llama-index

Pass

Audited by Gen Agent Trust Hub on May 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill defines architectural patterns for building RAG (Retrieval-Augmented Generation) pipelines and autonomous agents that process external, untrusted data. This configuration inherently introduces a risk of indirect prompt injection.
  • Ingestion points: Document loaders like SimpleDirectoryReader (loading from ./data) and CustomDataReader (fetching from api://endpoint) in SKILL.md represent entry points for untrusted data.
  • Boundary markers: The provided patterns do not implement boundary delimiters or 'ignore' instructions to prevent the model from obeying instructions embedded in the retrieved data.
  • Capability inventory: The skill demonstrates using ReActAgent and FunctionCallingAgent in SKILL.md with tools for calculations and searching, which could be abused if an indirect injection successfully overrides the agent's system prompt.
  • Sanitization: No examples or requirements for sanitizing or validating ingested data are included in the guide.
  • [SAFE]: All external downloads and references target well-known and trusted official repositories for the LlamaIndex framework and established vector database providers.
Audit Metadata
Risk Level
SAFE
Analyzed
May 6, 2026, 08:54 AM
Security Audit — agent-trust-hub — llama-index