ai-orchestration-llamaindex
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [SAFE]: The skill instructions proactively encourage security best practices, including the use of
dotenvfor secret management and a specific warning against usingeval()in tool implementations.\n- [INDIRECT_PROMPT_INJECTION]: The skill describes building RAG pipelines that ingest external documents, which is a potential surface for indirect prompt injection.\n - Ingestion points: Documents are loaded via
SimpleDirectoryReader(SKILL.md) and customApiDocumentReader(examples/ingestion.md).\n - Boundary markers: The examples include system prompts to constrain behavior but do not explicitly demonstrate the use of delimiters for retrieved chunks.\n
- Capability inventory: The patterns demonstrate network access (fetch), file-system writes (persistence), and tool execution.\n
- Sanitization: No explicit content sanitization is described, relying on framework defaults and LLM guardrails.\n- [PERSISTENCE]: The skill explains how to use local storage to persist vector indexes via
storageContextFromDefaults. This is a standard architectural pattern for RAG applications to minimize re-indexing costs and does not involve malicious persistence mechanisms like modifying shell profiles or system services.
Audit Metadata