memory-systems
Design AI memory architectures with layered persistence
Également disponible depuis: muratcankoylan,Asmayaseen,ChakshuGautam
AI agents lose all context when sessions end without proper memory systems. This skill provides battle-tested patterns for building layered memory from working context to temporal knowledge graphs.
Télécharger le ZIP du skill
Importer dans Claude
Allez dans Paramètres → Capacités → Skills → Importer un skill
Activez et commencez Ă utiliser
Tester
Utilisation de "memory-systems". Design a file-system memory structure for a writing assistant that tracks user documents, writing preferences, and revision history
RĂ©sultat attendu:
- memory/
- ├── users/
- │ └── {user_id}/
- │ ├── profile.json (preferences, style settings)
- │ ├── documents/
- │ │ └── {doc_id}/
- │ │ ├── current.md
- │ │ └── revisions/
- │ │ └── {timestamp}.md
- │ └── vocabulary.json (custom terms, frequently used phrases)
- └── indexes/
- └── user-documents.json (doc_id to user mapping)
Utilisation de "memory-systems". Query: Find all team members with Python skills who are available for tasks starting next week
RĂ©sultat attendu:
- Entity Query Result:
- - Alice Chen (Senior Developer) - Python, Django, FastAPI - Available from 2024-02-05
- - Marcus Rodriguez (Backend Engineer) - Python, PostgreSQL - Available from 2024-02-06
- - Sarah Kim (Full Stack) - Python, React, TypeScript - Available immediately
Audit de sécurité
SûrEducational documentation skill for AI memory architecture design. All 28 static analysis findings are false positives from pattern matching on documentation text. The skill contains no executable code, no network calls, no command execution, and no cryptographic operations. It provides conceptual guidance on memory layers, knowledge graphs, and retrieval patterns.
Score de qualité
Ce que vous pouvez construire
Customer Support Agent with Memory
Build a support agent that remembers customer preferences, past tickets, and product purchases across sessions using entity memory and temporal tracking
Research Assistant with Knowledge Accumulation
Create a research agent that builds a growing knowledge graph of papers, concepts, and relationships, enabling complex cross-domain reasoning queries
Personal AI Assistant with Long-Term Context
Develop a personal assistant that learns user preferences, tracks ongoing projects, and maintains consistent entity references across months of interactions
Essayez ces prompts
I need to add memory to my AI agent that persists across sessions. The agent helps users track their fitness goals and workout history. Design a two-layer memory system using file-system storage that tracks: (1) user profiles with goals, (2) workout session history with dates and exercises. Show the directory structure and JSON schemas for each layer.
My agent needs semantic search over 10,000+ customer support tickets while filtering by product category, date range, and resolution status. Design a vector RAG system with metadata filtering. Specify the embedding strategy, metadata schema, and query format that combines semantic similarity with metadata filters.
Build an entity memory system for a project management agent. The agent must track: team members (with roles and skills), projects (with deadlines and status), and tasks (with assignees and dependencies). Design the entity schema, relationship types, and query functions to answer: 'Who is available for a React task next week?' and 'What projects depend on the authentication refactor?'
Design a temporal knowledge graph for a healthcare agent that tracks patient conditions, medications, and lab results with validity periods. Each fact must support time-travel queries. Provide: (1) the triple schema with valid_from and valid_until timestamps, (2) Cypher queries for 'What medications was patient X on in March 2024?' and 'When did diagnosis Y become active?', (3) consolidation logic for updating facts when new lab results arrive.
Bonnes pratiques
- Match memory architecture to query complexity: file-system for simple persistence, vector RAG for semantic search, knowledge graphs for relationship reasoning, temporal graphs for time-aware queries
- Implement periodic memory consolidation to merge redundant facts, archive obsolete data, and rebuild indexes before retrieval performance degrades
- Use just-in-time memory loading with strategic context injection rather than stuffing all memories into context at once
Éviter
- Storing all memories in the context window - causes context bloat, increases latency, and wastes attention on irrelevant information
- Using vector-only storage for relationship-heavy data - loses entity connections and prevents graph traversal queries
- Never consolidating memories - leads to unbounded growth, duplicate facts, and contradictory information from different time periods