local_memory

Manage AI conversation memory locally with automatic extraction, retrieval, and manual commands, ensuring privacy without external APIs or fees.

Security Scan

The skill's code, files, and declared requirements align with its stated purpose (a purely local conversation memory manager), but it will autonomously record and inject conversation content locally — including any sensitive text users send — so treat its automatic behavior as a privacy risk and configure/remove sensitive data accordingly.

Purpose & Capability

Name/description, declared Python dependencies (jieba, scikit-learn, numpy), and implemented behavior (local sqlite DB, TF-IDF embeddings, local jieba tokenizer) are coherent: the code implements local extraction, storage, embedding, retrieval, and manual commands as described.

Instruction Scope

SKILL.md and the code instruct the agent to automatically extract and persist memory from every conversation and to inject retrieved memory into system prompts for subsequent messages. This matches the intended purpose, but the extraction is rule-based and will store any user-provided content (including credentials, IPs, or other secrets) unless the user disables auto_extract or filters content.

Install Mechanism

No install spec is provided (instruction-only install), and no remote downloads or archive extraction are present. The repo includes Python files that depend on common packages; installing those via pip is the expected next step. There are no surprising external URLs or installers.

Credentials

The skill requests no environment variables, no external credentials, and uses only local file paths (db/memory.db and lib/models/). Required permissions are limited to local disk I/O for its own data files.

Persistence & Privilege

The skill is configured to trigger on every message (skill.json pattern ".*" and SKILL.md '自动全局触发'), and the platform default allows autonomous invocation. While not set to always:true, this combination means the skill will run automatically for normal conversations and persist extracted memories locally — increasing privacy exposure. The skill does not modify other skills or system-wide settings.

Assessment

This skill appears to do exactly what it says (local memory extraction and retrieval). Before installing, consider the following: (1) it will automatically save anything users write into db/memory.db — do not send secrets (passwords, API keys, private server credentials) in chats or tests; (2) if you want to avoid automatic storage, disable auto_extract and/or auto_inject in config.json or set expire_days low; (3) secure the database file (permissions, backups, encrypted disk) if it will contain sensitive info; (4) installing requires Python packages (jieba, scikit-learn, numpy) — prefer installing in a virtualenv; (5) review and run test.py only with non-sensitive example data; (6) if you need stronger filtering of sensitive tokens, add explicit secret-detection or opt-out rules before enabling auto-extraction. Overall the skill is internally consistent and local-only, but its automatic capture of all conversation content is the primary privacy risk.

Local Memory 本地记忆技能

功能说明

纯本地运行的AI记忆管理技能，完全免费、数据本地存储，解决AI对话健忘、上下文不足的问题：

• 自动从对话中提取关键事实、用户偏好、项目进展、待办事项等记忆点
• 每轮对话前自动检索相关记忆注入上下文，让AI永远记得你的信息
• 支持手动管理记忆的命令
• 自动处理记忆更新、冲突和过期，不需要人工维护
• 纯本地运行，不需要任何外部API或付费服务，隐私安全

触发规则

自动全局触发，所有对话自动启用记忆功能，无需手动调用

使用命令

命令	说明	示例
/remember <内容>	手动保存信息到记忆库	/remember 我喜欢用TypeScript写代码
/recall <关键词>	查询相关记忆	/recall 部署相关的命令
/forget <关键词>	删除相关记忆	/forget 旧的服务器密码
/memory-list	查看所有记忆列表	/memory-list

配置

所有配置在config.json中修改：

{
  "auto_extract": true, // 自动提取记忆
  "auto_inject": true, // 自动注入上下文
  "max_memory_results": 5, // 每次最多注入的记忆数量
  "embedding_model": "bge-small-zh", // 本地embedding模型
  "expire_days": 90 // 记忆默认过期时间（天）
}

存储路径

• 数据库：db/memory.db
• 向量模型：lib/models/