eks-genai
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a collection of documentation and architectural guidance. It refers exclusively to official AWS documentation, validated workshops, and well-known open-source GenAI frameworks (such as vLLM, Ray, and LiteLLM).
- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for Agentic AI and Retrieval-Augmented Generation (RAG) which process untrusted external data.
- Ingestion points: Data enters the system via user queries through the AI Gateway (LiteLLM) and retrieved context from vector stores like Amazon S3 Vectors or Bedrock Knowledge Bases, as described in
references/agentic-and-rag.md. - Boundary markers: The guidance recommends implementing security controls such as Bedrock Guardrails and PII masking within the tracing layer (Langfuse) to mitigate injection risks.
- Capability inventory: While the recommended stack includes powerful tools like Ray Serve and vLLM capable of model execution, the skill itself provides no executable code or scripts that directly invoke these capabilities with unsanitized input.
- Sanitization: The skill explicitly includes a 'Security Baseline' in
references/security-and-compliance.md, mandating ECR image scanning, EKS Pod Identity for least-privilege access, and Secrets Manager for credential management.
Audit Metadata