🔥 MemSkill System

Dynamic Memory Strategy Selection for AI Agents — Stop Forgetting, Start Evolving.

MemSkill is a Skill-Content decoupled memory architecture that lets AI agents dynamically select the best memory strategy based on context. Instead of one-size-fits-all memory, MemSkill uses Top-K retrieval with dual embedding (retrieval + state encoding) to find what's relevant — and evolves its strategies over time through failure-driven feedback.

---

✨ Features

• Skill-Content Decoupling — Memory skills (how to remember) separated from memory content (what to remember)
• Top-K Dynamic Strategy Selection — Picks the right memory strategy from 20+ skill slots based on context
• Failure-Driven Evolution — Designer component monitors failure patterns, auto-generates new strategies
• Dual Embedding — Retrieval embedding + state encoding in parallel for richer matching
• 5000-capacity MemoryBank — With automatic promotion (short→long→core)

📦 Installation

git clone https://github.com/Youan-ai/MemSkill-System.git
cd MemSkill-System
# No pip package yet — drop the code into your project

🚀 Quick Start

from memskill import MemSkillSystem

# Initialize with default memory strategies
ms = MemSkillSystem()

# Store a memory — auto-selects best strategy
ms.remember(key="user_preference", value="钟明 prefers minimal UI", importance=0.9)

# Retrieve with context matching
result = ms.recall("What does the user like?")
print(result)  # "钟明 prefers minimal UI"

# Check current strategies in use
strategies = ms.get_active_strategies()
print(f"Active: {strategies}")

🏗️ Architecture

┌─────────────────────────────────────────────────┐
│                  MemSkillSystem                   │
├─────────────┬──────────────┬──────────────────────┤
│ MemoryBank  │ SkillPool    │ Designer             │
│ (5000 cap)  │ (20+ slots)  │ (Failure Analyst)    │
├─────────────┼──────────────┼──────────────────────┤
│ short→long→ │ retrieval +  │ monitors errors →    │
│ core promote │ state dual   │ auto-evolve skills   │
│             │ embedding    │                      │
└─────────────┴──────────────┴──────────────────────┘

📚 API Overview

Method	Description
remember(key, value, importance)	Store with auto-strategy
recall(query, k=5)	Retrieve top-K relevant memories
get_active_strategies()	List current memory strategies
evolve(feedback)	Trigger skill evolution from feedback

🤝 Contributing

PRs welcome! For major changes, open an issue first to discuss.

📄 License

Apache 2.0 — see LICENSE.