zvec
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill serves as a reference for the Zvec vector database, providing code examples for data operations, indexing configurations, and schema management.
- [EXTERNAL_DOWNLOADS]: The documentation indicates that built-in embedding and reranking functions (such as
DefaultLocalDenseEmbedding) will download pre-trained models (e.g.,all-MiniLM-L6-v2) from external repositories on first use. Additionally, the Flutter package downloads prebuilt native libraries for mobile platforms during the build process. These are standard operations for AI and vector database libraries. - [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external data through collection ingestion (
insert,upsert) and retrieval (query). While this is a common vector database functionality, it remains a potential vector for indirect prompt injection if the retrieved content is subsequently passed to an LLM without proper sanitization. - Ingestion points: Document fields and vectors ingested via
data-operations.md. - Boundary markers: None explicitly documented in the API examples.
- Capability inventory: File system writes (persistence), network access (via API-based embedding providers), and vector similarity search.
- Sanitization: Not addressed within the skill instructions, as it focuses on database operations.
Audit Metadata