Memory Work

An AI-first knowledge management system with evolvable memory - 一套 AI 时代的注意力保护架构

This skill is coherent with its purpose but operates by reading and writing your local knowledge files. Before enabling it: (1) review USER.md, SOUL.md, About Me/ and any files in the repo for sensitive personal data and remove or redaction any you don't want the agent to read; (2) back up your vault/files so you can recover from unwanted edits; (3) confirm how the agent's automatic behaviors work (startup sync, deep retrospective) and whether you can disable or require explicit approval for writes; (4) because the source is 'unknown' in the registry metadata, review the upstream GitHub repo link and change history to ensure it matches expectations; (5) consider running the skill with read-only permissions or in a sandbox until you're comfortable with its edits; and (6) watch for integrations (Obsidian, calendar import) that imply copying files out or generating external artifacts. If you want stricter control, require the agent to ask before any write to HIGH-sensitivity folders.

Type: OpenClaw Skill Name: memory-work Version: 1.0.1 The bundle is a comprehensive framework for a personal AI agent system ('Memory Work v2') designed for knowledge management and long-term memory. It implements a sophisticated multi-layered memory architecture (Persistent, Working, Dynamic, and Procedural) and uses 'Zone Agents' to manage data sensitivity and isolation. While the system is explicitly designed to handle highly sensitive personal information (as seen in USER.md and the example '家庭信息.md'), the instructions in CLAUDE.md, SOUL.md, and the various zone agents prioritize user consent, transparency, and explicit confirmation for any modifications to sensitive data. There is no evidence of malicious intent, data exfiltration, or unauthorized execution logic; the system appears to be a legitimate productivity tool for AI-assisted workflows.