Memory Sync Skill

How to manually sync learnings to memory systems after tasks.

Dual Memory Architecture

Ariff uses two complementary memory systems:

• mem0-memory-mcp - AI-native semantic memory for quick recall
• obsidian-memory - Structured vault for long-term knowledge

Manual Sync Commands

Quick Save to mem0

Use mcp__mem0-memory-mcp__add_memory to store:
- Task context and solution
- Key learnings and gotchas
- Useful commands discovered

Save to Obsidian

Use mcp__obsidian-memory__create_note with:
- path: "tasks/YYYY-MM-DD-task-slug.md"
- content: Structured task notes

Memory Entry Template

## Task: [Brief Title]
Date: [YYYY-MM-DD]
Device: [hostname]

### Problem
[What was the issue/request]

### Solution
[How it was resolved]

### Key Learnings
- [Learning 1]
- [Learning 2]

### Related
- Previous tasks: [links]
- Documentation: [links]

Creating Agent Instructions

When a mistake or pattern should be avoided in future:

1. Create instruction file at:

   Ariff-code-config/instructions/[category]-[topic].instructions.md
   

2. Use format:

   ---
   description: [What this instruction prevents/ensures]
   applyTo: [glob pattern like **/*.py or **/canvas/**]
   ---
   
   # [Rule Title]
   
   ## Context
   [When this applies]
   
   ## Rule
   [What to do/not do]
   
   ## Example
   [Good vs bad example]
   

3. Add to 00_INSTRUCTIONS_INDEX.md if significant

Reconciliation

If memories seem out of sync:

1. Query both systems for topic
2. Compare entries
3. Update older system with newer info
4. Remove duplicates with less detail

When to Save

Save memory when:

• ✅ Task completed successfully
• ✅ Discovered non-obvious solution
• ✅ Made mistake worth remembering
• ✅ Found useful tool/command
• ✅ Clarified user preference

Skip saving when:

• ❌ Trivial task (simple file edit)
• ❌ Already documented elsewhere
• ❌ One-off unique situation