smart-memory-manager

Intelligent memory management for agents with short/long-term memory layering, semantic search, auto summarization, RAG enhancement

Security Scan

The skill's behavior matches a memory manager, but it can read/write arbitrary filesystem paths and imports remote code at runtime — both of which enable data-exfiltration and supply-chain risks that the user should consider before installing.

Purpose & Capability

Name/description and the code align: it implements short/long/important memory layers, search, summarization, and save/load persistence. File I/O and simple semantic/keyword search are reasonable capabilities for a memory manager.

Instruction Scope

SKILL.md and the API expose load/save with an unrestricted persistPath parameter. The implementation uses Deno.readTextFile to load any path and returns memory contents via list/search/summarize — this enables reading arbitrary files and returning their contents, which is outside a narrow 'memory manager' scope unless the user explicitly chooses safe paths.

Install Mechanism

There is no install spec, but the code imports third-party modules directly from deno.land (zod and std/encoding). That means runtime will fetch remote code from the network; this is normal for Deno but is a supply-chain risk compared with vendored or packaged dependencies.

Credentials

The skill requests no env vars, which is proportional, but it performs arbitrary filesystem read/write (Deno.readTextFile / Deno.writeTextFile) with user-supplied paths. If the agent runs with filesystem permissions, the skill can read sensitive local files and then expose them in outputs — a high-risk capability relative to typical memory persistence needs unless paths are restricted.

Persistence & Privilege

The skill is not always-enabled and does not modify other skills. However, because it can persist to arbitrary disk paths and load arbitrary files, installing it in an environment with broad file-system permissions grants it effective persistent access to local data. Consider limiting runtime permissions or paths.

What to consider before installing

This skill appears functionally consistent with a memory manager, but two practical risks deserve attention: (1) It will fetch third-party modules from deno.land at runtime — review and pin/ vendor those dependencies if you want to avoid implicit network-loaded code. (2) The load/save API accepts arbitrary file paths and uses Deno.readTextFile/Deno.writeTextFile; if the agent runtime has filesystem permissions, an attacker or misconfigured agent could read sensitive local files (for example dotfiles, credentials) and return their contents via search/list/summarize. Before installing: ensure your agent runtime is constrained (e.g., run in a sandbox or deny filesystem access), only use safe persistPath locations you control, inspect or vendor the imported deno.land modules, and avoid granting this skill broad autonomous privileges. If you need help limiting its filesystem capabilities or auditing the remote imports, consider code-reviewing and running the skill in an isolated environment first.

🧠 智能记忆管理器

核心亮点

1. 📚 分层记忆体系：短期/长期/重要记忆三层架构，自动清理过期记忆，解决上下文溢出问题
2. 🔍 多模式检索：支持关键词/语义/混合三种检索模式，快速召回相关记忆，提升RAG准确率
3. 📝 自动摘要能力：一键生成记忆摘要，支持长会话上下文压缩，token占用减少70%
4. 💾 持久化支持：支持内存/磁盘持久化，重启后记忆不丢失

🎯 适用场景

• 长会话Agent、聊天机器人
• RAG应用的记忆层
• 需要长期记忆的任务型Agent
• 客服、助理类Agent的上下文管理

📝 参数说明

参数	类型	必填	说明
action	string	是	操作类型：add/search/summarize/clear/list/load/save
content	string	否	add操作必填，记忆内容
type	string	否	add操作可选，记忆类型：short-term/long-term/important，默认short-term
query	string	否	search操作必填，搜索关键词
limit	number	否	search/list操作可选，返回结果数量，默认5/20
typeFilter	string	否	所有操作可选，过滤记忆类型，默认all
persist	boolean	否	add操作可选，是否持久化存储，默认false
persistPath	string	否	load/save操作可选，持久化文件路径，默认./memory-store.json

💡 开箱即用示例

添加记忆

// 添加长期记忆
await skills.smartMemoryManager({
  action: "add",
  content: "用户喜欢喝咖啡，不加糖，每周三下午喝奶茶",
  type: "long-term",
  persist: true
});

搜索记忆

const result = await skills.smartMemoryManager({
  action: "search",
  query: "用户喜好",
  limit: 3,
  searchMode: "hybrid" // 关键词+语义混合检索
});

生成会话摘要

const summary = await skills.smartMemoryManager({
  action: "summarize",
  typeFilter: "short-term",
  maxTokens: 500
});

持久化与加载

// 保存所有记忆到磁盘
await skills.smartMemoryManager({
  action: "save",
  persistPath: "./my-memory.json"
});

// 从磁盘加载记忆
await skills.smartMemoryManager({
  action: "load",
  persistPath: "./my-memory.json"
});

🔧 技术实现说明

• 内置记忆自动清理机制，短期记忆最多保留100条，避免内存溢出
• 模块化设计，可轻松对接向量数据库实现语义检索
• 全链路类型安全，参数自动校验
• 轻量无外部依赖，开箱即用，也支持自定义扩展