coreweave-gpu-node-forensics
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill uses a deterministic diagnostic core (scripts/triage.py) which ensures that GPU triage actions are grounded in code rather than probabilistic model interpretation, minimizing hallucination risks.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted system logs (dmesg, nvidia-smi). This surface is mitigated by design as the skill passes raw data to a local Python script for regex-based extraction of error codes, rather than allowing the model to perform direct interpretation of potentially manipulated log text.
- [COMMAND_EXECUTION]: Tool access is restricted to read-only queries (nvidia-smi -q, kubectl get, dmesg). The skill correctly recommends but cannot execute destructive actions like resets or reboots, maintaining a safe permission boundary.
- [DATA_EXFILTRATION]: No sensitive data exposure or exfiltration patterns were identified. External URLs point to official and well-known documentation from NVIDIA and CoreWeave.
Audit Metadata