rag-knowledge-assistant

基于向量数据库的 RAG(检索增强生成) 知识库助手。支持语义检索、多格式文档 (PDF/Word/Excel/Markdown) 处理、智能问答。使用 Chroma 向量库 + BGE-M3 Embedding 模型。适用于从 knowledge 目录快速检索信息、回答基于文档的问题。触发词："从知识库查"、"...

This skill is mostly coherent for building a local RAG index and searching documents, but take these precautions before installing or running it: - Do NOT follow the PUSH_GUIDE recommendation to embed your GitHub Personal Access Token in the git push URL; that exposes the token to shell history/logs. Use a credential manager, SSH keys, or temporary tokens instead. - Inspect scripts/requirements.txt before pip installing. Consider running pip install inside a virtualenv or container to avoid affecting your system Python. - The code may download large HuggingFace models (BAAI/bge-m3) or expect a local Ollama service — ensure you have disk, bandwidth, and understand where models are fetched from. If you cannot or do not want remote downloads, use the Ollama/local models option but only if Ollama runs locally. - Only put documents you trust into the ./knowledge directory; the skill will read and index everything under that path (don’t index sensitive secrets, private keys, or system files). - Run the tooling in an isolated environment (VM/container) if you are unsure, and verify network activity if you need strong assurance that data is not being transmitted off-host. If you want a cleaner manifest: ask the author to list operational requirements (local Ollama service, approximate model download size, any optional env vars) and to remove or fix the insecure push instructions.

Type: OpenClaw Skill Name: rag-knowledge-assistant Version: 1.0.0 The bundle provides a functional RAG (Retrieval-Augmented Generation) system using LangChain, ChromaDB, and Ollama or HuggingFace embeddings. The scripts (index_knowledge.py, rag_query.py, etc.) implement standard document loading, text splitting, and vector search logic aligned with the stated purpose of a knowledge assistant. No evidence of data exfiltration, unauthorized network calls, or malicious prompt injection was found. Documentation files like PUSH_GUIDE.md and README.md provide standard setup instructions, although PUSH_GUIDE.md references a specific GitHub repository (AIxbinge/rag-skill.git) which appears to be for project-specific contribution rather than a malicious redirect.