Mem Search

Search and browse work memories with local MEMORY_INDEX.md as the primary source.

Execution Steps

1. Read Local Memory Index (Priority)

Read docs/dev/MEMORY_INDEX.md first:

if [ -f docs/dev/MEMORY_INDEX.md ]; then
  # File exists - use as primary source
  cat docs/dev/MEMORY_INDEX.md
fi

1.1 No search query provided → Show recent entries

Display the most recent 20 entries from MEMORY_INDEX.md:

• Show all entries from [Active Work] section
• Show recent entries from the latest completed phase section
• Format: time-descending order

📋 Memory Index (Recent 20 entries)

## [Active Work]
- 15:30 | Completed mem-save skill implementation
- 14:00 | Started dev-phase-manager v1.1.0

## Phase 4 - MCP Tool Server [COMPLETED 2026-02-22]
- 18:30 | Phase 4 complete: MCP server 30 tests all pass
- 16:45 | Architecture decision: spawn_blocking wraps Wasm sync execution
...

1.2 Search query provided → Filter matching entries

Search for entries containing the query keyword in MEMORY_INDEX.md:

• Case-insensitive matching
• Search across all sections (active + archived phases)
• Highlight matching keywords in output

🔍 Search results for "architecture" in MEMORY_INDEX.md:

- [Phase 4] 16:45 | Architecture decision: spawn_blocking wraps Wasm sync execution
- [Phase 0] 10:00 | Architecture design completed: bilingual Python + TypeScript

2. Search claude-mem

Use mcp__plugin_claude-mem_mcp-search__search to search MCP memory:

# If query provided
mcp__plugin_claude-mem_mcp-search__search \
  --query "$ARGUMENTS" \
  --limit 10

# If no query, search recent project memories
mcp__plugin_claude-mem_mcp-search__search \
  --query "[ProjectName]" \
  --limit 10

3. Search Memory Knowledge Graph

Use mcp__memory__search_nodes to search the knowledge graph:

# If query provided
mcp__memory__search_nodes \
  --query "$ARGUMENTS"

# If no query, search project entity
mcp__memory__search_nodes \
  --query "[ProjectName]"

4. Merge and Display Results

Present results in a unified view with local index first:

📋 Memory Search Results

━━━ Local Index (MEMORY_INDEX.md) ━━━
[Results from step 1]

━━━ claude-mem ━━━
📝 [2026-02-22 18:30] [memory] Phase 4 completed - MCP server all tests pass
📝 [2026-02-22 15:00] [memory] Phase 4 architecture decision - spawn_blocking

━━━ Knowledge Graph ━━━
🔗 Ouroboros
   - Adopted bilingual architecture (Python + TypeScript)
   - Using FastMCP as MCP server framework
🔗 MCP Server
   - Using FastMCP framework
   - Expose vault skills as MCP tools

💡 Tips:
- Use /mem-search <keyword> to search specific topics
- Use /mem-save to save current progress
- For full details on a memory, ask me to fetch it

5. Offer Detail Retrieval

If user wants more details on a specific memory:

# Fetch full observation details
mcp__plugin_claude-mem_mcp-search__get_observations \
  --ids [id1, id2]

Use Cases

Scenario 1: Browse all recent memories

/mem-search
# → Shows MEMORY_INDEX.md recent 20 entries
# → Shows claude-mem recent memories
# → Shows knowledge graph entities

Scenario 2: Search specific topic

/mem-search architecture
# → Filters MEMORY_INDEX.md for "architecture" entries
# → Searches claude-mem for "architecture"
# → Searches knowledge graph for "architecture"

Scenario 3: New session context recovery

# New session starts
/mem-search
# → Quickly see what happened in recent work
# → Understand current state without reading all files

Smart Features

1. Local Index Priority

MEMORY_INDEX.md is the primary source because:

• It's structured by phase with timestamps
• It's browsable without semantic search
• It provides a chronological overview
• It works offline without MCP servers

2. Deduplication

When merging results from multiple sources:

• Local index entries are always shown first
• claude-mem results that match local entries are marked as (indexed)
• Knowledge graph results supplement with entity relationships

3. Graceful Degradation

If MEMORY_INDEX.md doesn't exist:

• Skip local index display
• Fall back to claude-mem and knowledge graph
• Suggest running /mem-save to create the index

Notes

1. Read-only: This skill only reads data, never modifies files
2. Performance: Local index is fast; MCP searches may take longer
3. Fallback: If MCP servers unavailable, still shows local index
4. Project detection: Auto-detect project name from git remote or directory