setup-gpu-finops
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a system that adjusts infrastructure state based on external data sources, including Prometheus metrics and inference queue depths from engines like vLLM and Triton.
- Ingestion points: GPU telemetry data from DCGM exporters and queue metrics from inference servers (vLLM/Triton) as outlined in the Suggested Process.
- Boundary markers: The skill explicitly includes a security guardrail (ASI05) that warns against constructing configurations from external content without strict schema validation.
- Capability inventory: The skill utilizes
execute_command,write_file, andcreate_fileto manage Kubernetes resources like DaemonSets and ScaledObjects, which could be influenced by external metric inputs. - Sanitization: Instructions recommend the use of structured contracts (ASI07) and strict schema validation for all dynamically created objects to prevent manipulation by malicious data.
Audit Metadata