isaac-sim-remote

Pass

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill's primary utility, scripts/isaacsim_send.py, acts as an IPC client that sends arbitrary Python code to a local Isaac Sim TCP server for execution. This is an intentional design choice for the 'Isaac Sim Remote' developer control plane.- [COMMAND_EXECUTION]: The script scripts/execute_command.py provides an interface to dispatch any command registered within the Omniverse Kit command registry, allowing broad control over the simulation environment.- [DYNAMIC_EXECUTION]: The scripts/isaacsim_send.py client dynamically assembles Python code at runtime by injecting variable assignments into script templates based on CLI arguments provided by the user or agent.- [EXTERNAL_DOWNLOADS]: Several scripts, including scripts/set_asset_root.py and scripts/verify_asset.py, reference and configure access to external USD assets hosted on NVIDIA's S3 buckets (e.g., omniverse-content-production.s3-us-west-2.amazonaws.com).- [DATA_EXFILTRATION]: The scripts/console_log.py script allows the agent to read the application's session logs from the local filesystem, which can be used to extract debugging information and environment details.- [INDIRECT_PROMPT_INJECTION]: The skill processes external USD stages and application logs, which represents a potential surface for indirect prompt injection if those sources contain malicious instructions.
  • Ingestion points: scripts/open_stage.py (loading USD files), scripts/console_log.py (reading session logs).
  • Boundary markers: None identified in the processing of external content.
  • Capability inventory: Full arbitrary Python execution via scripts/isaacsim_send.py, file system access, and network operations through the Isaac Sim process.
  • Sanitization: No sanitization or validation of the content within processed USD files or log entries was detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 15, 2026, 08:12 AM