recall（recall 搜索）

Recall - Search Knowledge Graph

Search past decisions and patterns from the knowledge graph with optional cloud semantic search enhancement.

Graph-First Architecture (v2.1)

The recall skill uses graph memory as PRIMARY search:

1. Knowledge Graph (PRIMARY): Entity and relationship search via mcp__memory__search_nodes - FREE, zero-config, always works
2. Semantic Memory (mem0): Optional cloud search via search-memories.py script - requires MEM0_API_KEY, use with --mem0 flag

Benefits of Graph-First:

• Zero configuration required - works out of the box
• Explicit entity and relationship queries
• Fast local search with no network latency
• No cloud dependency for basic operation
• Optional cloud enhancement with --mem0 flag for semantic similarity search

Overview

• Finding past architectural decisions
• Searching for recorded patterns
• Looking up project context
• Retrieving stored knowledge
• Querying cross-project best practices
• Finding entity relationships

Usage

/recall <search query>
/recall --category <category> <search query>
/recall --limit <number> <search query>

# Cloud-enhanced search (v2.1.0+)
/recall --mem0 <query>                     # Search BOTH graph AND mem0 cloud
/recall --mem0 --limit 20 <query>          # More results from both systems

# Scoped search
/recall --agent <agent-id> <query>          # Filter by agent scope
/recall --global <query>                    # Search cross-project best practices

Advanced Flags

Context-Aware Result Limits (CC 2.1.6)

Result limits automatically adjust based on context_window.used_percentage:

Workflow

1. Parse Input

Check for --category <category> flag
Check for --limit <number> flag
Check for --mem0 flag → search_mem0: true
Check for --agent <agent-id> flag → filter by agent_id
Check for --global flag → search global scope
Extract the search query

2. Search Knowledge Graph (PRIMARY)

Use mcp__memory__search_nodes:

{
  "query": "user's search query"
}

Knowledge Graph Search:

• Searches entity names, types, and observations
• Returns entities with their relationships
• Finds patterns like "X uses Y", "X recommends Y"

Entity Types to Look For:

• Technology: Tools, frameworks, databases (pgvector, PostgreSQL, React)
• Agent: OrchestKit agents (database-engineer, backend-system-architect)
• Pattern: Named patterns (cursor-pagination, connection-pooling)
• Decision: Architectural decisions
• Project: Project-specific context
• AntiPattern: Failed patterns

3. Search mem0 (OPTIONAL - only if --mem0 flag)

Skip if --mem0 flag NOT set or MEM0_API_KEY not configured.

Execute the script IN PARALLEL with step 2:

!bash skills/mem0-memory/scripts/crud/search-memories.py \
  --query "user's search query" \
  --user-id "orchestkit-{project-name}-decisions" \
  --limit 10 \
  --enable-graph

User ID Selection:

• Default: orchestkit-{project-name}-decisions
• With --global: orchestkit-global-best-practices

Filter Construction:

• Always include user_id filter
• With --category: Add { "metadata.category": "{category}" } to AND array
• With --agent: Add { "agent_id": "ork:{agent-id}" } to AND array

4. Merge and Deduplicate Results (if --mem0)

Only when both systems return results:

1. Collect results from both systems
2. For each mem0 memory, check if its text matches a graph entity observation
3. If matched, mark as [CROSS-REF] and merge metadata
4. Remove pure duplicates (same content from both systems)
5. Sort: graph results first, then mem0 results, cross-refs highlighted

5. Format Results

Graph-Only Results (default):

🔍 Found {count} results matching "{query}":

[GRAPH] {entity_name} ({entity_type})
   → {relation1} → {target1}
   → {relation2} → {target2}
   Observations: {observation1}, {observation2}

[GRAPH] {entity_name2} ({entity_type2})
   Observations: {observation}

With --mem0 (combined results):

🔍 Found {count} results matching "{query}":

[GRAPH] {entity_name} ({entity_type})
   → {relation} → {target}
   Observations: {observation}

[GRAPH] {entity_name2} ({entity_type2})
   Observations: {observation}

[MEM0] [{time ago}] ({category}) {memory text}

[MEM0] [{time ago}] ({category}) {memory text}

[CROSS-REF] {memory text} (linked to {N} graph entities)
   📊 Linked entities: {entity1}, {entity2}

With --mem0 when MEM0_API_KEY not configured:

🔍 Found {count} results matching "{query}":

[GRAPH] {entity_name} ({entity_type})
   → {relation} → {target}
   Observations: {observation}

⚠️ mem0 search requested but MEM0_API_KEY not configured (graph-only results)

High Context Pressure (>85%):

🔍 Found 12 matches (showing 3 due to context pressure at 87%)

[GRAPH] pgvector (Technology)
   → USED_FOR → RAG
[GRAPH] cursor-pagination (Pattern)
[GRAPH] database-engineer (Agent)
   → RECOMMENDS → pgvector

More results available. Use /recall --limit 10 to override.

