vector-database-management
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides implementation patterns for building Retrieval-Augmented Generation (RAG) and semantic search architectures that ingest external documents. * Ingestion points: External text strings enter the agent context through the document collection and querying pipelines highlighted in
README.mdandSKILL.md. * Boundary markers: The reference prompt template provided inREADME.mddirectly interpolates context data into the LLM prompt without strict encapsulation boundaries or instructions to ignore nested directives. * Capability inventory: The ingested content interacts with vector operations and chat completion APIs, with no connections to code execution tools or lower-level file manipulation tools. * Sanitization: No escaping, filtering, or structural validation of the retrieved text data is featured in the educational examples.
Audit Metadata