Claude Code Source Code Leak: Tamagotchi Pet and Persistent Background Agent

The source code leak of Claude Code revealed the internal mechanisms of one of the most discussed AI development tools. Over 512,000 lines of TypeScript, Anthropic's system prompts, and traces of two undisclosed features — a virtual pet in the spirit of Tamagotchi and a constantly active background agent — ended up in public access after an accidental error in the 2.1.

88 update. Everything started with a scheduled release. After the release of Claude Code version 2.

1.88, one user on the X platform noticed that a source map was accidentally included in the package — an auxiliary file for debugging containing the original TypeScript code of the application. The user published the file contents, and the materials quickly spread across the developer community.

Ars Technica and VentureBeat were among the first to write about the leak. The data covers over 512,000 lines of code — the complete codebase of the tool, sufficient for a detailed study of its entire architecture. In the code, researchers found several layers of information.

First, system instructions — prompts that Anthropic embedded in Claude Code: they describe how the model should behave when working with code, what style to follow in responses, and what constraints to observe. Until now, the content of such instructions had remained closed. Second, the memory architecture became clear: it is now known how Claude Code stores context between sessions and accesses it during work — this has long interested developers actively using the tool in their daily practice.

The main findings were two undisclosed innovations. The first — a virtual pet, whose logic resembles the mechanics of classic Tamagotchi: an interactive element connected to user activity within the tool. Apparently, Anthropic is experimenting with gamification mechanics — such features foster attachment to the product and encourage its regular use.

The second — a mode of constantly active agent capable of working in the background without explicit commands from the user. This is a fundamentally different model of interaction with AI: not a reactive tool responding to requests, but a proactive assistant independently tracking context and acting on its own initiative. Anthropic did not publicly comment on the incident.

In subsequent updates, the source map disappeared from the package — the company silently eliminated the leak channel. The fate of the discovered features remains unknown: they may be released in final form, undergo changes, or remain internal experiments. Such cases happen in the industry, but such a scale — half a million lines from one of the most closed AI companies — is truly rare.

For competitors, this is a chance to study Anthropic's real technical solutions. For security researchers — valuable material. For ordinary users — a rare opportunity to look behind the scenes of a tool that many use daily without realizing what's happening inside.

Claude Code competes with Cursor, GitHub Copilot, and Windsurf in the rapidly growing segment of AI assistants for developers. If the constantly active background agent is released in final form, it could become a significant differentiator: most competitors still adhere to a reactive model. The Tamagotchi pet, in turn, indicates Anthropic's attempt to add an element of personal attachment to the tool — to transform a utility into something users return to not only for results.