ByteHouse 多模态检索 Skill，支持文本、图片、视频的向量化存储和混合检索。当用户需要在ByteHouse数据库中进行多模态向量化存储和混合检索时，使用此Skill。

This skill appears to implement the ByteHouse + Volcano Engine Ark multimodal search it claims, but there are important caveats to review before installing or running it: - Registry metadata omission: The registry declares no required env vars or primary credential, but the code needs ARK_API_KEY (for embeddings) and BYTEHOUSE_* credentials (host, user, password, database). Treat this omission as a transparency issue — do not provide high-privilege credentials without verifying the source. - Limit credentials and scope: Create and supply least-privilege credentials (a ByteHouse user limited to the specific database/tables and an API key with only embedding permission) rather than root/admin DB credentials. - Review and test code locally: The package includes Python scripts. Inspect them (you already have them) and run in a sandboxed environment before giving it network access or real credentials. - Beware SQL construction: insert_document and other methods build SQL using f-strings and interpolate table_name and content; avoid passing untrusted values or sanitize inputs first to prevent injection or malformed SQL. - Verify third-party packages: SKILL.md instructs pip installing clickhouse-connect and volcengine-sdk. Confirm these are the official packages, and consider installing into an isolated virtual environment. - MCP client behavior: The code tries to use an mcp_client if present (it may call other tools or skills). If you rely on MCP, understand what that client does; if unsure, initialize the search client with prefer_mcp=False. - Network/data exfiltration: The skill will send content (text/image/video URLs) and embeddings to external Ark endpoints and connect to your ByteHouse instance. Do not pass sensitive data (personal data, secrets) unless you accept that it will be transmitted to those endpoints. If you need higher assurance: ask the publisher for an explicit list of required env vars in the registry and for a signed package or known homepage. If you cannot verify the source, run in a restricted environment or decline installation.

Type: OpenClaw Skill Name: byted-bytehouse-multimodal-search Version: 1.0.0 The skill bundle provides multimodal search capabilities using ByteHouse and Volcengine Ark API, but it contains a significant SQL injection vulnerability. In `scripts/search_client.py`, multiple methods (such as `vector_search`, `hybrid_search`, and `_execute_sql`) construct SQL queries using f-strings with inadequate sanitization. Specifically, the `filter_condition` and `query_text` parameters are directly interpolated into SQL strings, which could allow an attacker or a malicious prompt to execute unauthorized database commands. While the code follows its stated purpose and lacks evidence of intentional malice or data exfiltration, the high-risk nature of the SQL injection flaw warrants a suspicious classification.