Every conversation leaves a trace. Every decision echoes.

智能记忆系统：支持用户自定义记忆类别、AI自动分析触发、记忆冲突解决。 当用户需要持久记忆跨会话信息、设定记忆规则、自动记忆重要内容、或解决记忆冲突时使用此技能。 支持L1/L2/L3三层记忆分层、JSON快速索引搜索、用户自定义记忆类别、自动分析触发、冲突时以用户为准。 详细文档见 README.md

This skill is a local, file-based auto-memory system and the code matches that description — it parses conversations and writes extracted facts to memory/*.md and an index.json. Before installing: 1) Understand it will ask you (via docs) to modify AGENTS.md / HEARTBEAT.md so it runs automatically each session — it will not auto-enable itself; enabling those rules will make it receive and store entire conversation transcripts. 2) Review the scripts and memory files (they are included) and test in an isolated workspace first; the tool stores potentially sensitive personal data in plaintext under the skill directory. 3) If you want manual control, do not add the AGENTS.md/HEARTBEAT.md entries and keep auto_trigger=false in config; run capture/auto_analyze manually. 4) Be cautious about the ambiguous 'MEMORY.md' reference — confirm which file would be read before granting access. 5) If you need help assessing privacy impact (what will be recorded, retention), ask for a walkthrough of where data is written, retention/cleanup behavior, and how to disable automatic capture.

Type: OpenClaw Skill Name: mnemosyne Version: 1.0.0 The 'mnemosyne' (auto-memory) skill is a highly intrusive conversation logging system designed to automatically extract and store personal and professional information (names, companies, roles, decisions, and emotions) without explicit user triggers. It instructs the AI agent via SKILL.md and AUTO_MEMORY_PROTOCOL.md to analyze and record every session's full content into a local database (memory/index.json and Markdown files). While the code (scripts/auto_analyze.py and scripts/memory_cli.sh) lacks evidence of external data exfiltration, its 'stealthy' operation and focus on sensitive data capture pose a significant privacy risk if the user is unaware of the extent of the logging.