OpenViking Light

轻量级 RAG 知识库 — 基于 BM25 全文检索 + MiniMax LLM 生成回答。 纯本地 Python 实现，无需 embedding API，不需要下载模型。 触发：用户问"你记得之前..."、"查一下之前..."、"关于...的知识"等需要检索记忆的问题。

This skill implements local BM25 retrieval but uses a remote MiniMax chat API for generation. Before installing: (1) expect to provide MINIMAX_API_KEY (not declared in metadata) — queries and retrieved context will be sent to the configured MINIMAX_API_HOST; do not store sensitive secrets or private data in the skill's JSON store unless you trust that endpoint. (2) The scripts auto-install 'jieba' via pip at runtime — consider pre-installing dependencies in a controlled environment. (3) If you want a truly offline solution, remove or modify the chat() calls to use a local model instead. (4) Ask the publisher to update the skill metadata to declare required environment variables and to document the network call to the MiniMax endpoint. If you don't trust the MiniMax provider or cannot provide a dedicated API key with limited scope, treat the generation feature as risky and use search-only commands instead.

Type: OpenClaw Skill Name: openviking-light Version: 1.0.0 The skill implements a local RAG (Retrieval-Augmented Generation) system using the BM25 algorithm and the MiniMax LLM. It is classified as suspicious because multiple files (ask.py, bm25.py, and search.py) use os.system to automatically install the 'jieba' library if it is missing. While this is likely intended for user convenience, programmatically executing shell commands to modify the environment is a high-risk behavior that could be exploited. The code also makes external network requests to the MiniMax API (api.minimaxi.com), which is consistent with its stated purpose, and no evidence of data exfiltration or intentional malice was found.