What is the benefit of using hooks over standard skills for observation?

While skills are probabilistic and might not always trigger, hooks are deterministic and fire 100% of the time, ensuring that no important user corrections or patterns are missed during a session.

How does project scoping prevent contamination?

The system automatically detects your project using git remote URLs or repository paths. This ensures that Python-specific conventions aren't mistakenly applied when you are working in a React or JavaScript environment.

What are 'instincts' in this context?

Instincts are small, atomic learned behaviors—such as a preference for functional programming—that are assigned a confidence score and can be scoped either to a specific project or globally.

Is my code or conversation data shared externally?

No. All observations and learned instincts stay local on your machine within the ~/.claude/homunculus/ directory. You have total control over what is exported or promoted.

How do instincts eventually become full skills?

Through the /evolve command, the system analyzes clusters of related high-confidence instincts and suggests promoting them into formal skills, reusable commands, or specialized agents.

Continuous Learning v2 (Instinct Architecture)

Name: Continuous Learning v2 (Instinct Architecture)
Author: oabdelmaksoud

byoabdelmaksoud

0•

Productividad y Flujo de Trabajo

Automates the capture and evolution of coding patterns and preferences into reusable, confidence-weighted instincts and skills.

Continuous Learning v2 is an advanced instinct-based architecture that transforms your Claude Code sessions into a structured knowledge base. By utilizing deterministic lifecycle hooks, it observes every tool interaction to distill atomic behaviors—called 'instincts'—which are weighted by confidence and context. Version 2.1 introduces project-scoped isolation, ensuring that specific framework conventions (like React patterns) remain within relevant projects while universal best practices (like security validation) are shared globally, allowing your AI assistant to grow more specialized and efficient the more you use it.

Características Principales

010 GitHub stars

02Deterministic observation via PreToolUse and PostToolUse hooks for 100% reliable data capture.

03Atomic instinct model with weighted confidence scoring (0.3 to 0.9) to track behavioral certainty.

04Project-scoped isolation using git-based detection to prevent cross-language or framework contamination.

05Intelligent promotion system that upgrades recurring project patterns to global status based on usage frequency.

06Automated evolution of clustered instincts into full-fledged skills, custom commands, and specialized agents.

Casos de Uso

01Capturing and automating frequent workflow patterns like specific git commit styles or testing procedures.

02Onboarding Claude to complex project-specific conventions by importing pre-existing instinct libraries.

03Maintaining strict architectural consistency across multiple repositories without manual prompt engineering.

Características Principales

010 GitHub stars

02Deterministic observation via PreToolUse and PostToolUse hooks for 100% reliable data capture.

03Atomic instinct model with weighted confidence scoring (0.3 to 0.9) to track behavioral certainty.

04Project-scoped isolation using git-based detection to prevent cross-language or framework contamination.

05Intelligent promotion system that upgrades recurring project patterns to global status based on usage frequency.

06Automated evolution of clustered instincts into full-fledged skills, custom commands, and specialized agents.

Casos de Uso

01Capturing and automating frequent workflow patterns like specific git commit styles or testing procedures.

02Onboarding Claude to complex project-specific conventions by importing pre-existing instinct libraries.

03Maintaining strict architectural consistency across multiple repositories without manual prompt engineering.