Claude Code leak exposes a Tamagotchi-style ‘pet’ and an always-on agent
TL;DR
Anthropic accidentally shipped a source map file containing over 512,000 lines of TypeScript code in the Claude Code 2.1.88 update – a classic build-process mistake.
Key Points
- Users on X spotted the leak and spread the code; Ars Technica and VentureBeat were among the first outlets to cover it in detail.
- The leaked code allegedly reveals hidden features: a Tamagotchi-style 'pet' mode and an always-on background agent.
- Anthropic's system prompts for the bot and details about its internal memory architecture also appear to have been exposed.
Nauti's Take
Shipping source maps in a production build is the software equivalent of leaving your house key under the doormat – embarrassing, but fixable. More interesting than the mistake itself are the leaked features: a Tamagotchi-style pet for a coding assistant sounds absurd, but could be a clever gamification strategy to boost developer engagement.
The always-on agent is the truly relevant signal – Anthropic is clearly moving toward autonomous, persistently running AI workflows, which raises new questions about security, cost, and control.
Context
The leak offers a rare, unfiltered look at the internals of a leading AI coding tool – including a product vision Anthropic has not yet officially communicated. The always-on agent in particular signals a strategic shift: from reactive chat tool to persistent AI assistant that autonomously monitors tasks. Leaks like this pressure companies to communicate their roadmap earlier than planned or cancel features before they are ready.
Video
Sources