continuous-learning-v2

Transforms Claude Code session observations into atomic, confidence-scored learning instincts.

Updated Mar 24, 2026

Search Tags:#automation #learning #claude-code #promotion #haiku #instincts #project-scoped

Ready to run

npx skills add https://github.com/Oruga420/claude-code-skills --skill continuous-learning-v2-oruga420

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: continuous-learning-v2
Source: https://github.com/Oruga420/claude-code-skills/tree/main/continuous-learning-v2
Command: npx skills add https://github.com/Oruga420/claude-code-skills --skill continuous-learning-v2-oruga420

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

Transform Claude Code sessions into atomic, confidence-weighted instincts that evolve into skills, commands, and agents, reducing repetitive decision-making and accelerating learning.

Core Features & Use Cases

Atomic instincts with confidence scoring that evolve into skills, commands, or agents.
Project-scoped instincts with global fallback, enabling isolated learning per repository and cross-project promotion.
Automatic detection of current project context from environment or Git, with hooks feeding observations to evolve patterns over time.
Optional CLI and observer components to inspect, evolve, and promote instincts across projects.

Quick Start

Install and enable the project-scoped observer to start harvesting instincts from Claude Code sessions.

Frequently Asked Questions about continuous-learning-v2

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate learning from Claude Code sessions to avoid repetitive decisions?▼

Claude Code session learning works by capturing session observations through a local Haiku-driven observer, which stores them as atomic, confidence-scored instincts that evolve into reusable skills, commands, or agents to reduce repetitive decision-making.

What are confidence-scored instincts and how do they evolve into reusable code?▼

Confidence-scored instincts are atomic learning patterns captured from your sessions. As their confidence scores increase through repeated observations, they automatically promote into standalone skills, commands, or agents for future use.

Can I scope automated learning to a specific project and share patterns across repositories?▼

Project-scoped learning isolates instincts per repository using environment or Git context, while a global fallback mechanism allows cross-project promotion of high-confidence instincts to reuse patterns across multiple projects.

How do I set up a Haiku-driven observer to capture Claude Code session observations?▼

Enable the project-scoped observer component to start harvesting observations from your sessions. The observer requires local storage under ~/.claude/homunculus to detect, store, and promote instincts with trigger, domain, and scope metadata.

Does continuous learning require any specific dependencies or external APIs to function?▼

No external dependencies are required. The Skill relies on a local Haiku-driven observer and local storage under ~/.claude/homunculus to manage the instinct lifecycle without needing external packages or APIs.

When should I avoid using automated instinct promotion for session learning?▼

You should avoid automated instinct promotion if you lack a local Haiku observer or cannot allocate storage under ~/.claude/homunculus, as these are strictly required to detect, store, and evolve session patterns.