The leaked source code of Claude Code has uncovered 44 distinct feature flags, each representing a capability that exists within the codebase but remains inactive in public builds. These features span from minor UI adjustments to ambitious system-level innovations, offering insight into Anthropic‘s evolving vision for its AI assistant.
Architectural Shifts: The KAIROS Daemon
The most prominent feature flag, KAIROS, appears over 150 times in the codebase. It enables a persistent background agent capable of running independently of active terminal sessions. This daemon mode includes an internal subsystem called autoDream, which consolidates memory from previous interactions to maintain contextual awareness across sessions. The implementation suggests a fundamental shift toward autonomous, continuous operation rather than reactive user input.
Developer Engagement: The BUDDY Companion System
The Buddy flag introduces a terminal companion system featuring 18 distinct animated species. These “terminal pets” respond to developer actions like test suite outcomes or build failures, creating an interactive environment that blends utility with engagement. The code includes logic for animation rendering, behavioral states, and interaction hooks, indicating substantial investment in this feature rather than a mere prototype.
Multi-Agent Orchestration: COORDINATOR Mode
Coordinator mode allows a single Claude Code instance to manage multiple worker agents. This system handles task decomposition, sub-task execution, and result aggregation through a structured task graph. The implementation implies a transition from individual assistance to coordinated workflow management, enabling parallel processing of complex development tasks.
Strategic Planning: ULTRAPLAN Sessions
The ULTRAPLAN flag introduces scheduled multi-agent planning sessions. These asynchronous, time-bound collaborations allow multiple AI agents to contribute to strategic decisions without requiring real-time user input. The feature suggests a shift toward proactive architectural decision-making, with outputs taking the form of structured plans rather than direct code execution.
Diverse Feature Categories
The remaining 37 flags fall into recognizable categories: UI enhancements (developer experience), expanded agent capabilities (reasoning modes and file system access), integration points (CI/CD pipelines and cloud platforms), and experimental features with ambiguous purposes. These indicate a layered approach to development, balancing immediate usability with long-term infrastructure goals.
The presence of these flags suggests Anthropic is positioning Claude Code as more than a coding assistant. The combination of persistent agents, parallel execution capabilities, and strategic planning tools points toward an evolving system that could function as an autonomous development team. While many features remain unactivated, their existence in the codebase underscores a deliberate path toward increasingly sophisticated AI collaboration frameworks.
