vector-db-search
Pass
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Python to execute local helper scripts and utilizes curl for a local health check on 127.0.0.1:8110. These operations are well-defined and serve the primary purpose of managing and querying the vector database.
- [EXTERNAL_DOWNLOADS]: The skill declares dependencies on several standard, well-known libraries such as chromadb, langchain, and huggingface. These are common components in AI agent development and do not pose a security risk in this context.
- [PROMPT_INJECTION]: As a retrieval-based tool, the skill presents an indirect prompt injection surface by processing natural language queries and repository content.
- Ingestion points: User natural language questions and retrieved documentation or code snippets from the vector store.
- Boundary markers: The skill implements a Source Transparency Declaration to manage retrieval failures, though it does not explicitly define delimiters for untrusted data.
- Capability inventory: Access to local files, execution of Python scripts, and local network operations.
- Sanitization: No explicit sanitization of retrieved repository content is documented in the provided logic files.
Audit Metadata