search-memory

X-McKay 的头像作者 X-McKay

Search the hierarchical memory system for relevant memories. Searches across all tiers (working, episodic, semantic) and returns ranked results. Use for context retrieval and pattern matching. Keywords: memory, search, retrieve, recall, context, similarity.

📖 中文导读:Search the hierarchical memory system for relevant memories. Searches across all tiers (working, episodic, semantic) and returns ranked results. Use for context retrieval and pattern matching. Keywords: memory, search, retrieve, recall, context, similarity. 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: search-memory description: > Search the hierarchical memory system for relevant memories. Searches across all tiers (working, episodic, semantic) and returns ranked results. Use for context retrieval and pattern matching. Keywords: memory, search, retrieve, recall, context, similarity. metadata: domain: general category: memory requires-approval: false confidence: 0.9 mcp-servers: []

Search Memory

Preconditions

Before applying this skill, verify:

  • Memory system is initialized
  • Search query is meaningful
  • Result limit is specified

Actions

1. Build Search Query

Prepare the semantic search query:

query: $search_text
user_id: $user_id
limit: $result_limit (default: 10)

2. Search Each Tier

Working Memory (exact + recency):

working_results = memory.search_working(query, limit=limit//3)

Episodic Memory (semantic + time decay):

episodic_results = memory.search_episodic(
    query=query,
    user_id=user_id,
    limit=limit//3
)

Semantic Memory (semantic + permanence):

semantic_results = memory.search_semantic(
    query=query,
    user_id=user_id,
    limit=limit//3
)

3. Merge and Rank Results

Combine results from all tiers:

  • Apply tier-specific weights
  • Remove duplicates
  • Sort by relevance score
  • Return top N results

Success Criteria

The skill succeeds when:

  • All accessible tiers searched
  • Results ranked by relevance
  • Metadata includes source tier

Failure Handling

If search fails:

  1. Return partial results from available tiers
  2. Log which tiers were unreachable
  3. Include error context in response

Examples

Input Context:

{
  "query": "nginx pod memory issues",
  "user_id": "k8s-monitor",
  "limit": 10
}

Expected Output:

{
  "results": [
    {
      "content": "Pod nginx-abc crashed due to OOM. Fixed by increasing limits.",
      "tier": "episodic",
      "score": 0.89,
      "timestamp": "2024-01-15T10:30:00Z"
    },
    {
      "content": "Nginx pods require 1Gi memory minimum for production workloads",
      "tier": "semantic",
      "score": 0.85,
      "pattern_type": "resource_requirement"
    }
  ],
  "total_searched": 3,
  "tiers_searched": ["working", "episodic", "semantic"]
}