coreweave-gpu-cost-leak-hunter

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill uses environment-provided credentials ($CW_TOKEN, $KUBECONFIG) to interact with official CoreWeave APIs and cluster endpoints, which is consistent with its stated purpose of cost auditing.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external command outputs (curl/PromQL results and kubectl node/pod status).\n
  • Ingestion points: Data is retrieved from the Grafana Prometheus proxy and Kubernetes API as specified in SKILL.md and categories definitions.\n
  • Boundary markers: The skill relies on structured JSON output from PromQL and utilizes a dedicated Python script for parsing and arithmetic rather than direct prompt interpolation of raw API strings.\n
  • Capability inventory: The skill utilizes curl, kubectl get, and a local Python script for data processing and report generation.\n
  • Sanitization: Input data is parsed using json.loads() and strictly coerced into numeric types (float/int) within scripts/rank-and-report.py before being formatted into the final CFO-level report template.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:43 AM
Security Audit — agent-trust-hub — coreweave-gpu-cost-leak-hunter