How does Continuous Learning V2 differ from standard system prompts?

Unlike static prompts, this skill uses deterministic hooks to observe your actual corrections and tool usage in real-time, building a dynamic, weighted model of your preferences that evolves automatically over time.

What is an 'Instinct' in this system?

An instinct is an atomic learned behavior with a confidence score. For example, if you repeatedly correct Claude to use functional components, it creates an instinct for 'prefer-functional-style' which it then applies to future tasks.

Is my source code shared with external servers during the learning process?

No. All observation data, session logs, and learned instincts are stored locally in your ~/.claude/homunculus directory, ensuring your private code remains secure.

Which models are required for the background observation?

The system is optimized to use lightweight models like Claude Haiku for background analysis of your observations, ensuring the learning process is fast and cost-effective.

Does this support multi-device synchronization?

While data is stored locally, the skill includes /instinct-export and /instinct-import commands, allowing you to manually sync your learned behaviors across different machines or team members.

Continuous Learning V2 | Adaptive Claude Code Skill

Continuous Learning V2 is an advanced behavioral adaptation framework for Claude Code that transforms daily interactions into a permanent knowledge base. By utilizing deterministic hooks to monitor tool usage and user corrections, the system creates 'instincts'—atomic, weighted patterns of behavior stored locally on your machine. As these instincts gain confidence through repetition, the system clusters them to automatically generate specialized commands, agents, and custom skills tailored to your specific coding style and project requirements. This ensures Claude becomes more efficient and aligned with your workflow the more you use it, without requiring manual prompt engineering.

Key Features

01Knowledge Portability: Support for importing and exporting learned instincts to share best practices across teams.

021 GitHub stars

03Atomic Instinct Model: Creates granular behavioral patterns with confidence scores ranging from 0.3 to 0.9.

04Automated Evolution: Clusters related instincts into sophisticated custom commands and specialized sub-agents.

05Privacy-Centric Local Storage: Keeps all session observations and learned data strictly on your local hardware.

06Deterministic Observation Hooks: Captures 100% of tool interactions (Pre/Post) for reliable behavior tracking.

Use Cases

01Streamlining complex, multi-step DevOps workflows by evolving repeated command sequences into a single skill.

02Automatically adopting project-specific architecture patterns after observing a few manual refactors.

03Developing personalized 'instincts' for specific library preferences (e.g., favoring Zod for validation over Joi).

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nixdorfer/claudecodetool continuous-learning-v2

For use in Claude.ai and ChatGPT

Download Skill