Memory V2

基于中文向量嵌入和知识图谱，实现语义搜索、实体提取及自动管理的高效记忆系统。

This skill appears coherent and implements what it claims. Key things to consider before installing: 1) It will run npm install (pulling packages from npm) and on first run will download large ML model files (~500MB) from external model hosts — expect network use and disk space consumption. 2) Data (vector DB and graph) is stored on disk by default (./vector-db and ./memory/ontology); point those paths to a suitable directory and ensure permissions are acceptable. 3) The skill requests no credentials and has no hidden network endpoints in the code, but the embedding/NER model selection is configurable — avoid pointing to untrusted models since a malicious model could change runtime behavior. 4) Review and run the included tests in a sandboxed environment if you want to validate behavior before integrating into production. If you need higher assurance, verify the exact model download hosts used at runtime and consider pinning model sources/versions.

Type: OpenClaw Skill Name: yuyonghao-memory-v2 Version: 0.1.0 The skill bundle implements a sophisticated memory system for OpenClaw agents, integrating vector storage (LanceDB), a knowledge graph (JSONL-based), and Named Entity Recognition (NER) via Transformers.js. The code in src/memory-system.js and src/vector-store.js follows standard patterns for local AI applications, including model downloading from Hugging Face and local file-based persistence. No evidence of data exfiltration, unauthorized network activity, or malicious prompt injection was found in the code or documentation (SKILL.md).