Skills
skills/nvidia/skills/omniverse-cad-to-simready/Gen Agent Trust Hub

omniverse-cad-to-simready

Fail

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs privilege escalation by executing docker exec --user root to run chmod commands that modify file system permissions within containers.- [DATA_EXFILTRATION]: The preflight script extracts sensitive information including API keys from the environment and writes them into local .env configuration files.- [DATA_EXFILTRATION]: The skill initiates network requests to user-defined service endpoints for asset rendering and processing tasks using standard libraries.- [REMOTE_CODE_EXECUTION]: The skill includes instructions to install dependencies via piped shell execution patterns (curl ... | sh) from external domains.- [EXTERNAL_DOWNLOADS]: The conversion logic fetches external tools and Gaussian splat converters from remote GitHub repositories and package registries.- [PROMPT_INJECTION]: The skill extracts text tokens and metadata from untrusted source assets like STEP and MJCF files to generate prompts for downstream AI agents, presenting a surface for indirect injection.- [COMMAND_EXECUTION]: Several components dynamically execute Python code by passing multiline string snippets to the interpreter via the python -c command.- [COMMAND_EXECUTION]: Extensive use of subprocess.run to execute external CLIs like git, uv, and simready-validate with user-supplied arguments.
Recommendations
  • HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 12, 2026, 08:08 PM
Security Audit — agent-trust-hub — omniverse-cad-to-simready