continuous-learning-v2

Captures Claude Code session activity via hooks and evolves observed patterns into reusable instincts, skills, and commands.

Updated Mar 25, 2026
One-click install
npx skills add https://github.com/Femad-6/my-skills --skill continuous-learning-v2-femad-6
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: continuous-learning-v2
Source: https://github.com/Femad-6/my-skills/tree/main/.github/skills/continuous-learning-v2
Command: npx skills add https://github.com/Femad-6/my-skills --skill continuous-learning-v2-femad-6

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Claude Code sessions contain valuable patterns—user corrections, error resolutions, repeated workflows—that are lost when the session ends. This Skill observes sessions deterministically via PreToolUse/PostToolUse hooks, distills them into atomic confidence-scored instincts, and evolves those instincts into reusable skills, commands, and agents, with project-scoped storage to prevent cross-project contamination. ## Core Features & Use Cases - Hook-Based Observation: PreToolUse/PostToolUse hooks capture 100% of tool calls into project-scoped observations.jsonl files, with secret scrubbing and automated-session filtering. - Instinct Management CLI: The instinct-cli.py script provides status, import, export, evolve, promote, projects, and prune commands for managing instincts with confidence scoring (0.3-0.9) and domain tagging. - Project Scoping & Promotion: Projects are detected via git remote URL hashing; instincts default to project scope and can be promoted to global when seen in 2+ projects with confidence >= 0.8. - Background Observer Agent: A Haiku-based observer loop analyzes accumulated observations on an interval, guarded by active-hours, cooldown, and idle-detection gates. - Use Case: A developer notices Claude repeatedly corrected on React component style; after several sessions, the system auto-creates a project-scoped instinct preferring functional components, which later evolves into a team-shareable skill. ## Quick Start Enable the observation hooks in your Claude Code settings and ask the assistant to show your learned instincts using the instinct status command.

Frequently Asked Questions about continuous-learning-v2

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

FAQPage Schema
How do I set up automatic learning from Claude Code sessions?

Register the observe.sh script as PreToolUse and PostToolUse hooks in your Claude Code settings.json, pointing to the skill's hooks directory. The hooks then capture every tool call into project-scoped observations files automatically.

How do I view and manage learned instincts?

Run the instinct-cli.py status command to list project and global instincts with confidence bars grouped by domain. Additional commands support export, import, evolve, promote, projects, and prune operations.

Does continuous learning work across multiple projects?

Yes, projects are detected via git remote URL hashing and each gets isolated instinct storage under ~/.claude/homunculus/projects/. Instincts seen in 2+ projects with confidence >= 0.8 can be promoted to global scope.

Is my code or conversation data shared externally?

No, observations stay local on your machine and only distilled instinct patterns can be exported, never raw observations or code. The observation hook also scrubs common secret patterns like API keys and tokens before persisting.

Why does the background observer not run on my machine?

The observer is disabled by default; set observer.enabled to true in config.json. It also skips cycles outside configured active hours, during cooldown periods, when the user is idle, or on Windows unless explicitly overridden.