6. Handle No Results

🔍 No results found matching "{query}"

Searched:
• Knowledge graph: 0 entities

Try:
• Broader search terms
• /remember to store new decisions
• --global flag to search cross-project best practices
• --mem0 flag to include cloud semantic search

Time Formatting

Examples

Basic Graph Search

Input: /recall database

Output:

🔍 Found 3 results matching "database":

[GRAPH] PostgreSQL (Technology)
   → CHOSEN_FOR → ACID-requirements
   → USED_WITH → pgvector
   Observations: Chosen for ACID requirements and team familiarity

[GRAPH] database-engineer (Agent)
   → RECOMMENDS → pgvector
   → RECOMMENDS → cursor-pagination
   Observations: Uses pgvector for RAG applications

[GRAPH] cursor-pagination (Pattern)
   Observations: Scales well for large datasets

Category Filter

Input: /recall --category architecture API

Output:

🔍 Found 2 results matching "API" (category: architecture):

[GRAPH] api-gateway (Architecture)
   → IMPLEMENTS → rate-limiting
   → USES → JWT-authentication
   Observations: Central entry point for all services

[GRAPH] REST-API (Pattern)
   → FOLLOWS → OpenAPI-spec
   Observations: Standard for external-facing APIs

Cloud-Enhanced Search

Input: /recall --mem0 database

Output:

🔍 Found 5 results matching "database":

[GRAPH] PostgreSQL (Technology)
   → CHOSEN_FOR → ACID-requirements
   Observations: Chosen for ACID requirements

[GRAPH] database-engineer (Agent)
   → RECOMMENDS → pgvector
   Observations: Uses pgvector for RAG

[MEM0] [2 days ago] (decision) PostgreSQL chosen for ACID requirements and team familiarity

[MEM0] [1 week ago] (pattern) Database connection pooling with pool_size=10, max_overflow=20

[CROSS-REF] [3 days ago] pgvector for RAG applications (linked to 2 entities)
   📊 Linked: database-engineer, pgvector

Agent-Scoped Search

Input: /recall --agent backend-system-architect "API patterns"

Output:

🔍 Found 2 results from backend-system-architect:

[GRAPH] backend-system-architect (Agent)
   → RECOMMENDS → cursor-pagination
   → RECOMMENDS → repository-pattern
   Observations: Use versioned endpoints: /api/v1/, /api/v2/

[GRAPH] repository-pattern (Pattern)
   Observations: Separate controllers, services, and repositories

Cross-Project Search

Input: /recall --global --category pagination

Output:

🔍 Found 3 GLOBAL best practices (pagination):

[GRAPH] cursor-pagination (Pattern)
   → SCALES_FOR → large-datasets
   → PREFERRED_OVER → offset-pagination
   Observations: From project: ecommerce, analytics, cms

[GRAPH] keyset-pagination (Pattern)
   → USED_FOR → real-time-feeds
   Observations: From project: analytics

[GRAPH] offset-pagination (AntiPattern)
   Observations: Caused timeouts on 1M+ rows

Relationship Query

Input: /recall what does database-engineer recommend

Output:

🔍 Found relationships for database-engineer:

[GRAPH] database-engineer (Agent)
   → RECOMMENDS → pgvector
   → RECOMMENDS → cursor-pagination
   → RECOMMENDS → connection-pooling
   → USES → PostgreSQL
   Observations: Specialist in database architecture

Related Skills

• remember: Store information for later recall

Error Handling

• If knowledge graph unavailable, show configuration instructions
• If --mem0 requested without MEM0_API_KEY, proceed with graph-only and notify user
• If search query empty, show recent entities instead
• If no results, suggest alternatives
• If --agent used without agent-id, show available agents
• If --global returns no results, suggest storing with /remember --global
• If --mem0 returns partial results (mem0 failed), show graph results with degradation notice