AgentSkillsCN

memory

海马体接口——面向大语言模型的向量记忆（LanceDB + FastEmbed）

SKILL.md
--- frontmatter
name: "memory"
version: "1.0.0"
description: "The Hippocampus Interface - Vector-based Memory for LLM (LanceDB + FastEmbed)"
routing_keywords:
  [
    "memory",
    "remember",
    "store",
    "save",
    "learn",
    "forget",
    "context",
    "persistent",
    "long-term",
    "recall",
    "embeddings",
    "vector",
    "note",
    "wisdom",
  ]
authors: ["omni-dev-fusion"]
intents:
  - "Remember or store information"
  - "Recall past experiences"
  - "Save learning to memory"
  - "Retrieve context from memory"

Memory Skill Policy

Router Logic

Scenario 1: User wants to store something

  1. •Analyze: Determine the type of memory (insight, rule, decision)
  2. •Store: Call save_memory(content, metadata)
  3. •Confirm: Show the saved memory ID

Scenario 2: User wants to remember/search

  1. •Search: Call search_memory(query, limit)
  2. •Format: Present results with relevance scores
  3. •Respond: "I found X memories about that..."

Scenario 3: User asks "What have you learned?", "Show memories"

  1. •List: Call get_memory_stats()
  2. •Recall: Call search_memory() with relevant keywords
  3. •Present: Show structured summary

Commands Reference

CommandDescriptionExample
save_memoryStore insight/recipe into vector memorysave_memory("Use semantic versioning", {"tag": "git"})
search_memorySemantic search in memorysearch_memory("git commit format", limit=5)
index_memoryOptimize vector index (IVF-FLAT)index_memory()
get_memory_statsGet memory countget_memory_stats()
load_skillLoad skill manifest into memoryload_skill("git")

Workflow: Store an Insight

code
User: Remember that for this project, all commit messages must be in English.

Claude:
  1. save_memory(
       content="All commit messages must be in English only",
       metadata={"domain": "git", "source": "user"}
     )
  2. → Saved memory [a1b2c3d4]: All commit messages must be in English only
  3. → "Got it! I'll remember that commit messages must be in English."

Workflow: Recall Past Learning

code
User: What do we use for git tags?

Claude:
  1. search_memory("git tags semantic versioning")
  2. → Found 2 matches:
     - [Score: 0.8921] Always use semantic versioning for git tags...
     - [Score: 0.7234] v1.2.3 format for releases
  3. → "I found memories about git tags:
       - Always use semantic versioning for git tags..."

Memory vs Knowledge Skill

AspectMemoryKnowledge
SourceLLM's own learningsProject documentation
StorageLanceDB (vector)File system (markdown)
QuerySemantic searchKeyword/pattern match
Purpose"What did I learn?""What are the rules?"
UpdateRuntime accumulationPre-indexed docs

Best Practices

  1. •Store actionable insights, not obvious facts
  2. •Include domain in metadata for filtering
  3. •Use clear, searchable phrasing in content
  4. •Recall before acting on project-specific patterns