Claude Just Leaked 5 HUGE Features (Kairos, UltraPlan...)
Anthropic's Claude Code has experienced a massive leak revealing 5 major unreleased features including Kairos (an always-on monitoring agent), Ultra Plan (30-minute deep thinking mode), a gamified buddy system, Dream memory management, and an undercover mode that hides AI contributions.
Summary
A significant leak from Anthropic's Claude Code repository has exposed over 500,000 lines of code across nearly 2,000 files, revealing five major unreleased features. Kairos represents a paradigm shift toward 'always-on agents' - functioning as a cron task with memory that continuously monitors projects, detects bugs, and proactively creates pull requests while users sleep, moving away from the current interactive model to autonomous background operation. Ultra Plan introduces a deep thinking feature where Claude can spend 30 minutes in cloud containers performing complex analysis for challenging tasks like database migrations or microservice decomposition, providing much more detailed and specific solutions than quick responses. The buddy system gamifies the coding experience with Tamagotchi-style virtual pets featuring 18 species, rarity systems (legendaries at 1%, commons at 60%), and five stats including debugging ability, patience, chaos, wisdom, and snark - designed to increase user engagement and lock users into Anthropic's ecosystem. The Dream system addresses Claude's biggest current weakness - memory persistence - by implementing a four-phase process (survey, gather, consolidate, prune) that merges duplicate memories, removes stale information, and maintains context across sessions, transforming Claude from a 'new hire every session' into a 'veteran colleague who remembers.' Finally, undercover mode ironically leaked as part of this revelation, designed to hide AI contributions to repositories and make autonomous agent edits appear human-generated.
About this episode
📈 ALL Systems: https://bit.ly/4kol0y5 🎁 Type with your Voice: https://Glaido.com 🎙️ My Goofy pod: https://bit.ly/46nLQ3U 👾 FULL Repo: https://github.com/instructkr/claude-code Anthropic just leaked their own source code. Anthropic accidentally shipped a source map file in their npm package, exposing 512,000 lines of Claude Code's source code. Inside we found 5 unreleased features they weren't ready to announce: KAIROS (always-on background AI), ULTRAPLAN (30-minute deep planning), BUDDY (AI Tamagotchi), DREAM (self-maintaining memory), and UNDERCOVER MODE (AI identity masking for Anthropic employees). Every claim was verified by cloning the repo and reading the actual source code. One missing line in a config file exposed everything.
Key Insights
- The speaker argues that Kairos represents a fundamental shift from interactive AI to autonomous 'always-on agents' that work proactively in the background rather than responding to user prompts
- The speaker claims that Ultra Plan solves the performance trade-off between quick responses and deep analysis by allowing Claude to spend 30 minutes thinking through complex problems in remote cloud containers
- The speaker suggests that Anthropic's memory issue is their largest current problem, with the Dream system designed to transform Claude from feeling like 'a new hire every session' into 'a veteran colleague who remembers' through progressive memory refinement
Topics
Transcript
Claude Code has just had the biggest leak ever, and what I read shocked me. In this video, I'll show you exactly what you need to know, the five things that are going to be relevant for you, what they are, and what we need to do about them now, based on that information. So if you haven't, grab that coffee. Let's go straight into number one, which is something called Kairos, which is a really interesting concept. And if you don't know, this leak itself isn't the model itself, it's the Claude tool that we use. So over 500,000 lines of code, almost 2,000 files, five unreleased features that we're going to cover in this video. And obviously, this…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Jack Roberts
100 hours of Hermes Agent lessons in 23 minutes
Jack walks through advanced features of Hermes Agent, an AI personal assistant, covering memory systems, background tasks, scheduled cron jobs, model switching, and integration with tools like Obsidian, GitHub, and various AI models. The video aims to help users unlock capabilities beyond basic chatbot usage. Key themes include connecting external memory systems, delegating tasks to specialized AI models, and building a persistent, context-aware personal assistant.
Claude Code = $10,000 Beautiful Slides
Jack demonstrates a GitHub-based system using Claude Code to generate professional slide decks from any website URL by codifying 20 universal design principles. The system uses Firecrawl to extract brand DNA from websites and can integrate AI-generated images via APIs like Krea.ai. The result is polished, brand-consistent HTML presentations created in minutes with minimal input.
The most valuable thing I got from working at McDonald's
The speaker reflects on their first job at McDonald's at age 17, earning £5 an hour. The most valuable takeaway was the realization that they never wanted to work such a job again, reinforced by calculating how long it would take to earn a million pounds at that wage.
Claude Code Memory System = CHEAT CODE
Jack Roberts presents a three-tier Claude memory system designed to give AI tools persistent context across all platforms and sessions. The system consists of short-term identity memory, mid-term project memory via structured folders and claude.md files, and long-term memory using either Obsidian or Pinecone for archiving conversations and expert knowledge. The goal is to eliminate the 'amnesia' problem where AI loses context between chats.
ChatGPT ads aren't actually a bad thing. Here's when I want them.
The speaker argues that ads in ChatGPT are not inherently bad, particularly when users are actively seeking product recommendations. They distinguish this as 'pull advertising,' where the user initiates the request and ChatGPT synthesizes results based on provided context.