skills/nvidia/skills/cdf-simulation/Gen Agent Trust Hub

cdf-simulation

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompt text to generate CDFJob configurations.
  • Ingestion points: User prompt text and Airflow-triggered catalog prompts (SKILL.md, "Inputs" section).
  • Boundary markers: The skill does not define explicit delimiters for user input, but it performs strict matching against a fixed set of five catalog rows and five reference flows.
  • Capability inventory: The skill builds Python scripts using the cdf and pcb_aoi libraries, accesses local USD files for metadata, and produces compiled job plans. It does not perform network operations or arbitrary command execution itself.
  • Sanitization: Input is validated against supported catalog shapes and specific component types read from the USD attachment. Unsupported requests are refused with detailed error messages.
  • [DYNAMIC_EXECUTION]: The skill generates Python source code for cdf.CDFJob scripts based on templates and user-provided parameters (SKILL.md, "Workflow" and "Templates"). These scripts are returned to the caller for execution by an external executor. This pattern involves generating executable content at runtime based on external input, though it follows structured templates.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 03:52 PM