isaac-sim-orchestrator
Warn
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes system commands such as 'nvidia-smi' for GPU profiling and 'pkill' to terminate kit processes. It also relies on shell-based execution of simulation scripts via 'isaac-sim.sh' and 'isaaclab.sh' and instructs the agent to manage path-based environment variables.
- [DYNAMIC_EXECUTION]: The skill describes a process of incremental Python script generation and subsequent execution within the Isaac Sim environment. It also handles the dynamic import of URDF and MJCF assets which are parsed and converted at runtime, posing a risk if these assets are malformed or maliciously crafted.
- [PERSISTENCE]: The 'Phase 4
- Distill' procedure mandates that the agent update the 'SKILL.md' file with iteration results, workarounds, and corrections. This self-modification capability allows the agent to persist changes to its own operating instructions across sessions, which could be exploited to maintain a compromised state.
- [INDIRECT_PROMPT_INJECTION]: The skill processes natural language requests and external asset files like URDF and MJCF. These ingestion points (documented in SKILL.md and api-cheatsheet.md) lack boundary markers or sanitization procedures. When combined with capabilities like filesystem writes and subprocess spawning, this creates a vulnerability where instructions embedded in processed data could override agent logic.
Audit Metadata