An advanced learning system that transforms AI coding sessions into reusable knowledge through atomic instincts and confidence scoring.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install lobster-continuous-learning-v2
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install lobster-continuous-learning-v2 using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Continuous Learning v2.1: Instinct-Based Architecture?
Continuous Learning v2.1 is a sophisticated framework designed to turn every interaction with your AI coding agent into a permanent asset. By utilizing a hook-based architecture, it observes sessions with 100% reliability to extract atomic behaviors known as instincts. These instincts represent small, learned behaviors—such as a preference for functional programming or specific error-handling patterns—that are assigned confidence scores based on frequency and user feedback. This system ensures that your AI becomes more personalized and efficient the more you use it, making it a vital component for power users of Openclaw Skills.
The v2.1 update introduces project-scoped learning, which effectively isolates knowledge to prevent cross-project contamination. For instance, React-specific patterns remain within your React repositories, while universal best practices like input validation can be promoted to a global scope. This architecture allows for a seamless evolution from raw observations to structured skills, commands, and specialized agents.
Continuous Learning v2.1: Instinct-Based Architecture Use Cases
- Automatically extracting reusable coding patterns from active development sessions.
- Managing distinct architectural conventions across different technology stacks.
- Evolving repeated manual workflows into automated skills and commands.
- Scaling personal coding preferences across multiple projects through instinct promotion.
- Reducing AI hallucination by providing high-confidence, project-specific evidence.
How Continuous Learning v2.1: Instinct-Based Architecture Works
- Deterministic hooks (PreToolUse and PostToolUse) capture all prompts, tool calls, and outcomes during a session.
- The system identifies the project context using git metadata or environment variables to ensure data isolation.
- A background observer agent analyzes the captured logs to detect patterns, user corrections, and successful error resolutions.
- Identified behaviors are saved as atomic YAML files (instincts) with metadata including domain, confidence score, and evidence links.
- Users interact with the CLI to review, evolve, or promote these instincts into full-scale skills or global behaviors.
Continuous Learning v2.1: Instinct-Based Architecture Setup
To integrate this into your Openclaw Skills environment, first initialize the directory structure:
mkdir -p ~/.claude/homunculus/{instincts/{personal,inherited},evolved/{agents,skills,commands},projects}
Next, configure your agent's settings.json to include the observation hooks. Add the following to your hooks configuration:
{
"hooks": {
"PreToolUse": [{
"matcher": "*",
"hooks": [{
"type": "command",
"command": "~/.claude/skills/continuous-learning-v2/hooks/observe.sh"
}]
}],
"PostToolUse": [{
"matcher": "*",
"hooks": [{
"type": "command",
"command": "~/.claude/skills/continuous-learning-v2/hooks/observe.sh"
}]
}]
}
}
Continuous Learning v2.1: Instinct-Based Architecture Data Schema & Taxonomy
The system organizes data using a hierarchical structure to balance global knowledge with project specificities:
| Component | Location | Purpose |
|---|---|---|
projects.json |
~/.claude/homunculus/ |
Maps project hashes to paths and metadata. |
observations.jsonl |
projects/<hash>/ |
Per-project logs of all interactions. |
instincts/ |
projects/<hash>/instincts/ |
YAML files defining project-specific atomic behaviors. |
evolved/ |
~/.claude/homunculus/evolved/ |
Global skills, agents, and commands derived from instincts. |
Continuous Learning v2.1: Instinct-Based Architecture Advanced Features
- Project-scoped isolation using git remote hashing for portable project IDs.
- Confidence-weighted logic (0.3 to 0.9) that adjusts based on reinforcement or corrections.
- Automatic promotion logic that identifies high-confidence patterns recurring across multiple projects.
- Evolution pipeline that clusters related instincts into complex skills or specialized agent personas.
- Background observer mode with configurable intervals and analysis thresholds.
SKILL.md
Loading
Related Openclaw Skills
VisNote Images Creator
An AI-driven automation skill for generating stylized XiaoHongShu social media images using template-based browser rendering.
HostLink
HostLink enables secure, authenticated execution of host-level commands directly from within an OpenClaw container.
Video Narration
An automated workflow that transforms silent screen recordings into polished, narrated product demos using neural text-to-speech.
Daily Poem
A literary AI assistant that delivers curated daily poems, translations, and technical rhythm analysis across Chinese and English traditions.
Novel Writer
A sophisticated prompt construction engine that synthesizes chapter outlines, character dossiers, and voice profiles into actionable instructions for AI writing agents.
WeChat Cover Image Generator
A professional-grade utility for creating perfectly sized, text-free WeChat official account cover images using AI generation and automated post-processing.
Featured*
