Auto Memory Manager

Automatic memory management based on built-in memory-lancedb-local-storage. Auto-extract key points, compress conversation, write to MEMORY.md and update vec...

Security Scan

The skill's code and instructions match its stated purpose (automatically summarizing sessions, appending to MEMORY.md, and updating the local vector index); nothing requested or installed is disproportionate or unrelated.

Purpose & Capability

Name/description promise automatic memory extraction, compression, write to MEMORY.md and vector index update. The code registers session:end and command:new hooks, invokes the LLM to extract key points, appends results to MEMORY.md in ctx.workspace, and runs 'openclaw memory index --force' — these actions directly implement the declared functionality.

Instruction Scope

Instructions and code automatically trigger at session:end (and on /new), gather the full session messages, call the default LLM to extract JSON-formatted key points, and append them to MEMORY.md. This is consistent with the stated scope, but it does mean full conversation content is sent to the configured LLM provider and is written to disk without additional confirmation. There are minimal heuristics (message count >=2 and keyPoints length >=20) to avoid trivial writes.

Install Mechanism

No install spec; this is an instruction+code skill that relies on runtime hooks and existing platform APIs. Nothing is downloaded or written to disk by an installer. Code included is small and readable; no external URL downloads or packaged installers are invoked.

Credentials

The skill requests no environment variables, no external credentials, and no config paths. Its operations use platform-provided ctx APIs (invokeLLM, exec, listSessions, workspace). The lack of extra credentials is proportional to its stated role.

Persistence & Privilege

always:false (not force-installed). The skill can run autonomously when session:end fires (normal for skills). It also registers an optional daily cron (if ctx.cron exists) and a CLI command for manual runs. These behaviors increase automatic activity but are coherent with a memory management skill; users should be aware it will write to MEMORY.md and invoke indexing automatically.

Assessment

This skill will automatically summarize finished sessions by sending the session text to the platform LLM, append the extracted summary to MEMORY.md in the agent workspace, and run 'openclaw memory index --force' to update local vector indexes. It requests no credentials and installs nothing. Before installing, consider: (1) Do you want automatic writes to MEMORY.md (back it up if needed)? (2) Are you comfortable the full conversation will be sent to your configured LLM provider? (3) If you prefer manual control, disable the skill or remove/modify its session:end and cron hooks. If you need more assurance, review or test the skill in a safe environment to confirm behavior and output formatting.

memory-auto-manager

基于内置 memory-lancedb-local-storage 的自动记忆管理 Skill。

功能

• ✅ 会话结束自动触发 → 无需手动输入指令，会话结束自动运行
• ✅ 自动压缩提炼 → 调用 LLM 去掉闲聊、重复内容，保留核心信息
• ✅ 自动分类 → 将内容分为 decision/fact/preference/entity 四类
• ✅ 自动写入 → 提炼后的核心内容自动追加到 MEMORY.md 永久记忆
• ✅ 自动更新索引 → 写入完成后自动执行 openclaw memory index --force 更新本地向量语义索引
• ✅ 完全兼容 → 不修改原有 memory-lancedb-local-storage 配置和数据，只是在上层增加自动管理功能

配置

无需额外配置，使用当前 OpenClaw 默认 LLM 进行提炼。

工作流程

1. 会话结束 → 触发 session:end hook
2. 获取完整会话对话
3. 调用 LLM 提炼核心要点，分类
4. 判断是否值得长期保存（内容太短/无价值跳过）
5. 值得保存 → 追加写入 MEMORY.md
6. 执行 openclaw memory index --force 更新向量索引
7. 完成，日志输出结果

作者

OpenClaw auto generated