AgentSkillsCN

memory-management

利用 memory_search 和 memory_get 工具，有效管理长期记忆与知识库。

SKILL.md
--- frontmatter
name: memory-management
description: Manage long-term memory and knowledge base using memory_search and memory_get tools.

Memory Management

Use this skill to interact with the long-term memory and knowledge base system.

When To Use

  • •"search my notes about..."
  • •"what do I know about..."
  • •"add this to my knowledge base"
  • •"find related information"
  • •"summarize my documents"

Available Tools

memory_search

Search the knowledge base for relevant information.

typescript
{
  query: "project requirements",
  maxResults: 10,
  minScore: 0.7
}

memory_get

Retrieve specific memory files by path.

typescript
{
  relPath: "projects/client-a/specs.md";
}

Search Strategies

Semantic Search

For conceptual queries, use natural language:

typescript
memory_search({
  query: "authentication implementation approaches",
  maxResults: 5,
});

Keyword Search

For specific terms:

typescript
memory_search({
  query: "API_KEY environment variable",
  maxResults: 10,
});

Filtered Search

Combine with context:

typescript
memory_search({
  query: "deployment process",
  maxResults: 8,
  minScore: 0.8, // Higher relevance threshold
});

Common Workflows

Finding Project Information

typescript
// Search for architecture docs
const results = await memory_search({
  query: "system architecture design",
  maxResults: 5,
});

// Read full document if found
if (results[0]) {
  await memory_get({ relPath: results[0].path });
}

Context Recovery

When user references past discussions:

typescript
// Search for previous context
await memory_search({
  query: "user requirement about notification system",
  maxResults: 10,
});

Knowledge Discovery

Find related information:

typescript
// Find all documents about a topic
await memory_search({
  query: "database migration",
  maxResults: 20,
});

Output Format

Search results include:

  • •path: File location in memory
  • •score: Relevance score (0-1)
  • •snippet: Preview of content
  • •metadata: Timestamps, tags, etc.

Best Practices

  1. •Start broad, then narrow down
  2. •Use specific keywords for technical terms
  3. •Check multiple results for comprehensive view
  4. •Read full files when snippets are promising
  5. •Combine searches for complex queries

Limitations

  • •Search is semantic (meaning-based), not exact match
  • •Results depend on embedding quality
  • •Very recent additions may not be indexed yet
  • •File paths use relative format from memory root

Example: Research Workflow

typescript
// 1. Search for relevant info
const searchResults = await memory_search({
  query: "microservices communication patterns",
  maxResults: 10,
});

// 2. Read most relevant documents
for (const result of searchResults.slice(0, 3)) {
  await memory_get({ relPath: result.path });
}

// 3. Search for specific implementation
await memory_search({
  query: "message queue vs REST API",
  maxResults: 5,
});