memvid-core

name	memvid-core
description	This skill should be used when the user asks to "remember this", "recall", "what did I say about", "note that", "for future reference", "store memory", "my preferences", "forget", or discusses memory storage and retrieval.
version	1.0.0

Memvid Memory Skill

This skill enables natural memory storage and retrieval using a dual-memory architecture.

Memory Scopes

Project Memory (.memvid_data/)

Stores project-specific knowledge:

• Code architecture decisions: "This project uses FastAPI for APIs"
• Bug reports: "Bug in auth.py line 42"
• Technical implementation: "We use MemvidMemory class for storage"
• Project configuration: "Dependencies managed via pyproject.toml"

User Memory (~/memvid_data/)

Stores personal preferences:

• Coding preferences: "I prefer pytest over unittest"
• Style preferences: "I like clean code with type hints"
• Tool preferences: "I use VS Code for development"
• General knowledge: "I always document functions with docstrings"

Automatic Classification

The system uses semantic similarity to auto-classify memories:

• Compares content against project and user examples
• Assigns to scope with higher similarity
• Falls back to user memory if confidence < 0.65

Commands

memvid:store

Store a new memory with automatic scope detection.

Trigger patterns:

• "remember that this project uses FastAPI"
• "note that I prefer pytest"
• "for future reference: bug in line 42"

Action: Call memvid_store tool with:

{
  "content": "<extracted content>",
  "scope": "auto"
}

Response format: "Stored in [project/user] memory: [brief summary]. (ID: abc-123)"

memvid:recall

Search both memories and return merged results.

Trigger patterns:

• "what did I say about testing?"
• "recall memories about authentication"
• "do you remember my coding preferences?"

Action: Call memvid_query tool with:

{
  "query": "<extracted query>",
  "limit": 10
}

Response format: Present results naturally: "Based on your memories:

• [Project] This project uses FastAPI for REST APIs
• [User] You prefer pytest over unittest"

memvid:forget

Delete a memory by ID.

Trigger patterns:

• "forget memory abc-123"
• "delete that memory about authentication"

Action:

1. If ID provided: Call memvid_delete with ID
2. If description: Call memvid_query first, confirm with user, then delete

memvid:stats

Show statistics for both memory stores.

Trigger patterns:

• "show memory stats"
• "how many memories do I have?"

Action: Call memvid_stats tool and present:

Memory Statistics:
Project Memory (.memvid_data/):
  - Total: 42 memories
  - By sector: episodic (15), semantic (20), procedural (5), emotional (1), reflective (1)

User Memory (~/memvid_data/):
  - Total: 103 memories
  - By sector: semantic (60), procedural (30), reflective (13)

Auto-Trigger Rules

Strong triggers (auto-store):

• "Remember that ..."
• "Note that ..."
• "For future reference: ..."
• "I always ..."
• "This project always ..."

Weak triggers (ask first):

• "I think ..."
• "Maybe ..."

Best Practices

1. Always show classification results: Tell user if memory was stored in project or user scope
2. Natural integration: Don't say "I will call memvid_store" - just say "I'll remember that..."
3. Contextualize recalls: When presenting recalled memories, explain which scope they came from
4. Deduplicate before storing: Check if similar memory exists before storing new one

Error Handling

• If storage fails: "I couldn't store that memory. Error: [details]"
• If no memories found: "I don't have any memories about [topic] yet."
• If ambiguous classification: "I'm not sure if this is project or personal knowledge. Storing in user memory for safety."