continuous-learning-v2

Convert repeated Claude Code tool behaviors into confidence-scored instinct files.

Updated May 4, 2026
One-click install
npx skills add https://github.com/gganbukim1/myskills --skill continuous-learning-v2-gganbukim1
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: continuous-learning-v2
Source: https://github.com/gganbukim1/myskills/tree/main/continuous-learning-v2
Command: npx skills add https://github.com/gganbukim1/myskills --skill continuous-learning-v2-gganbukim1

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires git, python3, shasum, sha256sum, and includes scripts (resource) components.

What problem does it solve?

This Skill eliminates the manual effort of repeatedly teaching your AI the same patterns by automatically observing Claude Code sessions and distilling them into reusable, confidence-scored instincts. It also prevents cross-project contamination by defaulting learned behaviors to the project where they were observed.

Core Features & Use Cases

  • Deterministic session observation via hooks: Captures tool use events 100% of the time using PreToolUse/PostToolUse hooks.
  • Atomic, confidence-scored “instincts”: Learns small behaviors with weighted confidence and evidence tracking.
  • Project-scoped learning (v2.1): Stores project-specific instincts under projects/<hash>/ to isolate React/Python/etc. conventions.
  • Global promotion when patterns generalize: Automatically or manually promotes project instincts to global scope when they appear confidently across multiple projects.
  • Workflow evolution pipeline: Clusters instincts into commands/skills/agents using an evolve step.

Quick Start

Tell your AI to run continuous learning by enabling the PreToolUse and PostToolUse hooks to execute observe.sh for this skill.

Frequently Asked Questions about continuous-learning-v2

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

FAQPage Schema
How do I capture Claude Code tool behaviors automatically during git development?

You can capture Claude Code tool behaviors deterministically by enabling PreToolUse and PostToolUse hooks to execute an observation script, which records tool use events into local instinct files.

What is the best way to keep AI learned behaviors isolated per git repository?

Project-scoped learning isolates AI behaviors by storing confidence-scored instincts under a hashed project directory, preventing cross-project contamination like mixing React and Python conventions.

How does confidence scoring work for AI instinct extraction from tool sessions?

Confidence scoring applies weighted values and evidence tracking to small observed behaviors, ensuring instincts are only promoted to global patterns when they appear reliably across multiple projects.

Do I need Python3 and git to use project-scoped continuous learning?

Yes, this workflow requires git for repository detection via remote hashing and Python3 to run the CLI-driven status, evolve, export, import, and promote operations for instinct files.

Can I promote project-specific AI instincts to global patterns across multiple repositories?

Yes, project-specific instincts can be promoted automatically or manually to a global scope when those learned behaviors demonstrate high confidence and appear consistently across multiple projects.

How do I evolve learned instincts into reusable commands or agents?

You can run the CLI-driven evolve step to cluster your accumulated YAML-frontmatter instincts into higher-level reusable commands, skills, or agents for future sessions.