Memory Evolver

记忆系统优化器 - 结合三层记忆与知识图谱的持续自我进化系统。自动诊断、优化、记录记忆系统状态，实现记忆的持续进化。

Key points to consider before installing or running: - The package appears to be incomplete/mismatched: SKILL.md mentions diagnose.py, knowledge_graph.py, KG rebuild and Feishu/EvoMap integrations that are not present in the shipped files. Ask the author for the missing modules or an explanation. - The script hard-codes BASE_DIR to C:\Users\Administrator\.openclaw\workspace. On other systems this will fail or may target the wrong user's workspace. Prefer a version that uses a configurable path or the runtime agent workspace. - The skill reads and writes files under your .openclaw workspace (MEMORY.md, PROJECTS.md, memory/*, knowledge_graph/graph.json and writes memory/optimization循环.md). Back up those files before running to avoid accidental overwrites. - There are no network calls or credential accesses in the current code, which limits immediate exfiltration risk. However the SKILL.md's claim of external reporting (Feishu, EvoMap) suggests future or missing code could add network behavior — treat that as a potential future risk. - Recommended actions: (1) request clarification or the missing modules from the author; (2) inspect the workspace path and consider modifying the script to use a safe, explicit path or ask for a parameterized version; (3) run in a sandbox or with backups first; (4) avoid scheduling automatic runs until you confirm the implementation matches the documented integrations and there are no external endpoints.

Type: OpenClaw Skill Name: memory-evolver Version: 1.0.2 The 'memory-evolver' skill is a utility designed to manage and optimize an AI agent's internal memory files (e.g., MEMORY.md, PROJECTS.md). The core logic in 'index.py' performs file-based diagnostics, generates optimization plans, and logs activities within a hardcoded workspace directory (C:\Users\Administrator\.openclaw\workspace). While the script uses an absolute path and some documented features (like knowledge graph reconstruction) are only partially implemented as placeholders, the code lacks any indicators of malicious intent, such as data exfiltration, unauthorized network access, or harmful prompt injection.