Does this version replace the previous learning system?

Continuous Learning V2 is backwards compatible with V1; it can run alongside existing learned skills while providing a more reliable observation path via system hooks.

How does the confidence scoring work?

Instincts start with lower scores (0.3) and increase as patterns are repeatedly observed or explicitly accepted by the user, while declining if behaviors are corrected, ensuring only high-confidence patterns become core behaviors.

What is the difference between an instinct and a skill?

An instinct is a single, atomic behavioral unit (e.g., 'always use Zod for validation'), while a skill is a mature collection of related instincts clustered together through the evolution process.

Is my code data sent to external servers for learning?

No, all observations and the resulting instinct files are stored locally on your machine. While an LLM like Haiku is used for analysis, the raw session data remains under your control.

Continuous Learning V2

Name: Continuous Learning V2
Author: unju-ai

byunju-ai

0•

Productivity & Workflow

Implements an instinct-based learning system that captures user patterns via hooks to evolve Claude's behavior automatically.

Continuous Learning V2 is a sophisticated framework designed to make Claude Code smarter over time by capturing 'atomic instincts'—small, specific behavioral patterns learned directly from your coding sessions. By utilizing deterministic hooks (PreToolUse/PostToolUse) rather than probabilistic triggers, it ensures 100% reliable observation of your workflows, preferences, and corrections. These instincts are assigned weighted confidence scores and can eventually cluster into full-fledged custom skills, commands, or specialized agents, creating a personalized AI assistant that adapts to your unique development style while keeping all data stored locally for privacy.

Key Features

01Hook-based observation for 100% reliable capture of tool usage and user corrections

02Atomic instinct model with weighted confidence scoring from 0.3 to 0.9

03Automated evolution engine that clusters instincts into specialized commands and agents

04Privacy-first local storage for all session observations and learned behaviors

05Background observer agent (Haiku) for asynchronous pattern analysis

060 GitHub stars

Use Cases

01Standardizing repetitive manual workflows into reusable automated commands

02Developing specialized 'refactor' or 'testing' agents based on historical successful patterns

03Automatically learning project-specific coding styles and architecture preferences from user corrections

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add unju-ai/ecc continuous-learning-v2

For use in Claude.ai and ChatGPT