Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents. v2.1 adds project-scoped instincts to prevent cross-project contamination.
The continuous-learning-v2 skill provides an instinct-based learning framework that observes user sessions and extracts small, reusable behaviors called "instincts." These atomic instincts are confidence-weighted, domain-tagged, and backed by evidence from session observations, enabling them to evolve into full-fledged skills, commands, or agents. By capturing patterns automatically, this skill helps prevent knowledge loss, standardizes best practices across projects, and reduces repetitive manual coding efforts. With the introduction of project-scoped instincts in version 2.1, learned behaviors remain isolated within their relevant projects, minimizing cross-project contamination and ensuring that domain-specific practices remain contextually appropriate.
Activate the skill by enabling hooks in your Claude Code sessions, configuring confidence thresholds, and allowing the observer agent to capture prompts and tool use. Once activated, instincts are automatically generated from session activity.
Yes, instincts can be exported and imported. Version 2.1 supports project-scoped instincts by default, but universal or promoted instincts can be shared globally once observed in multiple projects.
The skill is domain-agnostic but supports tagging instincts by domains such as code-style, testing, git, debugging, workflow, and language-specific conventions like Python or React patterns.
The skill requires a Git repository context to detect project scope and a Claude Code session environment with hooks enabled for session observation.
No, instincts are confidence-weighted (0.3-0.9) to indicate reliability. Lower confidence instincts may require user review before evolution into skills, commands, or agents.
Quick Setup:
.claude/skills/Repository
affaan-m/everything-claude-code