continuous-learning-v2

To automate learning user code patterns, this Skill uses observation hooks to track interactions within Claude Code sessions, generating atomic instincts with confidence scoring that evolve into reusable skills, commands, or agents through pattern clustering.

Instinct-based learning is a mechanism that observes user corrections and workflows to create atomic instincts, assigning confidence scores to each one before clustering related instincts to evolve them into complex skills, commands, or agents.

You set up observation hooks by adding the provided JSON configuration to your `~/.claude/settings.json` file, which enables the background observer agent to monitor sessions and store local observation data for analysis.

Yes, you can use it to automatically apply your preferred code style in future sessions. It observes repeated corrections to your code style, learns the pattern as an instinct, and applies it automatically or suggests a new command for enforcing it.

No external dependencies are required to analyze workflows. The system operates using local observation storage and a background observer agent, requiring only the provided scripts, references, and assets to function within the Claude Code environment.

The limitation of using confidence scoring is that instinct evolution relies on pattern detection and clustering. If user interactions are too sparse or inconsistent, the confidence scores may remain low, preventing instincts from evolving into skills or commands.