brown wrapped RileyCode in a desktop app built from scratch, adding persistent memory, a built-in browser, and a custom editing interface. The video walks through each layer of that build, timestamps at 08:50 through 16:03, showing how much of Claude Code's architecture is reusable scaffolding rather than proprietary logic.
Watch this if you want to understand what Claude Code actually is under the hood, not what Anthropic's marketing says it is. The gap between 'leaked source code' and 'shippable desktop product in under 17 minutes' tells you something real about the current state of AI tooling.
[WATCH ON YOUTUBE →]