Semantic Memory

OpenClaw Agent 中文长期记忆系统。jieba TF-IDF + 向量检索三轨混合，中文语义优先，支持多Agent记忆协同。触发词：向量数据库、记忆检索、长期记忆、语义搜索、vector search、memory retrieval

This skill implements the described memory/search functionality, but take these precautions before installing/running: 1) Do not run the ChromaDB server bound to 0.0.0.0 on a machine you don't trust or that is network-accessible; prefer localhost (127.0.0.1) or enable authentication/proxying. 2) Be careful with import_memory.py: if you pass a directory path, it will recursively read and upload .md files — avoid pointing it at system, home, or other sensitive directories. 3) The TF-IDF cache uses Python pickle; treat the cache directory as sensitive and ensure it's not writable by untrusted users/processes (an attacker-modified pickle could lead to code execution on load). 4) Review and, if needed, harden defaults in start_chroma.sh and README (host, port, API auth) before use. 5) Only run this skill and its scripts if you trust the source; if unsure, request the author to change defaults to bind to localhost and to use a safer cache format (e.g., JSON) or validate pickle integrity.

Type: OpenClaw Skill Name: semantic-memory Version: 1.0.0 The skill provides a legitimate Chinese-optimized memory system, but it contains significant security vulnerabilities. Specifically, `scripts/vector_search.py` uses the `pickle` module to load and save TF-IDF indices, which is a well-known vector for Remote Code Execution (RCE) if the cache directory is compromised. Additionally, `scripts/start_chroma.sh` configures the database to listen on all network interfaces (`0.0.0.0`) by default, increasing the network attack surface. While these appear to be unintentional design flaws rather than intentional malware, they represent high-risk behaviors.