large-language-models
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides implementation examples for Retrieval-Augmented Generation (RAG) systems that process external data sources.
- Ingestion Point: The
Chroma.from_documents(documents, embeddings)function call inSKILL.mddemonstrates how external documents are loaded into the vector database. - Boundary Markers: The provided code snippets do not include instructions for using prompt delimiters or boundary markers to isolate retrieved text from the primary instructions.
- Capability Inventory: The demonstrated code focuses on model fine-tuning and retrieval; it does not contain dangerous capabilities such as arbitrary shell execution, system modification, or file writing.
- Sanitization: The skill lacks examples of input validation, filtering, or sanitization logic for the data ingested into the RAG pipeline.
Audit Metadata