continuous-learning-v2

You turn Claude Code sessions into reusable instincts by activating automated observation to extract atomic patterns, storing them locally with YAML frontmatter that includes confidence scoring and domain scope.

Atomic instincts with confidence scoring are structured patterns extracted from coding sessions, tracked with a numerical confidence value to determine when they mature into reusable skills, commands, or agents.

You isolate learning per repository by applying project-scoped instincts, which restrict extracted observations and evolution workflows to the local project boundary instead of global scope.

Yes, you need YAML frontmatter in instinct files to manage instincts and evolution workflows, requiring specific fields including id, trigger, confidence, domain, and scope.

Yes, you can automate the evolution of instincts into skills and commands through promotion workflows that evaluate confidence scores to mature and promote atomic observations into team resources.

Project scope confines instinct storage and application to a single repository to prevent cross-project contamination, while global scope allows instincts to apply across multiple projects and sessions.