llama-index
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents methods for ingesting data from external sources (local directories, APIs, databases) and processing it through LLM query engines and agents.
- Ingestion points: Uses
SimpleDirectoryReaderto load files from./dataand provides templates forCustomDataReaderto fetch data from remote API endpoints. - Boundary markers: Code examples do not demonstrate the use of delimiters or specific instructions to the model to ignore embedded commands in the retrieved context.
- Capability inventory: The documented agents (
ReActAgent,FunctionCallingAgent) have the capability to execute tools and query knowledge bases, which could be manipulated by instructions hidden within the ingested data. - Sanitization: The provided patterns lack input validation or sanitization for the content being indexed or queried, making the resulting application vulnerable to data-driven instruction overrides.
Audit Metadata