Skills
Navigate
skills/wshobson/agents/spark-training-gotchas/Gen Agent Trust Hub

spark-training-gotchas

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides instructions to execute system-level maintenance commands.
  • Evidence: sync; echo 3 > /proc/sys/vm/drop_caches in references/gotcha-checks.md for memory management.
  • Evidence: High-resource benchmarking script in references/gotcha-checks.md (G5) that allocates significant GPU memory.
  • Impact: Writing to /proc/sys/vm/drop_caches requires root privileges and clears the page cache for all system processes.
  • [PROMPT_INJECTION]: The skill uses the GitHub CLI to read untrusted external issue data, which creates an indirect prompt injection surface.
  • Ingestion points: references/gotcha-checks.md calls gh issue list --repo NVIDIA/dgx-spark-playbooks.
  • Boundary markers: Absent.
  • Capability inventory: The agent has shell execution and system diagnostic capabilities.
  • Sanitization: None.
  • [EXTERNAL_DOWNLOADS]: The skill references official NVIDIA and PyTorch resources for installation and verification.
  • Details: References to nvcr.io/nvidia/pytorch containers, download.pytorch.org wheels, and github.com/NVIDIA/dgx-spark-playbooks.
  • Note: These sources are well-known or trusted and are used for legitimate configuration and troubleshooting.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 07:33 PM
Security Audit — agent-trust-hub — spark-training-gotchas