from yesterday Surprise leak of Anthropic’s Claude Code source code revealed a lot about the vibration-coding scaffolding the company has built around its patented Claude model. But observers investigating more than 512,000 lines of code in more than 2,000 files have also discovered references to disabled, hidden or inactive features that provide a glimpse into the potential roadmap for future features.
The main of these characteristics is Kairosa persistent daemon that can operate in the background even when the Claude Code terminal window is closed. The system would use newspapers “
Kairos uses a file-based “memory system” designed to allow persistent operation between user sessions. a notice Hidden behind a disabled “KAIROS” flag in the code, it is explained that the system is designed to “have a complete picture of who the user is, how they would like to collaborate with you, what behaviors to avoid or repeat, and the context behind the work the user provides you.”
To organize and consolidate this memory system between sessions, the Claude Code source code includes references to an AutoDream system with an evocative name. When a user becomes inactive or manually tells Anthropic to sleep At the end of a session, the AutoDream system Tell Claude Code that “you are realizing a dream: a reflective pass over your memory files.”
This message describing this AI “sleep” process asks Claude Code to scan the day’s transcripts for “new information worth preserving,” to consolidate that new information in a way that avoids “near duplicates” and “contradictions,” and to remove existing memories that are too detailed or recently outdated. Claude Code would also be asked to keep an eye out for “existing memories gone astray,” a problem we’ve seen before. when Claude users attempted to graft memory systems into their harnesses. The overall goal would be to “synthesize what you’ve recently learned into lasting, well-organized memories so that future sessions can be quickly oriented,” according to the message.
