Memory Consolidation Skill - Pattern Extraction &

Tinker

Home Epics Memory Consolidation Skill - Pattern Extraction & DeduplicationEdit

Memory Consolidation Skill - Pattern Extraction & Deduplication

Cancel

Title *

Project *

Description

**Problem:**
The research analysis of memories revealed significant redundancy and lost opportunity for knowledge extraction:

1. **Duplicate patterns across 6+ memories**: Scope creep captured in #13, #17, #18, #19; MCP registration in #9, #39, #42, #43
2. **Specific PR reviews** contain valuable technical patterns but are stored as historical summaries instead of extractable patterns
3. **No systematic process** for consolidating while preserving data
4. **Manual consolidation is error-prone** and loses context

**What We're Losing:**
- Systematic UI rollout pattern (8 successful phases from #22-32)
- Authorization guard patterns from security reviews
- Architectural patterns from terminal streaming work
- Recurring failure patterns that indicate systemic issues

**Solution:**
Build a **memory consolidation skill** that can:
1. Cluster semantically similar memories (using embeddings/similarity)
2. Extract higher-level patterns from specific instances
3. Create consolidated "pattern memories" from grouped content
4. Create a **memory_cleanup proposal** to delete the original redundant memories
5. Human reviews and approves the proposal before deletion

**Workflow:**
```
Researcher identifies duplicates
    → Creates consolidated memory
    → Creates memory_cleanup proposal with IDs to delete
    → Human verifies
    → Human approves
    → Originals deleted
```

This uses the existing proposal system for safety - no risky batch deletion tool needed.