continuous-learning-v2

Instinct-based learning uses hooks to observe interactive sessions, capturing atomic instincts with confidence scoring and metadata like trigger, domain, and source. These small learning signals are stored as reusable building blocks for future automation.

You evolve captured instincts through an evolution pipeline that clusters related atomic instincts into higher-level skills, commands, and agents. This creates scalable automation capabilities from observed session learning signals.

Yes, the system supports local-first observation with export and import functionality for instincts. You can share instinct data across environments while maintaining on-device session capture and confidence-weighted knowledge building.

Enable the observer hook, create the instincts directory, and start the agent to begin capturing atomic instincts. The system automatically models each instinct with id, trigger, confidence, domain, and source for evolving into reusable skills.

No external dependencies are required. The system is self-contained, relying on hook-based session observation and local instinct storage to capture learning signals and evolve them into skills, commands, and agents.

Instinct-based learning is designed for interactive work sessions where observation and confidence scoring are meaningful. It is not suited for one-off tasks or environments where hook-based session observation cannot be enabled.