continuous-learning-v2

You can capture reusable patterns from Claude Code sessions by using hooks to observe interactions and automatically save them as atomic instincts with confidence scores. These instincts are stored under ~/.claude/homunculus for later evaluation.

Instinct-based learning is a process that observes your coding sessions, captures recurring patterns as atomic instincts with confidence levels, and evolves them into scalable skills, commands, and agents to improve your software engineering workflow.

To prevent cross-project pattern contamination, you can use project-scoped instincts to isolate learning by repository. This ensures patterns captured in one software project do not affect the automation workflows of another.

Yes, you can promote project-specific instincts to global skills. When cross-project consensus is achieved, the system supports automated promotion of these project instincts to a global scope for broader reuse.

You can convert session-based patterns into Claude Code agents by recording project-scoped and global instincts with tracked confidence scores. Once validated, these instincts evolve into functional commands and agents.

Yes, instinct-based learning works without external dependencies. It operates autonomously using internal scripts to track confidence and scope, storing all captured patterns locally under the ~/.claude/homunculus directory.