continuous-learning-v2

Instinct-based learning works by observing your coding sessions via hooks and extracting atomic instincts with confidence scoring. These related instincts are then clustered and evolved automatically into reusable skills, commands, or agents.

Prevent cross-project contamination by using project-scoped instincts to isolate learned patterns within specific project contexts. This isolation ensures that learned behaviors remain targeted to their specific projects without bleeding into others.

Promote project-specific code patterns to a global scope by identifying universal instincts within your isolated project context. The system supports targeted promotion of these universal patterns for reuse across other projects.

You can use session hooks to build reusable skills from any development workflow, including React. The hooks observe your preferred component patterns and state management techniques, automatically creating instincts that evolve into skills.

You need to configure settings.json to start learning from your coding sessions. Add the provided JSON configuration to your ~/.claude/settings.json file to enable the observation hooks required for extracting instincts.

Confidence scoring for instinct development relies on observing repeated tool usage and user corrections to validate patterns. Instincts require sufficient session data to achieve high confidence before they can be clustered and evolved into skills.