continuous-learning-v2

Continuous learning from Claude Code sessions is automated by observing session hooks, detecting code patterns, and creating instincts with confidence scoring. These instincts are then evolved into reusable skills, commands, or agents for development workflows.

Evolving code patterns into full Claude Code skills involves detecting recurring project actions, scoring their confidence as instincts, and promoting them into standalone skills, commands, or agents based on usage frequency and project context.

Yes, you need Python, git, and Claude Code installed to detect code patterns. Python and git provide the environment for session observation, hook execution, and instinct evolution within your development workflows.

Yes, you can share code instincts globally. The system keeps instincts project-scoped by default to prevent cross-project contamination, but you can promote project instincts to a global scope to share useful patterns across your projects.

You manage detected code instincts via command line using the `clinstinct` command. This CLI provides commands to check instinct status, trigger evolution into skills, and promote instincts between project and global scopes.

You should avoid automated instinct evolution in development workflows where code patterns are entirely unique or non-repeating. Since the mechanism relies on confidence scoring from recurring session observations, one-off tasks will not generate useful instincts.