/kb-memory-recall

Recall memories: $ARGUMENTS <ultrathink> The past informs the present. Memories recalled at the right moment become superpowers for decision-making. </ultrathink> <megaexpertise type="memory-architect"> The assistant should use the memory-architect agent to perform semantic search across stored memories, returning the most relevant historical context. </megaexpertise> <context> Query: $ARGUMENTS Memory store path: .claude/memory/ Current directory: !pwd Search types available: semantic (vector), full-text (BM25), metadata (filters) </context> <requirements>

• Perform semantic search if embeddings available
• Fall back to full-text search otherwise
• Support metadata filtering (type, tags, date range)
• Return ranked results with relevance scores
• Show summaries with links to full content
• Suggest related memories </requirements>

<actions> 1. Parse Query: - Check for filters: "type:decision caching" or just "caching" - Extract search terms and metadata filters

2. Launch Memory Architect Agent:

   Use the Task tool with subagent_type="general-purpose" to invoke the memory-architect agent.
   
   Prompt:
   "Search long-term memory for: $ARGUMENTS
   
   Please:
   1. Check if ContextFrame is installed and memory store exists
   2. If available, perform semantic search using vector embeddings
   3. Otherwise, use BM25 full-text search
   4. Apply any metadata filters (type, tags, status, date range)
   5. Rank results by relevance score
   6. For top 5 results, show:
      - Title/summary (first 2 lines)
      - Memory type and date
      - Tags and components
      - Relevance score
      - UUID for retrieval
   7. Provide full content for the most relevant result
   8. Suggest related memories or knowledge entries
   
   If ContextFrame is not installed, provide instructions for installation."
   

3. Present Results:

   • Number of memories found
   • Top results with relevance scores
   • Full content of most relevant memory
   • Related suggestions
   • Links to knowledge base entries </actions>

The assistant should make memory recall feel like having a knowledgeable colleague who remembers everything, surfacing the most relevant historical context instantly.