skills/nvidia/skills/hsb-test/Gen Agent Trust Hub

hsb-test

Pass

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses SSH to execute shell commands and Docker operations on a remote devkit to manage the test environment. It maintains session state on the remote host in /tmp and interacts with hardware via hololink CLI tools.
  • [EXTERNAL_DOWNLOADS]: Users can provide a URL for a test plan document, which the skill fetches and parses to generate the executable test list.
  • [REMOTE_CODE_EXECUTION]: The skill executes applications and scripts inside a Docker container on the remote target based on test plan instructions. It uses docker run with high privileges (--privileged, --net host) for hardware access. This risk is managed via multiple required user approval gates before and during execution.
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection (Category 8) because it processes untrusted test documents from user-supplied paths or URLs.
  • Ingestion points: Phase 1 fetches test plan content from local paths, remote devkit paths, or external URLs.
  • Boundary markers: No specific delimiters are used to wrap the ingested test plan content.
  • Capability inventory: The skill has high capabilities, including remote shell execution, privileged container access, and network operations.
  • Sanitization: No technical sanitization of test content is performed; the skill relies on the human operator to approve the parsed test list before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 12, 2026, 07:42 PM
Security Audit — agent-trust-hub — hsb-test